Generalized Rellich's Lemmas

• Generalized Rellich's lemmas are a family of results that extend classical inequalities, uniqueness theorems, and compactness properties across various analytical and PDE frameworks.
• They encompass higher-order Hardy–Rellich inequalities, discrete analogues, and scattering uniqueness results via techniques like Carleman estimates and spectral analysis.
• Methodologies include factorization methods, weighted identities, and analytic selections in singular systems that yield sharp coercive bounds and optimal constants.

Searching arXiv for the cited works and closely related papers on generalized Rellich lemmas and Rellich-type results. “Generalized Rellich’s lemmas” denotes a family of extensions of classical results associated with Franz Rellich rather than a single theorem. In current usage, the term covers at least four major lineages: higher-order Hardy–Rellich inequalities and their weighted refinements; uniqueness statements for Helmholtz and scattering problems on unbounded domains; compactness theorems of Rellich–Kondrachov type in generalized Sobolev frameworks; and smooth selection results for kernels of parametrized singular linear systems. Recent arXiv literature develops these themes in continuous, discrete, degenerate, Riemannian, many-body, and complex-frequency settings (Gesztesy et al., 2017, Vesalainen, 2014, Ito et al., 2016, Liu et al., 6 Jul 2025).

1. Classical lineages and the scope of the term

A first classical lineage is inequality-theoretic. Birman’s sequence $\{I_n\}_{n\in\mathbb N}$ extends the Hardy inequality ($n=1$) and the Rellich inequality ($n=2$) to all orders, with

$$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$

For $n=1$ this gives the classical Hardy constant $1/4$, and for $n=2$ it gives the Rellich constant $9/16$ on the half-line (Gesztesy et al., 2017).

A second lineage is uniqueness for Helmholtz-type equations. In this setting, a Rellich lemma states that an outgoing solution whose far-field or large-radius boundary behavior is sufficiently small must vanish identically. This perspective is central both in half-space and cone uniqueness theorems for $(\Delta+\lambda)u=f$ on unbounded domains and in recent extensions to complex wavenumbers associated with scattering poles (Vesalainen, 2014, Liu et al., 6 Jul 2025).

A third lineage is compactness. The classical Rellich–Kondrachov theorem appears as a special case of abstract compact embeddings for generalized Sobolev spaces defined by nonnegative quadratic forms $Q(x,\xi)$, including degenerate and weighted cases (Chua et al., 2011).

A fourth lineage is algebraic and analytic rather than PDE-theoretic. Rellich’s 1969 lemma concerns analytic families of singular matrices $n=1$0 with $n=1$1, asserting the existence of a locally analytic unit vector $n=1$2 satisfying $n=1$3. This has been generalized to $n=1$4 and multiparameter settings, to inhomogeneous systems, and to orthonormal frame fields of kernel vectors (Bellido et al., 2022).

This multiplicity of meanings is a recurrent source of confusion. The modern literature does not use “generalized Rellich’s lemma” for a single canonical extension; instead, it refers to a class of results sharing a common role: converting weak decay, weak regularity, or structural singularity into rigidity, compactness, or sharp coercive control.

2. Higher-order Hardy–Rellich inequalities and weighted identities

In the higher-order inequality setting, Gesztesy, Littlejohn, Michael, and Wellman place Birman’s inequalities in a natural Hilbert-space framework. On the half-line,

$n=1$5

with inner product $n=1$6, one has the key inclusion $n=1$7. Repeated application of the classical Hardy inequality to $n=1$8 yields the entire Birman sequence, and the constants $n=1$9 are sharp. Equality occurs only for the trivial function. The same inequalities hold on finite intervals in the standard Sobolev space $n=2$0, and the constants are again sharp (Gesztesy et al., 2017).

The same paper gives a spectral-operator interpretation through the continuous $n=2$1-fold Cesàro operator

$n=2$2

whose norm satisfies $n=2$3. Moreover, $n=2$4 on a suitable dense domain, and Mellin-transform methods show that $n=2$5 is normal with purely absolutely-continuous spectrum (Gesztesy et al., 2017).

