sign in sign up
<<
Home , Functional analysis

arXiv:2005.07561v1 [math-ph] 15 May 2020 stesmercdcesn eragmn e h monogra the see – rearrangement decreasing variati symmetric of m the The field is growing optimisation. th spectral ever by and and approximation minimised vast eigenvalue is the in Laplacian result Dirichlet archetypal negative the of value h prto pcrlotmsto,wihcnma ohthe both mean can which optimisation, spectral of spirit the ls h prtra oe eibuddon semi-bounded lower as the alise we emtyadseta rpris nie htcnb t be can [ that conjecture idea celebrated an Rayleigh’s properties, spectral and inequalit tween Faber-Krahn-type underlying concept general The therein. references the ot h ifrniloperator differential the forth preci is on respect conditions this boundary in Dirichlet focus natural first the Historically, t Ω states inequality foot Faber-Krahn same formulation original the its on In results bo both in present similar to are like utilised should we tools reason technical the t and interest, proofs of the operators distinct two concern applications l,otmsto eutdet .Fbr[ Faber famo G. a of to generalisations due of result types optimisation two produce old, we work this In AE-RH NQAIISFRSCHR FOR INEQUALITIES FABER-KRAHN ⊂ R d oooiiypoete ftelws ievlei lone also is eigenvalue lowest caref the a of properties model monotonicity first the in techniques; decreasing symmetric rearrangement on Steiner based are t results at such attraction of Coulomb proofs The of terms in given Co being optimiser attractive the with Faber- Schr¨odinger operator three-dimensional two-body i establish and point we one- the Next, with ball centre. the its is at optimiser the interaction: point Schr First, the for domain. inequality bounded Faber-Krahn a three-dimensional and on Laplacian Dirichlet the of tions A , IHPITADWT OLM INTERACTIONS COULOMB WITH AND POINT WITH d BSTRACT ≥ LDMRLTRIHKADAESNR MICHELANGELI ALESSANDRO AND LOTOREICHIK VLADIMIR 2 ihtesm ie nt oue h oet(rnia)eig (principal) lowest the volume, finite given same the with , eoti e ae-rh-yeieulte o eti p certain for inequalities Faber-Krahn-type new obtain We . 1. akrudadoutline and Background − ∆ 58 h iu inbigicue nodrt re- to order in included being sign minus the , Ω ]. b Lpain esaludrtn hence- understand shall we ‘Laplacian’ (by 1 31 n .Kan[ Krahn E. and ] L 2 (Ω) .Ohrstig fitrs in interest of settings Other ). DNE OPERATORS ODINGER ¨ rh nqaiisfor inequalities Krahn ¨odinger with operator eded. trcinsupported nteraction ecneta ceeof scheme conceptual he laayi fcertain of ul a mns l domains all amongst hat ecnr fteball. the of centre he eyteLpainwith Laplacian the sely lm interactions, ulomb eetbihatwo- a establish we eragmn and rearrangement hcss hc sthe is which cases, th e sterlto be- relation the is ies pe n h lower the and upper i oli h proof the in tool ain ing. ae akt Lord to back raced 49 al hsi an is This ball. e nlmtosfor methods onal h [ phs .Weesour Whereas ]. s n century one us, erturba- 6 , 46 , 4 and ] en- 2 V. LOTOREICHIK AND A. MICHELANGELI bound depending on the problem, are the Neumann Laplacian, whose lowest non-trivial eigenvalue is maximised by the ball [62, 63], Laplacians with Robin [27, 28, 33, 19, 43, 18, 52, 20], Wentzell-Robin [44, 45], Stekloff [64, 15], or mixed [48] boundary conditions, and more generally for Laplace-Beltrami operators for domains in compact Riemannian with various boundary conditions [54, 23, 65, 47, 60], just to scratch the surface of a huge and branched out research field. Of course, all this come up with a variety of techniques that may differ sub- stantially from the rearrangement scheme of the original Faber-Krahn inequality. Beside investigating different sorts of boundary conditions, also other sorts of differential operators have been studied which give rise to Faber-Krahn-type in- equalities, significantly p-Laplacians [11, 17, 45, 26, 40], magnetic Laplacians [32], and Dirac operators [8, 3]. Certain analogous spectral optimisation results have been established also for Robin Laplacians on the exterior of compact sets [50, 51, 29] and on unbounded cones [47]. The direction we are concerned with here is the emergence of Faber-Krahn-type inequalities for suitable perturbations of the Dirichlet Laplacian on bounded do- mains Ω. The first playground one may think of are of course Schr¨odinger operators −∆+V for suitable measurable potential V :Ω → R. For instance (see, e.g., [9, Sect. 4]) a straightforward adaptation of Faber-Krahn inequality holds, stating that amongst all Ω’s with same finite volume, the lowest eigenvalue of −∆+ V with V non- negative in L1(Ω) and with Dirichlet boundary conditions always exceeds the lowest eigenvalue of the analogous Schr¨odinger operator on the ball with poten- tial given by the symmetric increasing rearrangement of V . In this framework, optimisation (say, of the first eigenvalue, of the fundamental gap, etc.) has been investigated with respect to various classes of potentials at fixed Ω (see, e.g., the survey in [39, Chapter 8] and the references therein) Here our focus splits into two lines. Dirichlet boundary conditions shall be as- sumed throughout. On the one hand, we are concerned with the question of optimising the lowest eigenvalue of a Schr¨odinger operator with localised im- purity. There are significant precursors [22, 38, 25], which qualified the optimal placement of an obstacle or a well within a fixed bounded domain Ω, meaning, a positive or negative bump-like potential V supported inside Ω. We push this line further, by modelling the impurity with an operator of point interaction, that is, a singular delta-like perturbation of the Dirichlet Laplacian supported at a point y ∈ Ω. The connection with bump-like potentials of finite size is clear by analogy with the case of a point interaction Hamiltonian on the whole Rd, d ∈ {1, 2, 3}: the latter can be indeed constructed as a suitable of Schr¨odinger operators 2 d −∆+ Vn on L (R ) along a sequence of sufficiently localised and regular poten- tials Vn shrinking and spiking up to a delta-like profile as n →∞ [2]. FABER-KRAHN INEQUALITIES WITH INTERACTIONS 3

We thus consider a bounded domain Ω ⊂ Rd with C∞-boundary, d ∈ {2, 3}, and HΩ 2 R the operator α,y on L (Ω) for given y ∈ Ω and α ∈ , namely the self-adjoint operator of point interaction supported at y and with inverse scattering length α. HΩ Loosely speaking, α,y corresponds to the formal differential expression

−∆+ ναδy with Dirichlet boundary condition at ∂Ω and some coupling να (of which α is HΩ a suitable renormalisation). In fact, α,y is rigorously defined as a self-adjoint extension in L2(Ω) of the Dirichlet Laplacian restricted to smooth functions van- ishing on neighbourhoods of y, a construction obtained in [13] (see also [24]). Ba- HΩ sic spectral properties and an amount of further results on α,y were established in [13, 30, 57]. As the above singular perturbation at y does not alter the lower semi-boundedness of the unperturbed Dirichlet Laplacian, it still makes sense to α HΩ investigate the principal eigenvalue λ1 (Ω,y) of α,y. In particular, [30] proved α strict monotonicity of λ1 (Ω,y) with respect to certain directions along which y is moved, thus a first partial answer to the question where to locate a point interac- α tion of given strength so as to minimise λ1 (Ω,y). Our first main result in the present analysis is the solution to a problem that merges the above question with the isoperimetric question for domains with the α same volume, namely the problem of optimising λ1 (Ω,y) with respect to a simul- taneous variation of the domain Ω and the point y ∈ Ω. We demonstrate the Faber-Krahn inequality

