Maximal monotonicity and contraction semigroup for the quantum drift-diffusion (Derrida-Lebowitz-Speer-Spohn) equation

Abstract: We study the quantum drift-diffusion, or Derrida-Lebowitz-Speer-Spohn (DLSS), equation for a nonnegative density $\varrho$ on a bounded convex domain with Neumann boundary conditions, in the square-root variable $u=\sqrt\varrho$. We show that the DLSS operator, defined and monotone on smooth strictly positive functions, admits a unique maximal monotone extension in $L<sup>2(Ω)$, explicitly given by the minimal (defect-free) operator plus the normal cone of the positivity constraint. The generated semigroup, which contracts the Hellinger distance between the densities, thus yields a canonical solution - existing, unique, and stable for every nonnegative $L<sup>2$ initial datum and in every space dimension - independent of any approximation scheme: it is in fact the unique contraction semigroup extending the classical evolutions that emanate from smooth, uniformly positive data. The implicit Euler scheme converges to it, and $\sqrt u\in L^2_{\rm loc}(H<sup>2)$ along the flow. When the datum belongs to the domain of the operator, the solution is strong and satisfies the equation pointwise, with no reaction term created on the vacuum ${u=0}$. We characterize the trajectories in several equivalent ways - as Bénilan integral solutions and through one-sided weak formulations - prove the maximality of the operator also in the $H<sup>2$-$H<sup>{-2}$ duality and, in dimension $d\le3$, identify the flow with the weak solutions in the uniqueness class of Fischer. A second-order estimate of independent interest underlies the construction: on a convex domain with Neumann conditions the dissipation $\int_Ω(Δu)^{2/u\,\mathrm{d}} x$ is finite exactly when $\sqrt u\in H<sup>2(Ω)$, and it then controls the full Hessian of $\sqrt u$, in every dimension.

• The paper constructs the unique maximal monotone extension of the DLSS operator in L² as the classical operator plus the positive-cone normal cone, establishing a canonical contraction semigroup for all nonnegative data and dimensions.
• It proves that finite dissipation is equivalent to square-root H² regularity on convex domains, with explicit Hessian estimates, and uses defect measures to characterize behavior on vacuum regions without adding reaction terms.
• The resulting implicit Euler scheme preserves nonnegativity, converges to the semigroup, and yields a weak-solution theory that matches Fischer’s formulation in dimensions d≤3 while leaving uniqueness and identification open for d≥4.

Setting and motivation

The quantum drift-diffusion, or Derrida–Lebowitz–Speer–Spohn (DLSS), equation governs a nonnegative density $\varrho$ on a bounded convex domain $\Omega\subset\mathbb R^d$ with homogeneous Neumann conditions:

$$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$

equivalently $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$. Introduced in the study of interface fluctuations and arising as the field-free, pressureless case of the quantum drift-diffusion system of semiconductor modelling, the equation is analytically delicate: it is fourth order, violates the maximum principle, and its nonlinearity is singular on the vacuum set $\{\varrho=0\}$. Prior theory supplied weak solutions via a variety of approximation schemes — each carrying its own weak formulation and typically requiring finite entropy of the datum — and uniqueness for $d\le 3$ under periodic boundary conditions in the regularity class $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$ due to Fischer [Fischer13]. The flow is known to conserve mass, dissipate the Boltzmann entropy and the Fisher information, and to contract the Hellinger distance between densities.

The paper under discussion by Matthes, Savaré, and Schlichting takes the Hellinger contraction as its point of departure: in the square-root variable $u=\sqrt\varrho$, that contraction is precisely an $L^2$-contraction, suggesting that the DLSS operator $u\mapsto\Delta^2 u-(\Delta u)^2/u$ be treated as a maximal monotone operator in $\Omega\subset\mathbb R^d$0, with the equation read as the inclusion $\Omega\subset\mathbb R^d$1 solved by the implicit Euler scheme. Three obstacles stand in the way: the operator is a priori defined and monotone only on smooth strictly positive functions; it is unclear how to extend it to the whole positive cone $\Omega\subset\mathbb R^d$2; and, unlike the Wasserstein formulation for $\Omega\subset\mathbb R^d$3, the $\Omega\subset\mathbb R^d$4-level formulation has no variational (minimization) structure, so even a single implicit step is not a minimization problem. The paper resolves all three.