A different but complementary generalization is the weighted $n=2$6-Rellich identity for quasilinear second-order degenerate elliptic operators. For vector fields $n=2$7, $n=2$8, $n=2$9, and $$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$0, the paper “Sharp remainder formulae for general weighted Hardy and Rellich type inequalities for $$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$1” proves an exact identity of the form

$$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$2

under the structural condition

$$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$3

with $$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$4. The remainder is organized through the nonnegative functional

$$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$5

which allows the argument to work for all $$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$6, not only $$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$7. Discarding the nonnegative terms yields a sharp weighted Rellich inequality, and equality in the vanishing-remainder sense occurs only in the virtual case $$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$8 (Shaimerdenov et al., 2 Mar 2026).

In the Euclidean Laplacian case, taking $$\int_0^\infty |f^{(n)}(x)|^2\,dx \;\ge\; \frac{[(2n-1)!!]^2}{2^{2n}} \int_0^\infty \frac{|f(x)|^2}{x^{2n}}\,dx.$$9, $n=1$0, and $n=1$1 with $n=1$2, one obtains explicit sharp identities. For $n=1$3 and $n=1$4, the paper derives a new $n=1$5-Rellich identity with coefficient $n=1$6, and dropping the nonnegative remainders recovers the classical sharp Rellich inequality (Shaimerdenov et al., 2 Mar 2026).

A Riemannian analogue appears in work on complete noncompact manifolds. For a positive weight $n=1$7 satisfying $n=1$8 almost everywhere and suitable lower bounds on $n=1$9 or on $1/4$0, the generalized Rellich inequality on $1/4$1 takes the form

$1/4$2

for $1/4$3, where $1/4$4, $1/4$5, and

$1/4$6

When $1/4$7 and $1/4$8, this recovers the Euclidean weighted Rellich inequality, and the constants agree with the Euclidean literature in the flat case (Choudhury, 2024).

3. Discrete Rellich inequalities and factorization methods

The discrete theory replaces derivatives by lattice differences and the Laplacian by the discrete Dirichlet Laplacian. For finitely supported sequences $1/4$9 with $n=2$0, the one-dimensional operator is

$n=2$1

and the bi-Laplacian is $n=2$2. The classical one-dimensional discrete Rellich inequality states that, for sequences supported in $n=2$3,

$n=2$4

with sharp constant $n=2$5 (Das et al., 2023).

The paper “On improvements of the Hardy, Copson and Rellich inequalities” strengthens this by a factorization identity. For $n=2$6 with $n=2$7, there exists a nonnegative remainder operator $n=2$8 and a weight sequence $n=2$9 such that

$9/16$0

Hence

$9/16$1

and an explicit recursion shows $9/16$2 for all $9/16$3 (Das et al., 2023).

The proof proceeds by seeking a decomposition

$9/16$4

and writing

$9/16$5

The coefficients are chosen so that $9/16$6 annihilates the trial sequence $9/16$7, which yields a three-term recursion for the $9/16$8. The paper then proves positivity and quantitative bounds,

$9/16$9

establishing the existence of the factorization (Das et al., 2023).

The optimality issue is subtle. Gerhat–Krejčiřík–Štampach had shown that $(\Delta+\lambda)u=f$0 cannot be enlarged pointwise for all but finitely many $(\Delta+\lambda)u=f$1 without violating the inequality, and the present paper emphasizes that the same bounds reappear in infinitely many factorization presentations. In this sense, $(\Delta+\lambda)u=f$2 is presented as the best possible pointwise Rellich weight in the discrete half-line setting. The same discrete Rellich improvement feeds directly into sharpened Knopp inequalities of orders $(\Delta+\lambda)u=f$3 and $(\Delta+\lambda)u=f$4, as well as improvements of generalized Hardy-type inequalities (Das et al., 2023).