α B 0 α λ1 ( , ) ≤ λ1 (Ω,y), where B ⊂ Rd is the ball of the same volume as Ω centred at the origin 0 ∈ Rd. This is achieved by means of rearrangement techniques (for which we provide a concise survey in Section 2), combined with an accurate analysis of certain crucial α features of λ1 (Ω,y) and its associated , in particular the monotonic- ity of the former with respect to α, and some convenient representations of the latter (an analysis that we develop in Sections 3 and 4). We present the proof of this first main result in Section 5. It is worth emphasizing that we demon- strate the above Faber-Krahn inequality following two alternative routes: a gen- eral one that relies on certain estimates available in the literature for Green func- tions for Dirichlet Laplacians on domains, and an additional one, applicable when α λ1 (Ω,y) > 0, that has the virtue of exploiting rearrangement techniquesas for the original Faber-Krahn, and in fact allows us to re-prove independently the above mentioned Green estimates. In this respect, it is remarkable that the lat- ter are so intimately connected with the of the Hamiltonian with a point interaction in bounded domain. In the second line of investigation of this work, on the other hand, we are con- cerned with the optimisation of the lowest eigenvalue of certain three-dimensional 4 V. LOTOREICHIK AND A. MICHELANGELI

Schr¨odinger operators on bounded domain and with attractive Coulomb poten- tial. We actually examine two models. First, for generic bounded domain Ω ⊂ R3 with C∞-boundary, y ∈ Ω, and q > 0, we consider the operator q TΩ = −∆ − q,y x |x − y| in its natural self-adjoint realisation on L2(Ω) with Dirichlet boundary conditions at ∂Ω. Here too the Coulomb perturbation produces a lower semi-bounded oper- q ator, with lowest eigenvalue µ1(Ω,y). For the latter we establish the Faber-Krahn inequality q B 0 q µ1( , ) ≤ µ1(Ω,y) . Thus, the configuration with Coulomb attraction at the centre of the ball min- imises the principal Schr¨odinger-Coulomb eigenvalue among all domains with equal volume and generic interaction centre y. HΩ Next, we examine the two-body counterpart of the previous model. Whereas α,y TΩ and q,y above are naturally interpreted as Hamiltonians for one non-relativistic quantum particle confined in Ω and subject to an interaction (of contact or Coulomb type) centred at a fixed point y, we now study the quantum Hamiltonian for two particles confined in Ω and coupled among themselves by a two-body, isotropic, Coulomb attraction of intensity q > 0. The Hamiltonian of interest becomes

TΩ q q = −∆x1 − ∆x2 − , |x1 − x2| canonically realised as a self-adjoint operator on L2(Ω × Ω) with Dirichlet bound- ary conditions at ∂(Ω × Ω). Here one should think of q as the product of the TΩ (absolute values of) the charges of the two particles. Again, q is lower semi- q q bounded, with lowest eigenvalue ν1(Ω). In the optimisation of ν1 (Ω) over all Ω’s with the same volume we establish the lower bound q/2 B 0 q 2µ1 ( , ) ≤ ν1(Ω) . The l.h.s. above formally expresses the sum of the lowest energy levels of two identical particles in the ball B, each of which evolves uncoupled from the other and is subject instead to a Coulomb attraction from the centre of the ball, where now the product of the charge of the centre and the charge of each particle is half of the original q. Such analysis is carried on in Section 6. The conceptual scheme for the one-body case goes along the same line as for our Faber-Krahn inequality for the one-body point interaction. Exporting that scheme to the two-body case requires the re- placement of the standard symmetric rearrangement tools with the Steiner re- arrangement (which is also concisely reviewed in Section 2). FABER-KRAHN INEQUALITIES WITH INTERACTIONS 5

In conclusion, we provide three new non-trivial examples of variational eigen- value estimates for Schr¨odinger self-adjoint operators on bounded domains, the first two of which are in the form of novel Faber-Krahn-type inequalities. As the models we considered here appear not to have had previous scrutiny as far as spectral optimisation is concerned, an amount of interesting open ques- tions obviously arise, on some of which we are committed to, as counterparts of the corresponding analysis for the free Laplacian on L2(Ω) with given boundary conditions of self-adjointness. This includes, for example, the question of unique- ness of the minimiser, or the problem of optimisation over a restricted class of domains with definite geometry (such as parallelepipeds with the same volume of a prescribed cube), or the behaviour with respect to different boundary con- ditions other than Dirichlet, or the optimisation of the fundamental gap, just to mention a few typical ones. It would be of interest also to supplement our anal- TΩ TΩ ysis of q,y and q by including the case of negative coupling q < 0, and by q q B comparing ν1 (Ω) with ν1 ( ). We believe that these attractive topics deserve future investigation.

Notation. Beside an amount of fairly standard notation, as well as further conve- nient shorthand that will be introduced in due time, we shall adopt the following conventions throughout. |A| Lebesgue of a measurable set A ⊂ Rd ∂Ω boundary of a domain Ω ⊂ Rd supp f support of the function f 2 (·, ·)L2(Ω) L -scalar product, anti-linear in the first entry, linear in the second

½ identity operator (on the that will be clear from the context) dom domain of an operator or a quadratic form σ(T ) spectrum of the operator T w.r.t. the underlying T1 ⊗ T2 tensor products of operators T1, T2 (w.r.t. the underlying Hilbert spaces) Unless when it becomes relevant to emphasize that, we shall tacitly understand all identities f = g between L2-functions in the sense of almost everywhere iden- tities.

2. Preparatory materials

2.1. Symmetric decreasing rearrangement. Let us start with introducing the sym- decreasing rearrangement and recalling some of its fundamental proper- ties. This is standard material; we refer to the monographs [6, 55, 46, 53] for additional details. 6 V. LOTOREICHIK AND A. MICHELANGELI

Let A be a measurable set of finite volume in the Rd of d ≥ 2. Its symmetric rearrangement A∗ is the open ball B ⊂ Rd centred at the origin 0 ∈ Rd and such that |A| = |B|. Let u: Rd → R be a non-negative that vanishes at infinity, in the sense that all its positive level sets have finite measure: (2.1) {x ∈ Rd u(x) >t} < ∞, ∀ t> 0. We define the symmetric decreasing rearrangement u∗ of u by symmetrizing its level sets as ∞ ∗ ∗ d (2.2) u (x) := χ{u>t} (x) t. Z0 d d Here χA : R → R is the characteristic function of a measurable set A ⊂ R . The rearrangement u∗ has a number of straightforward properties.

Lemma 2.1. Let u: Rd → R, d ≥ 2, be a non-negative measurable function vanishing at infinity. Let A ⊂ Rd be a measurable set of finite volume. Then:

(i) u∗ is non-negative; (ii) u∗ is radially symmetric and non-increasing; (iii) u and u∗ are equi-measurable, i.e., |{x ∈ Rd u(x) >t}| = |{x ∈ Rd u∗(x) >t}| for all t> 0;

(iv) supp u ⊂ A implies supp u∗ ⊂ A∗; (v) (u∗)2 = (u2)∗.

Let us collect further standard properties of the symmetric decreasing rearrange- ment that we shall use throughout.