Second-order calculus: the dissipation controls the Hessian

A technical result of independent interest underpins the whole construction. Define the dissipation functional

$\Omega\subset\mathbb R^d$5

with the integrand interpreted via the lower semicontinuous perspective function $\Omega\subset\mathbb R^d$6 (equal to $\Omega\subset\mathbb R^d$7 at $\Omega\subset\mathbb R^d$8, $\Omega\subset\mathbb R^d$9 and $$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$0 at $$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$1, $$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$2), so finiteness of $$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$3 also forces $$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$4 on the vacuum. The central theorem states that on a convex domain with Neumann conditions, $$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$5 if and only if $$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$6, and in that case

$$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$7

valid in every dimension, with the lower constant degenerating no faster than $$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$8. The proof combines a pointwise sum-of-squares inequality for symmetric matrices (proved by Sylvester's criterion on a $$\partial_t\varrho = -2\,\nabla\cdot\Big(\varrho\,\nabla\frac{\Delta\sqrt\varrho}{\sqrt\varrho}\Big),$$9 form whose principal minors are computed explicitly), the convexity bound $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$0, an integration-by-parts identity for the quartic term $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$1, and Ioffe's lower semicontinuity theorem. A relaxed functional $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$2 built from the Neumann heat semigroup coincides with $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$3 and is convex and weakly lower semicontinuous on $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$4 — the property that makes the a priori theory close under limits. The consequence is that the class $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$5 is exactly the domain of finiteness of $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$6, and it essentially coincides with Fischer's uniqueness class.

The DLSS operator and its defect measure

The operator is first realized between $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$7 and $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$8. A pair $$\partial_t\varrho + \Delta^2\varrho - \sum_{ij}\partial^2_{ij}\big(\partial_i\varrho\,\partial_j\varrho/\varrho\big)=0$$9 belongs to the graph of the operator $\{\varrho=0\}$0 when $\{\varrho=0\}$1, $\{\varrho=0\}$2 (so $\{\varrho=0\}$3), and $\{\varrho=0\}$4 where $\{\varrho=0\}$5 is a positive measure in $\{\varrho=0\}$6 satisfying the constraint $\{\varrho=0\}$7 via the quasi-continuous representative $\{\varrho=0\}$8. The excess $\{\varrho=0\}$9 is a defect measure, necessarily concentrated on the vacuum set $d\le 3$0. When the datum $d\le 3$1 itself is a measure in $d\le 3$2, the structure is completely identified:

$d\le 3$3

so the defect is exactly the restriction of $d\le 3$4 to the vacuum, and $d\le 3$5 there. In $d\le 3$6 terms, this yields the normal-cone structure

$d\le 3$7

with $d\le 3$8: all elements of $d\le 3$9 coincide with the minimal section $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$0 on $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$1 and are nonpositive on the vacuum, and $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$2 is a singleton exactly when $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$3 is positive a.e. The paper stresses that this normal-cone term is not an artifact: any maximal monotone operator whose domain lies in a closed convex set necessarily contains it, and maximality here shows that nothing further must be added — in particular, no reaction term is created when the solution touches zero.

Monotonicity of $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$4 in the $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$5–$\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$6 duality is proved by a completing-the-square argument adapted to the measure-valued defect: the monotonicity gap reduces to a weighted $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$7 square, and the constraint $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$8 supplies exactly the cross terms needed. At the very weak level (the form $\varrho^{1/2},\varrho^{1/4}\in L^2_{\rm loc}(H^2)$9 on $u=\sqrt\varrho$0), genuine monotonicity does not survive, but an asymmetric inequality retains its core: $u=\sqrt\varrho$1 equals the negative of a pointwise square whenever $u=\sqrt\varrho$2 and $u=\sqrt\varrho$3 is smooth and uniformly positive. This asymmetric inequality later drives the characterization of the semigroup trajectories.