4. Uniqueness on unbounded domains and scattering at complex frequencies

In PDE and scattering theory, generalized Rellich lemmas are uniqueness statements for solutions of Helmholtz-type equations under decay and support conditions. Vesalainen studies

$(\Delta+\lambda)u=f$5

with $(\Delta+\lambda)u=f$6 and $(\Delta+\lambda)u=f$7 the Agmon–Hörmander spaces, and proves several generalizations of the classical uniqueness theorem on unbounded domains. In the half-space case, if $(\Delta+\lambda)u=f$8 is super-exponentially decaying and supported in $(\Delta+\lambda)u=f$9, then the scattered wave $Q(x,\xi)$0 vanishes in the lower half-space $Q(x,\xi)$1. The proof uses a Carleman estimate extracted from Sylvester–Uhlmann,

$Q(x,\xi)$2

together with cutoff arguments and unique continuation (Vesalainen, 2014).

The same paper proves a discrete-lattice analogue on $Q(x,\xi)$3. For

$Q(x,\xi)$4

if $Q(x,\xi)$5 is super-exponentially decaying and vanishes in the discrete cone

$Q(x,\xi)$6

then $Q(x,\xi)$7 for all $Q(x,\xi)$8. The key inputs are a discrete Paley–Wiener theorem and repeated use of the lattice equation to propagate the value $Q(x,\xi)$9 arbitrarily deep into the cone, where super-exponential decay forces it to vanish (Vesalainen, 2014).

Vesalainen also gives complex-variable proofs under weaker hypotheses: exponential decay of $n=1$00 in a half-space and a polynomial-decay case with exponentially thin support. In Fourier variables, one uses

$n=1$01

analytic continuation of $n=1$02, vanishing on the real sphere $n=1$03 via the Rellich–Vekua lemma, and division by $n=1$04 to continue $n=1$05. These results feed into the discreteness of non-scattering energies for non-compactly supported potentials with suitable decay and support assumptions (Vesalainen, 2014).

A further extension concerns complex wavenumbers $n=1$06, where outgoing fields may grow exponentially at infinity. Liu, Sun, and Zhang study exterior scattering in $n=1$07 for a bounded Lipschitz obstacle $n=1$08, defining outgoing solutions by the Green representation with $n=1$09. They prove two generalized Rellich lemmas. The first assumes

$n=1$10

for any $n=1$11 solution of $n=1$12, and concludes $n=1$13. The second assumes that $n=1$14 is outgoing and only requires

$n=1$15

When $n=1$16 and $n=1$17, these reduce to the classical Rellich lemma (Liu et al., 6 Jul 2025).

The proofs use Fourier–Bessel expansions,

$n=1$18

together with large-argument asymptotics of the Hankel functions. The new feature is the exponential reweighting, which compensates exactly for the growth of $n=1$19 and $n=1$20 when $n=1$21. This addresses a common misconception: the classical unweighted Rellich condition is not adequate for complex poles with negative imaginary part, because outgoing solutions no longer decay in the usual sense (Liu et al., 6 Jul 2025).

These lemmas are then used to prove uniqueness of obstacles from far-field data at non-real $n=1$22, and they underpin an inside-out duality for scattering poles. In the latter, the near-field operator $n=1$23 factorizes as

$n=1$24

where $n=1$25 is the single-layer operator and $n=1$26 is an interior solution operator. Off the set of scattering poles, $n=1$27 is injective with dense range in the appropriate trace space, so $n=1$28 is injective; at scattering poles, injectivity fails. The linear sampling method then identifies poles through the blow-up of regularized solution norms for the equation $n=1$29 (Liu et al., 6 Jul 2025).

5. Compactness and spectral Rellich theorems

The generalized compactness theory begins with abstract Sobolev spaces built from possibly degenerate quadratic forms. Let $n=1$30 be a measurable family of symmetric nonnegative quadratic forms on a finite measure space $n=1$31, and define

$n=1$32

Chua, Rodney, and Wheeden prove an abstract compact embedding theorem: if a bounded set $n=1$33 satisfies localization, finite-overlap control, and a Poincaré-type estimate on finitely many small sets covering all but an $n=1$34-portion of $n=1$35, then the projection $n=1$36 is compact from $n=1$37 into $n=1$38 for every $n=1$39. This gives the classical Rellich–Kondrachov theorem when $n=1$40, but it also yields compactness for weighted boundary-degenerate spaces, $n=1$41-John domains with distance weights, and even quasimetric spaces without any gradient structure, provided local oscillation can be controlled in the required way (Chua et al., 2011).