Proposition 2.2. [55, Theorem 3.4 and Lemma 7.17] Let u, v : Rd → R be non- negative measurable functions vanishing at infinity. Then the following hold.

∗ 2 (i) kukL2(Rd) = ku kL2(Rd) (conservation of L -). d ∗ ∗ d (ii) Rd u(x)v(x) x ≤ Rd u (x)v (x) x (Hardy-Littlewood inequality). (iii) IfR ∇u ∈ L2(Rd) existsR in the sense of distributions, then ∇u∗ has the same property ∗ k∇ukL2(Rd) ≥ k∇u kL2(Rd) (P`olya-Szeg˝oinequality).

In particular, u ∈ H1(Rd) implies that u∗ ∈ H1(Rd) as well.

In view of Lemma 2.1 (iv), the operation of taking symmetric decreasing rearrange- ment can be naturally extended to functions defined on domains. FABER-KRAHN INEQUALITIES WITH INTERACTIONS 7

Let Ω ⊂ Rd be a bounded domain with C∞-smooth boundary. For a non-negative measurable function u:Ω → R, we denote by u: Rd → R its extension by zero to the whole Rd and define the symmetric rearrangement u∗ of u as ∗ ∗ ∗ e u := u |Ω∗ :Ω → R. In this respect Proposition 2.2 has the following corollary. e Corollary 2.3. Let Ω ⊂ Rd, d ≥ 2, be a bounded domain with C∞-smooth boundary and let the ball B := Ω∗ be its symmetric rearrangement. Let u, v :Ω → R be non-negative measurable functions. Then:

∗ (i) kukL2(Ω) = ku kL2(B); d ∗ ∗ d (ii) Ω u(x)v(x) x ≤ B u (x)v (x) x; 1 ∗ 1 B ∗ (iii) Rif additionally u ∈RH0 (Ω), then u ∈ H0 ( ) and k∇ukL2(Ω) ≥ k∇u kL2(B).

Next, following the lines of [14, 16, 21], we introduce for a non-negative measur- 2d d able function u = u(x1,x2): R → R, x1,x2 ∈ R , d ≥ 2 its Steiner rearrangements with respect to the first d and the last d variables, that is, ∗ (S1u)(·,x2) = (u(·,x2)) (2.3) ∗ (S2u)(x1, ·) = (u(x1, ·)) .

Here, we implicitly assume that u(x1, ·) and u(·,x2) are vanishing at infinity for d d almost all x1 ∈ R and x2 ∈ R , respectively. These rearrangements have an amount of properties reminiscent of those for the standard symmetric decreasing rearrangement and follow from the latter via sim- ple arguments (see [16, Theorem 8.2]).

Proposition 2.4. Let u ∈ H1(R2d) be real-valued and non-negative. Let A ⊂ Rd be a measurable set of finite volume. Then 1 2d S1u, S2u, S1S2u, S2S1u ∈ H (R ). Moreover, the following hold.

∗ ∗ (i) If supp u ⊂ A × A, then supp(S1u) ⊂ A × A, supp(S2u2) ⊂ A × A and ∗ ∗ supp(S1S2u), supp(S2S1u) ⊂ A × A . 2 2 2 2 2 2 2 2 (ii) (S1u) = S1u , (S2u) = S2u , (S1S2u) = S1S2u , and (S2S1u) = S2S1u . d (iii) If u(x1,x2)= u(x2,x1) for a.e. x1,x2 ∈ R , then S1S2u = S2S1u.

(iv) kukL2(R2d) = kS1ukL2(R2d) = kS2ukL2(R2d) = kS1S2ukL2(R2d) = kS2S1ukL2(R2d).

(v) k∇ukL2(R2d) ≥ k∇S1S2ukL2(R2d) and k∇ukL2(R2d) ≥ k∇S2S1ukL2(R2d).

2.2. Green functions. Let us now recall basic properties of the Green function associated with the Dirichlet Laplacian on a bounded smooth domain, focusing in 8 V. LOTOREICHIK AND A. MICHELANGELI particular on the connections between Green function and symmetric decreasing rearrangement [5]. Let Ω ⊂ Rd, d ∈ {2, 3}, be a bounded domain with C∞-smooth boundary. Con- HΩ 2 sider the self-adjoint Dirichlet Laplacian D in the Hilbert space L (Ω) HΩ 2 1 dom D := H (Ω) ∩ H0 (Ω) (2.4) HΩ Du := −∆u . HΩ The operator D is lower semi-bounded and with purely discrete spectrum, and its lowest eigenvalue λ1(Ω) is strictly positive. Let RΩ HΩ −1 C HΩ (2.5) D(z) := D − z , z ∈ \ σ( D), HΩ RΩ 2 be the resolvent of D at the point z. D(z) acts on L (Ω) as a compact Ω R operator with kernel Gz : Ω × Ω → , called the Green function associated with HΩ D. In the sense of distributions one has Ω ((−∆ − z)Gz )(x,y)= δ(x − y), where δ(·) is the standard Dirac distribution in Rd, supported at the origin. We will also need to refer to the Green function S of the free Laplacian on Rd, Ω S defined in complete analogy to Gz . In fact, is given explicitly by − ln t , d = 2, S(x,y) := F (|x − y|) F (t) := 2π (2.6) with 1 ( 4πt , d = 3. Correspondingly, we define HΩ Ω S (2.7) z (x,y) := Gz (x,y) − (x,y) . Rd Ω Proposition 2.5. [5, Sect. 1] Let Ω ⊂ , λ1(Ω), and Gz be as above. Assume that z ∈ (−∞, λ1(Ω)). Then:

HΩ (i) for fixed y ∈ Ω, the function x 7→ z (x,y) is continuous on Ω, whence also Ω S HΩ (2.8) Gz (x,y)= (x,y)+ z (y,y)+ o(1) x → y ; Ω (ii) Gz (x,y) = 0 for any x ∈ ∂Ω and all y ∈ Ω; Ω (iii) Gz is positive in Ω × Ω.

In what follows, d B ≡ BR := {x ∈ R |x| < R} d stands for the ball centred at 0 ∈ R and being such that |Ω| = |B|. We shall also use the shorthand Ω HΩ hz,y := z (y,y) (2.9) Ω Ω R+ gz,y := Gz (·,y): Ω → for any fixed z < λ1(Ω) and y ∈ Ω. FABER-KRAHN INEQUALITIES WITH INTERACTIONS 9

Ω Here are relevant properties of the Green function Gz with respect to the sym- metric decreasing rearrangement. Proposition 2.6. Ω ⊂ Rd, d ∈ {2, 3}, be a bounded domain with C∞-smooth boundary. Let z < λ1(Ω) and y ∈ Ω. Then:

Ω 2 (i) gz,y ∈ L (Ω) ; Ω ∗ B (ii) 0 ≤ (g0,y) ≤ g0,0 ; Ω B (iii) hz,y ≤ hz,0 ; Ω (iv) h0,y < 0 if d = 3 .

Ω Proof. (i) Square-integrability of gz,y is a consequence of Proposition 2.5 (i) and of the fact that the function Ω ∋ x 7→ S(x,y) with fixed y ∈ Ω is square-integrable. (ii) follows directly from [5, Theorem 2.1 with p = 0]. (iii) Recall that R> 0 is the radius of the ball B. By[5, Lemma 2.3] there is R′ ≤ R B ′ R Ω Br such that hz,0 = hz,y. According to [5, proof of Lemma 2.3] the function r 7→ hz,0 B ′ B R Ω is increasing. Hence, we conclude that hz,0 ≥ hz,0 = hz,y. B  (iv) follows from (iii) (with z = 0) and from h0,y < 0 (see [6, §II.2.2]).