Maximality and canonicity

The maximality of $u=\sqrt\varrho$4 in $u=\sqrt\varrho$5 is obtained through an $u=\sqrt\varrho$6-regularized variational inequality in $u=\sqrt\varrho$7, $u=\sqrt\varrho$8, penalized by a smooth convex barrier $u=\sqrt\varrho$9 combining a $L^2$0 gradient term and a singular weight $L^2$1 that forces strict positivity of the approximants. The choice of a smooth barrier is deliberate: a hard two-sided obstacle $L^2$2 would yield only a one-sided variational inequality, blocking the nonlinear tests ($L^2$3, heat-flow perturbations) from which the entropy and $L^2$4 estimates are extracted. Existence for fixed $L^2$5 follows from a Baiocchi–Capelo type theorem; uniform $L^2$6 and entropy estimates (the latter via a Gagliardo–Nirenberg interpolation giving $L^2$7) allow the passage $L^2$8, the defect measure being recovered by lower semicontinuity. The same scheme with the duality map $L^2$9 in place of the identity proves maximality of $u\mapsto\Delta^2 u-(\Delta u)^2/u$0 in the $u\mapsto\Delta^2 u-(\Delta u)^2/u$1–$u\mapsto\Delta^2 u-(\Delta u)^2/u$2 duality; the two maximality properties are shown to be logically independent.

Two canonicity statements follow. First, $u\mapsto\Delta^2 u-(\Delta u)^2/u$3 is the unique maximal monotone operator in $u\mapsto\Delta^2 u-(\Delta u)^2/u$4 extending the smooth core $u\mapsto\Delta^2 u-(\Delta u)^2/u$5 (smooth, uniformly positive functions) whose domain consists of nonnegative functions; the key propagation lemma shows that monotonicity against the smooth core already forces monotonicity against the whole minimal section, via a heat-semigroup commutation inequality $u\mapsto\Delta^2 u-(\Delta u)^2/u$6 derived from a Jensen inequality for Markov kernels. Second, dynamically: $u\mapsto\Delta^2 u-(\Delta u)^2/u$7 is the unique strongly continuous contraction semigroup on $u\mapsto\Delta^2 u-(\Delta u)^2/u$8 whose generator agrees with the classical DLSS operator at $u\mapsto\Delta^2 u-(\Delta u)^2/u$9 on the smooth core. Hence the semigroup is not the by-product of a particular approximation scheme, and the solution concept is unambiguous.

The semigroup, the Euler scheme, and regularity

Generation theory then applies verbatim. For every $\Omega\subset\mathbb R^d$00 there is a unique Bénilan integral solution, given by the exponential formula as the limit of implicit Euler iterates, and the map $\Omega\subset\mathbb R^d$01 is an $\Omega\subset\mathbb R^d$02-contraction — exactly the Hellinger contraction for the densities $\Omega\subset\mathbb R^d$03. For $\Omega\subset\mathbb R^d$04 the solution is strong, Lipschitz in time, right-differentiable everywhere, satisfies the pointwise equation $\Omega\subset\mathbb R^d$05 with no reaction term on the vacuum, and $\Omega\subset\mathbb R^d$06 is nonincreasing. The inhomogeneous problem with a source $\Omega\subset\mathbb R^d$07 is well posed as an immediate corollary of the abstract theory.

The discrete scheme inherits the structural features of the flow: each Euler step involves only the defect-free operator $\Omega\subset\mathbb R^d$08 (the defect measure vanishes), preserves nonnegativity without recourse to a maximum principle, cannot enlarge the vacuum set up to Lebesgue-negligible sets, and loses mass at rate $\Omega\subset\mathbb R^d$09 — consistent with exact conservation of $\Omega\subset\mathbb R^d$10 in the limit.

Regularity of the trajectories is dimension-free at the level of the square root: along every trajectory, $\Omega\subset\mathbb R^d$11, so $\Omega\subset\mathbb R^d$12 (i.e. $\Omega\subset\mathbb R^d$13) for a.e. $\Omega\subset\mathbb R^d$14, for arbitrary $\Omega\subset\mathbb R^d$15 data and in every dimension. If the initial entropy $\Omega\subset\mathbb R^d$16 is finite, a dimension-free entropy dissipation inequality $\Omega\subset\mathbb R^d$17 upgrades this to $\Omega\subset\mathbb R^d$18; for $\Omega\subset\mathbb R^d$19 this finiteness holds automatically for a.e. $\Omega\subset\mathbb R^d$20 (since $\Omega\subset\mathbb R^d$21 by Sobolev embedding), so the semigroup itself generates Fischer's regularity class from bare $\Omega\subset\mathbb R^d$22 data.