A spectral version appears in the many-body Schrödinger setting. Ito and Skibsted consider a generalized $n=1$42-body Hamiltonian

$n=1$43

where $n=1$44 is a hard-core configuration space and the pair potentials may be singular, with a decomposition $n=1$45 satisfying the soft-potential hypotheses of Condition 1.2. They define the optimal Besov-type space

$n=1$46

and prove the generalized Rellich theorem: if $n=1$47 and $n=1$48 with $n=1$49 a real non-threshold spectral point, then $n=1$50 (Ito et al., 2016).

The proof uses a Mourre estimate with a rescaled Graf vector field,

$n=1$51

yielding positivity of $n=1$52 up to compact errors on a non-threshold spectral window, together with functional-calculus localization for a propagation observable $n=1$53. The argument proceeds in two stages: first proving $n=1$54, then bootstrapping to $n=1$55. According to the paper, the spaces $n=1$56 and $n=1$57 are optimal, and the theorem covers Coulomb atomic and molecular models with hard-core nuclei (Ito et al., 2016).

These compactness and spectral results illustrate a broader structural point. In one direction, generalized Rellich–Kondrachov theorems convert local oscillation estimates into precompactness. In another, generalized Rellich theorems in scattering and many-body analysis convert weak generalized-eigenfunction control into genuine square integrability. The terminology is shared because both types of result express a rigidity principle beyond the classical Euclidean Sobolev setting.

6. Parametrized singular systems, frame fields, and sensitivity

Rellich’s name also appears in the smooth dependence of solutions to parametrized singular linear systems. Let

$n=1$58

be an $n=1$59 real matrix with analytic entries near $n=1$60, and assume $n=1$61 for all $n=1$62 in a neighborhood. Rellich’s original lemma asserts the existence, after possibly shrinking the neighborhood, of analytic functions $n=1$63 such that the vector $n=1$64 satisfies

$n=1$65

The proof uses nonvanishing minors, cofactor vectors, a vanishing-order argument, and normalization (Bellido et al., 2022).

Bellido and Prieto-Martínez extend this in three directions. First, if the entries $n=1$66 are only $n=1$67, one still obtains a $n=1$68 null-vector field $n=1$69, and if $n=1$70 equals the maximal rank in the neighborhood, then $n=1$71 can be normalized to a nonvanishing $n=1$72 unit kernel vector locally. Second, the same conclusion holds for several real parameters $n=1$73, both in the analytic and $n=1$74 categories, again provided one localizes at maximal-rank points. Third, an inhomogeneous version is proved: if

$n=1$75

and $n=1$76 in a neighborhood, then $n=1$77 admits a local analytic or $n=1$78 solution obtained from Cramer’s rule on a suitable invertible minor (Bellido et al., 2022).

The same paper develops frame fields of kernel vectors. If $n=1$79 is the nullity of $n=1$80, then under the hypotheses of the analytic theorem there exists an analytic $n=1$81-frame field

$n=1$82

such that $n=1$83 and $n=1$84. The inductive step adjoins a rank-one term,

$n=1$85

which increases the rank by one and reduces the nullity by one, allowing iteration (Bellido et al., 2022).

A further development concerns $n=1$86-deficient systems, where $n=1$87. Differentiating

$n=1$88

at $n=1$89 gives

$n=1$90

where $n=1$91. This yields a direct sensitivity method. The paper also introduces an adjoint method based on the zero-Lagrangian

$n=1$92

leading to the adjoint system

$n=1$93

and the sensitivity formula

$n=1$94

The paper emphasizes that this is asymptotically more efficient when many different scalar functionals are considered for the same system matrix (Bellido et al., 2022).

The examples in that work also clarify limitations. Analyticity cannot in general be replaced by mere $n=1$95 without rank constancy, and in the multiparameter case a unit null-vector may fail to extend continuously through points where maximal-rank localization is unavailable. These obstructions are part of the modern meaning of generalized Rellich lemmas in the algebraic setting: they are not only existence theorems, but also sharp descriptions of when smooth or analytic kernel selection is possible (Bellido et al., 2022).