Denoting by F −1 the inverse of the function F in (2.6), we set Ω −1 Ω (2.10) Ry := F (−h0,y) > 0 . Ω The quantity Ry is actually the conformal radius of Ω at the point y when d = 2, or the harmonic radius of Ω at y when d = 3. (We refer to [7] and the references therein for a detailed discussion on conformal and harmonic radii.) In the case Ω B z = 0, Proposition 2.6 (iii) reduces to the inequality Ry ≤ R0 for all y ∈ Ω.

3. Dirichlet Laplacian on bounded domain with point interaction

In this Section we review the construction of the Dirichlet Laplacian on a bounded domain with a point interaction and we collect an amount of relevant properties. This combines two complementary languages: the operator theoretic self-adjoint extension scheme and the quadratic form approach. Let in the following Ω be a bounded domain in Rd with C∞-boundary, d ∈ {2, 3}, and let the point y ∈ Ω be fixed. It is standard to see (see, e.g., [13, Lemma 1]) that S 2 1 dom := u ∈ H (Ω) ∩ H0 (Ω) u(y) = 0 (3.1) S u := − ∆u is a densely defined, closed, symmetric operator on L2(Ω), with lower bound HΩ λ1(Ω) (the strictly positive, lowest eigenvalue of the Dirichlet Laplacian D from 10 V. LOTOREICHIK AND A. MICHELANGELI

(2.4)), with deficiency indices (1, 1) and with deficiency subspace S∗ Ω (3.2) ker( − z) = span {gz,y} , z < λ1(Ω) , Ω HΩ where gz,y is the Green function (2.9). It is also standard to see that D, obviously a self-adjoint extension of S, is precisely the Friedrichs extension of S (the operator domain of the former is contained in the form domain of the latter, namely in the H1-closure of dom S). As a consequence of these facts and of the Viˇsik-Birman decomposition formula (see, e.g., [36, Theorem 1]), any u ∈ dom S∗ decomposes as RΩ Ω Ω C S (3.3) u = u0 + c1 D(z)gz,y + c0gz,y, c0, c1 ∈ , u0 ∈ dom , RΩ HΩ for any z < λ1(Ω), where D(z) is the resolvent (2.5) of D. At each fixed pa- rameter z, the decomposition (3.3) is unique in terms of the u- and z-dependent elements c0, c1, u0, and S∗ S RΩ Ω Ω (3.4) u = u0 + c1z D(z)gz,y + (c1 + c0z)gz,y.

The self-adjoint extensions of S form a one-real-parameter family {Sβ |β ∈ R}∪ HΩ { D}, where S RΩ Ω Ω S C dom β := u = u0 + c β D(z)gz,y + gz,y u0 ∈ dom , c ∈ (3.5) S u := Sn∗u .  o β Formula (3.5) is a direct application to the present unit-deficiency-index case of the general classification formula for the self-adjoint extensions of a lower semi- bounded (and densely defined) symmetric operator (see, e.g., [36, Theorem 5] or HΩ [61, Sect. 14.8]). The Friedrichs extension D formally corresponds to β = ∞.

The extensions Sβ are equivalently characterised in terms of their quadratic forms: the form sβ associated with each Sβ is given by Ω 1 C dom sβ = u = v + ξgz,y v ∈ H0 (Ω),ξ ∈ (3.6) 2 Ω 2 2 2 2 sβ[u] = k∇ vk 2 + βk g k 2 · |ξ| + z kuk 2 − kvk 2 L (Ω) z,y L (Ω) L (Ω) L (Ω) (see, e.g., [36, Theorem 7] or [61, Theorem 14.24]).  Ω The combination of the asymptotics (2.8) for gz,y with the decomposition (3.5) implies that any function u ∈ dom Sβ behaves in the vicinity of the point y as Ω 2 S Ω (3.7) u(x)= c βkgz,ykL2(Ω) + (x,y)+ hz,y + o(1), x → y. This short-scale behaviour, as argued already in [13, Sect. III], has the form (3.8) u(x)= c (S(x,y)+ α + o(1)) , x → y which is typical of the low-energy scattering of a quantum particle over a scatter- ing centre with zero-range interaction and with s-wave scattering length (−α)−1, in suitable units, as originally identified by Bethe and Peierls [10] (whence also the FABER-KRAHN INEQUALITIES WITH INTERACTIONS 11 nomenclature of Bethe-Peierls contact condition – see, e.g., [56, Sect. 2]). It is there- fore meaningful to re-parametrise the Sβ’s in terms of the new, physical grounded extension parameter

Ω 2 Ω (3.9) α = βkgz,ykL2(Ω) + hz,y.

Upon plugging (3.9) into (3.5) and (3.6), we can summarise the above considera- tions as follows.

Proposition 3.1. Let Ω be a bounded domain in Rd with C∞-boundary, d ∈ {2, 3}, and let y ∈ Ω. Correspondingly, let S be as in (3.1).

(i) The self-adjoint extensions of S in L2(Ω) constitute the one-real-parameter family HΩ R HΩ α,y α ∈ ∪ { D} HΩ  with each element α,y given, fixed z < λ1(Ω), by Ω −2 Ω RΩ Ω Ω u = u0 + ckg k 2 α − h (z)g + cg dom HΩ = z,y L (Ω) z,y D z,y z,y α,y S C (3.10) ( for some u0 ∈ dom , c ∈ ) HΩ Ω −2 Ω RΩ Ω Ω Ω α,yu = −∆u0 + ckgz,ykL2(Ω) α − hz,y z D(z)gz,y + gz,y + czgz,y . Ω HΩ   (ii) The quadratic form hα,y of α,y is given, fixed z < λ1(Ω), by Ω Ω 1 C dom hα,y = u = v + ξgz,y | v ∈ H0 (Ω),ξ ∈ (3.11) Ω 2 Ω 2 2 2 hα,y[u] = k∇ vkL2(Ω) + α − hz,y |ξ| + z k ukL2(Ω) − kvkL2(Ω) .   HΩ Fixed the parameter z, the decompositions in (3.10) and (3.11) of u are unique. D is the Friedrichs extension.

HΩ We shall refer to each α,y as a self-adjoint Dirichlet Laplacian with point interaction on Ω with interaction centre y ∈ Ω and interaction strength (−α)−1. It is convenient to introduce further shorthand notation: Ω Ω gy = g0,y Ω Ω hy = h0,y (3.12) Ω Ω hα,y = hα,0,y Ω Ω −2 γy = kgy kL2(Ω) . This allows one to re-write (3.10) and (3.11), with the choice z = 0, respectively as

u = u + cγΩ α − hΩ RΩ(0)gΩ + cgΩ dom HΩ = 0 y y D y y α,y for some u ∈ dom S, c ∈ C (3.13)  0   HΩ Ω Ω Ω α,yu = −∆u0 + cγy α − hy gy  12 V. LOTOREICHIK AND A. MICHELANGELI and Ω DΩ Ω 1 C dom hα,y = y := u = v + ξgy | v ∈ H0 (Ω),ξ ∈ (3.14) Ω 2 Ω 2 hα,y[u] = k∇vkL2(Ω) + α − hy |ξ| .  HΩ Let us work out certain useful properties of the Hamiltonian α,y. Proposition 3.2.