Weak formulations and the low-dimensional identification

The paper introduces a new notion of weak solution, requiring only: constant $\Omega\subset\mathbb R^d$23-norm; $\Omega\subset\mathbb R^d$24; and the one-sided inequality

$\Omega\subset\mathbb R^d$25

where $\Omega\subset\mathbb R^d$26. This formulation makes sense for arbitrary nonnegative data in every dimension, involves no assumption on the time derivative, and is shown equivalent to being a Bénilan integral solution: the forward direction uses the asymmetric inequality, the converse a delicate compactness argument in which the perspective term is passed to the limit via Ioffe's theorem, with $\Omega\subset\mathbb R^d$27 for $\Omega\subset\mathbb R^d$28 supplying the strong/weak convergence split. The solution class is closed under locally uniform $\Omega\subset\mathbb R^d$29 convergence and the solution map is 1-Lipschitz in the datum.

For $\Omega\subset\mathbb R^d$30, the picture sharpens: every semigroup trajectory satisfies the equality in the $\Omega\subset\mathbb R^d$31-form equation $\Omega\subset\mathbb R^d$32 (a defect-free renormalized formulation, stronger than the one-sided inequality), together with a positivity condition. Since in Fischer's class the $\Omega\subset\mathbb R^d$33-form coincides with the renormalized weak formulation for which uniqueness holds, the semigroup trajectories are exactly the weak solutions of the $\Omega\subset\mathbb R^d$34-form equation in that class. The paper is explicit that the final identification step is conditional: Fischer's theorem is stated for periodic boundary conditions, and its adaptation to the homogeneous Neumann setting — natural in view of the heat-semigroup regularization device used throughout — is not carried out here.

Limitations and open questions

The paper states its boundaries plainly. Whether $\Omega\subset\mathbb R^d$35 coincides with the weak solutions produced by the earlier approximation schemes is settled only for $\Omega\subset\mathbb R^d$36, through the $\Omega\subset\mathbb R^d$37-form characterization and Fischer's uniqueness theorem, the latter conditionally on the Neumann adaptation of Fischer's approximation lemmas. For $\Omega\subset\mathbb R^d$38 no uniqueness theory for weak solutions is available, and the identification question remains open — though in every dimension the paper does deliver one distinguished, canonically determined solution with continuous dependence on the datum. The extensions sketched (weighted setting with $\Omega\subset\mathbb R^d$39 curvature condition, and the full quantum drift-diffusion operator with isothermal pressure and drift) are argued at the level of monotonicity and algebraic structure only: the weighted transfer requires the Bochner bound and modifies the boundary condition to Robin unless $\Omega\subset\mathbb R^d$40; only drift potentials of the form $\Omega\subset\mathbb R^d$41 (possessing a positive zero mode) are reachable by the ground-state transform; and the maximal monotone extension of the pressure-perturbed operator is asserted as likely, not proved. The constants in the second-order estimates are explicitly not optimized (in $\Omega\subset\mathbb R^d$42 the lower bound refines to an identity).

Conclusion

The paper converts the formal inclusion $\Omega\subset\mathbb R^d$43 for the DLSS equation into a theorem: the classical DLSS operator on smooth positive functions admits a unique maximal monotone extension in $\Omega\subset\mathbb R^d$44, explicitly $\Omega\subset\mathbb R^d$45, generating a canonical contraction semigroup that exists, is unique, and is stable for every nonnegative $\Omega\subset\mathbb R^d$46 datum in every space dimension, contracts the Hellinger distance, creates no reaction on the vacuum, and is reachable by a convergent implicit Euler scheme whose steps are variational inequalities rather than minimizations. The supporting second-order estimate — $\Omega\subset\mathbb R^d$47 finite exactly when $\Omega\subset\mathbb R^d$48, controlling the full Hessian with a $\Omega\subset\mathbb R^d$49 constant on convex domains — is a standalone contribution. The main unresolved issue left by the paper is the identification of the canonical semigroup with the broader weak-solution theory beyond dimension three, and the completion of the Neumann version of Fischer's uniqueness argument.