HΩ (i) α,y is lower semi-bounded and has compact resolvent. HΩ Ω (ii) α,y ≥ 0 if and only if α ≥ hy . HΩ (iii) The map α 7→ α,y is a non-decreasing operator-valued function in the sense of ordering of forms. HΩ (iv) The map α 7→ α,y is continuous in the strong resolvent sense. HΩ HΩ (v) α,y → D as α → +∞ in the strong resolvent sense.

HΩ HΩ Proof. (i) Since α,y and D are both self-adjoint extensions of the symmetric op- erator S, and S has unit deficiency indices, the difference of their resolvents is HΩ a rank-one operator. Hence, semi-boundedness of α,y and compactness of its HΩ resolvent follow from respective properties of D. Ω HΩ Ω (ii) When α ≥ hy , non-negativity of α,y follows from the fact that hα,y[u] ≥ 0 for Ω Ω all u ∈ dom hα,y, which can be seen in (3.14). Conversely, when α < hy , (3.14) yields Ω Ω Ω hα,y[gy ]= α − hy < 0. HΩ Hence, the min-max principle implies that the negative spectrum of α,y is non- empty. Ω (iii) The claimed property follows from the fact that dom hα,y is independent of α DΩ hΩ hΩ and that for any u ∈ y one has α1,y[u] ≤ α2,y[u] whenever α1 ≤ α2. HΩ (iv) Continuity of the operator-valued function α 7→ α,y in the strong resolvent Ω sense is a consequence of continuity of the scalar-valued function α 7→ hα,y[u] for DΩ any u ∈ y combined with [42, Theorem XIII.3.6]. (v) The claim (iii), combined with the monotone convergence theorem for qua- HΩ HΩ dratic forms [59, Theorem S.14] imply α,y → D as α → +∞ in the strong resolvent sense. 

HΩ 4. The lowest eigenvalue and the ground state of α,y

α In this Section we discuss the properties of the lowest eigenvalue λ1 (Ω,y) of the HΩ Hamiltonian α,y, and of the corresponding eigenfunction. FABER-KRAHN INEQUALITIES WITH INTERACTIONS 13

For our later purposes, crucial features to analyse are the dependence of such objects (and associated quantities) on the extension parameter α, as well as the convenient representations of the ground state eigenfunction. We start with deriving a first set of results in this spirit.

Proposition 4.1. Let Ω be a bounded domain in Rd with C∞-boundary, d ∈ {2, 3}, and R HΩ α let y ∈ Ω, α ∈ . Correspondingly, let α,y be as in Proposition 3.1, and let λ1 (Ω,y) be its lowest eigenvalue. (As a reference, let us recall that λ1(Ω) > 0 is the lowest eigenvalue HΩ of the Dirichlet Laplacian D from (2.4)). One has: α (i) λ1 (Ω,y) < λ1(Ω); HΩ (ii) the spectrum of α,y in the interval (−∞, λ1(Ω)) consists of a unique simple eigenvalue; R α (iii) the function ∋ α 7→ λ1 (Ω,y) is continuous; α α (iv) lim λ (Ω,y)= −∞ and lim λ (Ω,y)= λ1(Ω). α→−∞ 1 α→+∞ 1 α HΩ α Let now u1 be the eigenfunction of α,y corresponding to λ1 (Ω,y) (up to a multiplicative constant). Then:

(v) one has the representation α Ω α C u1 = cgz,y with z = λ1 (Ω,y) and c ∈ \ {0} , α hence u1 can be chosen to be positive on Ω; α (vi) when in particular λ1 (Ω,y) > 0, one has the representation α Ω 1 u1 = v + ξgy with v ∈ H0 (Ω) , v ≥ 0 on Ω , and ξ ≥ 0 .

1 2 HΩ Proof. (i) Let u1 ∈ H0 (Ω) ∩ H (Ω) be the ground state eigenfunction of D. With- out loss of generality, one can assume that u1 is positive on Ω. Consider a family of test functions Ω R u1,ε := u1 + εgy , ε ∈ . DΩ L Owing to (3.14), u1,ε ∈ y . Differentiating the scalar-valued function (ε) := Ω 2 hα,y[u1,ε] − λ1(Ω)ku1,εkL2(Ω) at ε = 0 gives

L′ Ωd (0) = −2λ1(Ω) u1gy x . ZΩ Ω L′ As gy is positive on Ω (Proposition 2.5 (iii) and (2.9)), we conclude that (0) < 0 α and the min-max principle yields the inequality λ1 (Ω,y) < λ1(Ω). HΩ HΩ (ii) As argued already for the proof of Proposition 3.2 (i), α,y and D differ in the resolvent sense by a rank-one operator. From this, and from the fact that HΩ inf σ( D) = λ1(Ω), one can deduce [12, §9.3, Theorem 3] that the rank of the HΩ spectral projection for α,y corresponding to the interval (−∞, λ1(Ω)) is either 0 α or 1. Taking into account that λ1 (Ω,y) < λ1(Ω), we eventually conclude that this rank equals to one, which is equivalent to the claim. 14 V. LOTOREICHIK AND A. MICHELANGELI

α (iii) The continuity of λ1 (Ω,y) with respect to α directly follows from the spectral convergence result [42, Theorem VIII.1.14] and from Proposition 3.2 (iv). (iv) By the min-max principle and (3.14), hΩ [gΩ] λα ,y α,y y γΩ α hΩ α→−∞ , 1 (Ω ) ≤ Ω 2 = y − y −−−−−→ −∞ kgy kL2(Ω)  whence the first of the two claimed limits follows. The second limit is a conse- quence of the strong resolvent convergence in Proposition 3.2 (v) and the spectral convergence result [42, Theorem VIII.1.14]. (v) For the proof of this part and of the next one, let us switch to the shorthand α α S∗ λ1 = λ1 (Ω,y). As obviously u1 ∈ dom , then α RΩ Ω Ω u1 = u0 + c1 D(λ1)gλ1,y + c0gλ1,y for suitable u0, c0, c1 (owing to (3.3) above), whence HΩ λ uα S∗ λ u c RΩ λ gΩ ( α,y − 1) 1 = − 1 0 + 1 D( 1) λ1,y HΩ λ u c RΩ λ gΩ . = ( D − 1) 0 + 1 D( 1) λ1,y HΩ uα = λ uα u + c RΩ(λ )gΩ = 0  λ < inf σ(HΩ) Now, α,y 1 1 1 implies 0 1 D 1 λ1,y , because 1 D . uα = c gΩ c > 0 uα Therefore, 1 0 λ1,y. Moreover, the choice 0 yields a positive 1 , as gΩ > 0 λ1,y (Proposition 2.5 (iii)). Ω α (vi) By assumption λ1 > 0 and therefore (Proposition 3.2 (ii)) α ≥ hy . As u1 ∈ HΩ Ω dom α,y ⊂ dom hα,y, then α Ω u1 = v + ξgy 1 C for some v ∈ H0 (Ω) and ξ ∈ (owing to (3.14) above). It is not restrictive to assume that ξ ≥ 0 and that consequently (based on (v)) v is real-valued. It remains Ω to show that v ≥ 0. To this aim, we pick the test function u := |v| + ξgy . One has 2 2 Ω 2 Ω 2 kukL2(Ω) = kvkL2(Ω) + 2ξ(|v|, gy )L2(Ω) + |ξ| kgy kL2(Ω) (4.1) 2 Ω 2 Ω 2 α 2 ≥ kvkL2(Ω) + 2ξ(v, gy )L2(Ω) + |ξ| kgy kL2(Ω) = ku1 kL2(Ω), Ω where gy > 0 was used (owing to (2.9), (3.12), and Proposition 2.5 (iii)). Then

Ω 2 Ω 2 Ω α h [u] k∇vkL2(Ω) + α − hy |ξ| h [u ] λ ≤ α,y = ≤ α,y 1 = λ , 1 2 2  α 2 1 kukL2(Ω) kukL2(Ω) ku1 kL2(Ω) having applied the min-max principle in the first step, the identity k∇|v|kL2(Ω) = Ω k∇vkL2(Ω) in the second, and the inequalities α ≥ hy and (4.1) in the third. The α eigenvalue λ1 being simple, then necessarily u1 = u, and thus v = |v| is non- negative. 

Remark 4.2. Proposition 4.1 shows that when passing from the Dirichlet Laplacian HΩ HΩ D to any of its singular perturbations α,y, an eigenvalue is always created below FABER-KRAHN INEQUALITIES WITH INTERACTIONS 15 the threshold λ1(Ω) irrespective of the sign and magnitude of α. Thus, each self- HΩ S adjoint extension α,y of the symmetric operator defined in (3.1) has bottom HΩ α S inf σ( α,y) = λ1 (Ω,y) strictly below the bottom of the Friedrichs extension of , HΩ i.e., below inf σ( D) = λ1(Ω). Or, in other words, there are no other extensions besides the Friedrichs one with the same lower bound of S. The above behaviour is not generic. For instance, singular perturbations of the self-adjoint Laplacian on Rd when d = 3 only produce an eigenvalue below the Friedrichs threshold for a specific range of the interaction strength [1, Chapter I.1] (and the same holds for singular perturbations of the Schr¨odinger operator −∆+ q|x|−1 with q > 0 – see, e.g., [1, Chapter I.2] or [34]), while when d = 2 every self-adjoint extension does have an eigenvalue below the Friedrichs threshold [1, Chapter I.5]. A general characterisation of the possibility of having or not a self-adjoint extension with the lower bound strictly below the Friedrich’s lower bound may be found in [35].

Ω Next we focus on the map z 7→ hz,y defined in (2.7) and (2.9) above. We shall α show that it determines a suitable spectral condition on λ1 (Ω,y) and displays convenient monotonicity; based on such properties we can finally deduce the α strict monotonicity of λ1 (Ω,y) with respect to α.

Proposition 4.3. Same assumptions as in Proposition 4.1. Then:

α Ω (i) one has λ1 (Ω,y)= z with z ∈ (−∞, λ1(Ω)) if and only if hz,y = α; Ω (ii) the function (−∞, λ1(Ω)) ∋ z 7→ hz,y is continuous and monotone increasing;

α1 α2 (iii) one has λ1 (Ω,y) < λ1 (Ω,y) for α1 < α2.

Ω Ω S∗ HΩ Proof. (i) If hz,y = α, then the function u = gz,y ∈ ker( − z) belongs to dom α,y α (Proposition 3.1 (i)). Proposition 4.1 (v) then implies λ1 (Ω,y) = z. Conversely, HΩ α HΩ Ω α if α,yu = λ1 (Ω,y)u for some u ∈ dom α,y, then u = cgz,y with z = λ1 (Ω,y) and some c 6= 0 (Proposition 4.1 (v)), whence, using again the characterisation of HΩ Ω y ∈ dom α,y from Proposition 3.1 (i), α = hz,y.

(ii) Let z1 < z2 < λ1(Ω) be arbitrary. Owing to Proposition 4.1 (iii) and (iv), such values are surely attained (a priori multiple times) by the function R ∋ α 7→ α α1 λ1 (Ω,y), and in fact it is always possible to select α1 < α2 such that z1 = λ1 (Ω,y) α2 Ω and z2 = λ1 (Ω,y). Applying part (i) one then finds hzj ,y = αj, j ∈ {1, 2}. Thus, Ω Ω hz1,y < hz2,y.

α1 α2 (iii) Proposition 3.2 (iii) and the min-max principle imply λ1 (Ω,y) ≤ λ1 (Ω,y) α1 whenever α1 < α2. We are left with excluding the case of equality. If λ1 (Ω,y)= α2 α λ1 (Ω,y) =: z held, then for any α ∈ (α1, α2) we would have λ1 (Ω,y) = z. Ω Hence, by (i) we would get α = hz,y for all α ∈ (α1, α2), thus yielding an obvious contradiction.  16 V. LOTOREICHIK AND A. MICHELANGELI HΩ 5. The Faber-Krahn inequality for the operator α,y

We are finally in the condition to formulate and prove the first main result of the present work, namely the Faber-Krahn inequality for the Dirichlet Laplacian on a bounded domain with a point interaction.

Theorem 5.1. Let Ω be a bounded domain in Rd with C∞-boundary, d ∈ {2, 3}, and R α let y ∈ Ω, α ∈ . Correspondingly, let λ1 (Ω,y) be the lowest eigenvalue of the operator HΩ B Rd 0 Rd α,y qualified in Proposition 3.1. Let ⊂ be a ball centred at the origin ∈ and satisfying |B| = |Ω|. Then α B 0 α λ1 ( , ) ≤ λ1 (Ω,y) .

Thus: under fixed volume of the domain and fixed interaction strength parameter α ∈ R, and with generic position of point interaction centre, the principal eigen- value is minimised by the ball with the point interaction supported at the ball’s centre. The ordinary Faber-Krahn inequality is retrieved from Theorem 5.1 in the limit α → +∞.

Remark 5.2. We have already argued in Section 1 that in a sense the result ex- pressed by Theorem 5.1 has the same spirit of the optimal placement of an ‘ob- stacle’ or ‘impurity’ within the considered domain. In fact, Theorem 5.1 implies that the position y of the point interaction’s support which minimises the lowest HB eigenvalue of the Hamiltonian α,y on the ball is precisely the ball’s centre. This holds irrespective of the sign of α, hence of the attractive or repulsive nature of the interaction supported at y (here attraction or repulsion is meant in the ordinary sense of scattering theory, thus based on the sign of the scattering length). This is not in contradiction with the fact (see the already-mentioned analysis [38]) that the optimal placement, inside the ball, of a bounded potential Vy localised around HΩ y, in order to minimise the lowest eigenvalue of D + Vy, is achieved by putting y at the centre in case of negative potential well, and y at the boundary in case of positive bump-like potential. In this respect, the point interaction modelled by HB α,y has the same effect of a negative and localised potential well (whence indeed HΩ HΩ the lowering of the lowest eigenvalue of α,y with respect to D, Remark 4.2). In fact, by inspection of the explicit (computed in [13, Sect. 4]) one HΩ sees that they concentrate around y, analogously to the eigenfunctions of D + Vy with Vy < 0.

For the proof of Theorem 5.1 it is convenient to introduce the new shorthand α α B 0 λΩ := λ1 (Ω,y) , λB := λ1 ( , ) at fixed α ∈ R and y ∈ Ω. The thesis to prove is therefore λB ≤ λΩ. Prior to that, we can rule out the possibility λB > 0 and λΩ ≤ 0. FABER-KRAHN INEQUALITIES WITH INTERACTIONS 17

Lemma 5.1. Under the assumptions of Theorem 5.1 it is impossible that simultaneously λB > 0 and λΩ ≤ 0.

Proof. Assuming λB > 0, Proposition 3.2 (ii) combined with Proposition 4.3 (i) B B Ω imply α > h0 . Moreover, h0 ≥ hy (Proposition 2.6 (iii) and (3.12)). Hence, α > Ω  hy . Then Proposition 3.2 (ii) and Proposition 4.3 (i) imply λΩ > 0.

Theorem 5.1 is therefore to be proved in the only possible non-trivial scenario that λB and λΩ have the same sign (either λB, λΩ < 0, or λB, λΩ > 0), because the case λB ≤ 0, λΩ ≥ 0 is trivial, and the case λB > 0, λΩ ≤ 0 is impossible. At this point, as announced in Section 1, we find it instructive to present two al- ternative routes. The first applies to all cases and is entirely based on the Bandle’s Ω B inequality hz,y ≤ hz,0 underlying Proposition 2.6 (iii). For the first proof of The- orem 5.1, we exploit Bandle’s inequality in combination with our monotonicity HΩ analysis for α,y developed in Section 4. Proof of Theorem 5.1 – first version, based on the Bandle’s inequality. B Owing to Proposition 4.3 (i), hΩ = α = h . Moreover, Proposition 2.6 (iii) λΩ,y λB,0 hΩ ≤ hB hΩ ≤ hΩ implies λB,y λB,0. Hence, λB,y λΩ,y. Taking into account the increasing Ω monotonicity of the function (−∞, λ1(Ω)) ∋ z 7→ hz,y (Proposition 4.3 (ii)), we conclude from hΩ ≤ hΩ that λB ≤ λ .  λB,y λΩ,y Ω

Next, we present an independent proof, applicable to the case λΩ > 0, which has a two-fold virtue. First, it provides a more direct adaptation of the origi- nal Faber-Krahn inequality’s demonstration scheme to the present playground of point interaction Hamiltonian. Second, it is completely independent of Bandle’s inequality and in fact reproves it by alternative means (Corollary 5.3 below).

Proof of Theorem 5.1 – second version for λΩ > 0, Bandle-inequality-independent. α R HΩ Let u1 :Ω → be the eigenfunction of α,y corresponding to its lowest eigen- α value λΩ. By additional assumption, λΩ > 0. Then u1 decomposes as α Ω u1 = v + ξgy 1 ∗ for some v ∈ H0 (Ω), v ≥ 0 and ξ ≥ 0 (Proposition 4.1 (vi)). Furthermore, v ∈ 1 B H0 ( ) (Corollary 2.3 (iii)), whence ∗ B B v + ξg0 ∈ dom hα,0 (formula (3.14) above). Then applying the min-max principle to the test function ∗ B v + ξg0 yields B ∗ B hα,0 v + ξg0 (5.1) λB ≤ . v∗ ξgB 2 k + 0 kL2(B)

It suffices to show that this upper bound does not exceed λΩ. To this aim we estimate the numerator and the denominator separately. 18 V. LOTOREICHIK AND A. MICHELANGELI

Concerning the numerator, we find B ∗ B ∗ 2 B 2 hα,0 v + ξg0 = k∇v kL2(B) + α − h0 |ξ| 2 B 2 (5.2)   ≤ k∇vkL2(Ω) + α − h0 |ξ| 2 Ω 2 Ω Ω ≤ k∇vkL2(Ω) + α − hy |ξ| = hα,y v + ξgy , having used (3.14) in the first step, the P´olya-Szeg˝oinequality   (Corollary 2.3 (iii)) in the second, the bound B Ω α − h0 ≤ α − hy . in the third step (which in turn follows from Proposition 2.6 (iii) and (3.12)), and again (3.14) in the last step. Concerning the denominator in (5.1), first we observe that Ω Ω −2 Ω ∗ −2 B −2 B (5.3) γy = kgy kL2(Ω) = k(gy ) kL2(B) ≥ kg0 kL2(B) = γ0 , as a consequence of (3.12), of Corollary 2.3 (i), and of Proposition 2.6 (ii). We then estimate ∗ B 2 ∗ 2 ∗ B B −1 2 kv + ξg0 kL2(B) = kv kL2(B) + 2ξ(v , g0 )L2(B) + (γ0 ) |ξ| 2 ∗ B Ω −1 2 = kvkL2(Ω) + 2ξ(v , g0 )L2(B) + (γy ) |ξ| 2 ∗ Ω ∗ Ω −1 2 (5.4) ≥ kvkL2(Ω) + 2ξ(v , (gy ) )L2(B) + (γy ) |ξ| 2 Ω Ω −1 2 ≥ kvkL2(Ω) + 2ξ(v, gy )L2(Ω) + (γy ) |ξ| Ω 2 = kv + ξgy kL2(Ω) . For (5.4) we used that v, v∗ ≥ 0 and ξ ≥ 0, and we applied Corollary 2.3 (i) and (5.3) in the second step, Proposition 2.6 (ii) and (3.12) in the third, and the Hardy- Littlewood inequality (Corollary 2.3 (ii)) in the fourth. Last, plugging (5.2) and (5.4) into (5.1), and using (3.14), yields Ω Ω hα,y v + ξgy λB ≤ = λ , kv + ξgΩk2 Ω  y L2(Ω) which completes the proof. 

Corollary 5.3. Let Ω be a bounded domain in Rd with C∞-boundary, d ∈ {2, 3}, and let y ∈ Ω, α ∈ R. Let B ⊂ Rd be a ball centred at the origin 0 ∈ Rd and satisfying B Ω B | | = |Ω|. For z ∈ (0, λ1(Ω)), the relative Green functions hz,y and hz,0 defined in (2.7) and (2.9) satisfy B Ω hz,0 ≥ hz,y .

Proof. As announced, this is rather a corollary of the second proof of Theorem α R 5.1. A generic z ∈ (0, λ1(Ω)) can be written as z = λ1 (Ω,y) for some α ∈ , α because the function α 7→ λ1 (Ω,y) attains with continuity all real values below α λ1(Ω) (Proposition 4.1 (iii) and (iv)). For such α and such positive value λ1 (Ω,y), FABER-KRAHN INEQUALITIES WITH INTERACTIONS 19

α α B 0 the second proof of Theorem 5.1 shows that λ1 (Ω,y) ≥ λ1 ( , ). Now, on the one hand B Ω Ω α = h α B 0 0 = h α = h λ1 ( , ), λ1 (Ω,y),y z,y α α B 0 (Proposition 4.3 (i)). On the other hand, the inequality λ1 (Ω,y) ≥ λ1 ( , ) implies B B B h 0 = h α 0 ≥ h α B 0 0 z, λ1 (Ω,y), λ1 ( , ), (Proposition 4.3 (ii)). Combining these formulas yields the conclusion. 

B Ω It is admittedly remarkable that the estimate hz,0 ≥ hz,y (Proposition 2.6 (iii), Corollary 5.3), involving Green functions of Dirichlet Laplacians, is so intimately connected with the spectral theory of the Hamiltonian with a point interaction in bounded domain.

6. Systems with Coulomb interactions

Let us move now on to the second main focus of the present work, namely Faber- Krahn-type inequalities for one- and two-body systems with Coulomb interac- tions.

6.1. One-body case. As outlined already in Section 1, we shall establish the fol- lowing result: under fixed volume of the domain and fixed strength of the at- tractive Coulomb interaction between the particle and the impurity, the principal eigenvalue is minimised by the ball with the impurity located in its center. For the present setting, let Ω ⊂ R3 be a bounded C∞-smooth domain and let the point y ∈ Ω. For q ∈ R, we consider, on the Hilbert space L2(Ω), the quadratic form Ω 1 dom tq,y := H0 (Ω) (6.1) 2 Ω 2 |u(x)| t [u] := k∇uk 2 − q dx . q,y L (Ω) |x − y| ZΩ Ω In tq,y, the symmetric perturbation term |u(x)|2 H1(Ω) ∋ u 7→ dx 0 |x − y| ZΩ 1 2 is form-bounded with respect to the kinetic energy term H0 (Ω) ∋ u 7→ k∇ukL2(Ω) Ω with bound < 1 (see, e.g., [42, Lemma VI.4.8b]). Therefore, tq,y is closed and semi- ∞ Ω bounded and the subspace C0 (Ω) is a core for it. As a consequence, tq,y is the quadratic form of a uniquely determined self-adjoint and lower semi-bounded 2 TΩ operator on L (Ω), which we shall denote by q,y. 1 2 Now, the embedding H0 (Ω) ⊂ L (Ω) is compact (see, e.g., [37, Theorem 1.4.3.2]), TΩ therefore by, e.g., [61, Proposition 10.6] the spectrum of q,y is purely discrete. We q TΩ denote by µ1(Ω,y) the lowest eigenvalue of q,y. 20 V. LOTOREICHIK AND A. MICHELANGELI

Theorem 6.1. Let Ω ⊂ R3 be a bounded C∞-smooth domain and let y ∈ Ω. Let B ⊂ R3 be the ball centred at the origin 0 ∈ R3 and satisfying |B| = |Ω|. Let q ≥ 0. Then q B 0 q µ1( , ) ≤ µ1(Ω,y) .

Remark 6.2. The choice q = 0 in Theorem 6.1 corresponds to the ordinary Faber- Krahn inequality.

Proof of Theorem 6.1. Let us introduce the shorthand q q B 0 µΩ := µ1(Ω,y) , µB := µ1( , ) .

The thesis to prove is therefore µB ≤ µΩ. We also set q V : R3 → R+ ,V (x) := q,y q,y |x − y| and observe that

∗ (6.2) Vq,y = Vq,0 .

q 1 TΩ Let u1 ∈ H0 (Ω) be the eigenfunction of q,y corresponding to its lowest eigen- q value µΩ. Non-restrictively, the function u1 can be chosen to be real-valued and non-negative on Ω, and moreover it satisfies

Ω q q 2 (6.3) tq,y[u1] = µΩku1kL2(Ω) .

q In the following we shall simply write u := u1. Let u ∈ H1(R3) be the extension of u by zero. Owing to Proposition 2.2 (i) and (iii), also u∗ ∈ H1(R3) with

e ∗ ku k 2 R3 = kuk 2 R3 e L ( ) L ( ) (6.4) ∗ k∇u kL2(R3) ≤ k∇ukL2(R3) . e e Moreover, owing to Lemma 2.1e (iv) and (v), e supp u∗ ⊂ B (6.5) (u∗)2 = (u2)∗ . e ∗ 1 R3 ∗ B From u ∈ H ( ) and supp u ⊂ e, we infere that ∗ 1 v u B H B , (6.6) e e := | ∈ 0 ( ) and using (6.5) and the Hardy-Littlewoode inequality (Proposition 2.2 (ii)) we find

∗ 2 2 ∗ 2 (6.7) Vq,0(x)(u (x)) dx = Vq,0(x)(u ) (x)dx ≥ Vq,y(x)u (x)dx . R3 R3 R3 Z Z Z e e e FABER-KRAHN INEQUALITIES WITH INTERACTIONS 21

TB Now, using v as a trial function for the Rayleigh quotient for the operator q,0, we finally find

B ∗ 2 0 ∗ 2d k∇u kL2(R3) − Vq, (x)|u (x)| x tq,0[v] R3 µB ≤ = Z kvk2 kuk2 L2(B) e L2(R3) e 2 2d k∇ukL2(R3) − Vq,y|u(x)| x Ω eR3 tq,y[u] ≤ Z = = µ , kuk2 kuk2 Ω e L2(R3) e L2(Ω) where we used the min-max principle in the first step, (6.4) in the second, (6.6), e (6.7) in the third, (6.1) in the fourth, and (6.3) in the last. 

6.2. Two-body case. Next, we examine the model for a two-body quantum sys- tem on a bounded domain with inter-particle attractive Coulomb interaction. As before, let Ω ⊂ R3 be a C∞-smooth domain. For q ∈ R, we consider, on the Hilbert space L2(Ω × Ω), the quadratic form Ω 1 dom tq := H0 (Ω × Ω) (6.8) 2 Ω 2 |u(x1,x2)| t [u] := k∇uk 2 − q dx dx . q L (Ω×Ω) |x − x | 1 2 ZΩ×Ω 1 2 Ω In tq , the symmetric perturbation term |u(x ,x )|2 H1(Ω × Ω) ∋ u 7→ 1 2 dx dx 0 |x − x | 1 2 ZΩ×Ω 1 2 is form-bounded with respect to the kinetic energy term 1 2 H0 (Ω × Ω) ∋ u 7→ k∇ukL2(Ω×Ω) Ω with the bound < 1 (see, e.g., [41, Lemma 4]). Therefore, tq is closed and semi- ∞ Ω bounded and the subspace C0 (Ω × Ω) is a core for it. As a consequence, tq is the quadratic form of a uniquely determined self-adjoint and lower semi-bounded 2 TΩ 1 operator on L (Ω), which we shall denote by q . As the embedding H0 (Ω×Ω) ⊂ 2 TΩ q L (Ω × Ω) is compact, the spectrum of q is purely discrete. We denote by ν1 (Ω) the lowest eigenvalue. Theorem 6.3. Let Ω ⊂ R3 be a C∞-smooth domain and let B ⊂ R3 be a ball centred at the origin 0 ∈ R3 and satisfying |B| = |Ω|. Let q ≥ 0. Then q/2 B 0 q 2µ1 ( , ) ≤ ν1 (Ω) . Remark 6.4. With the choice q = 0 in Theorem 6.3, namely when the two particles 0 are uncoupled, the quantity ν1 (Ω) is obviously twice as the ground state energy of a single particle confined in Ω, namely, with Hamiltonian given by the Dirichlet Laplacian on Ω. In this case Theorem 6.3 implies 0 0 B 0 B 2λ1(Ω) = ν1 (Ω) ≥ 2µ1( , ) = 2λ1( ) , 22 V. LOTOREICHIK AND A. MICHELANGELI and one retrieves the ordinary Faber-Krahn inequality.

Proof. Let us introduce the shorthand q q/2 B 0 νΩ := ν1 (Ω) , µB := µ1 ( , ) .

The thesis to prove is therefore 2µB ≤ µΩ. We also set R3 R+ q Vq,x1 : → ,Vq,x1 (x2) := |x1 − x2| and observe that ∗ 0 (6.9) Vq,x1 = Vq, independently of x1. q 1 TΩ Let u1 ∈ H0 (Ω × Ω) be the eigenfunction of q corresponding to its lowest eigen- q value νΩ. Non-restrictively the function u1 can be chosen to be real-valued and non-negative, and moreover it satisfies Ω q q 2 (6.10) tq [u1] = νΩku1kL2(Ω×Ω) . It can also be easily shown that q q (6.11) u1(x1,x2) = u1(x2,x1) (x1,x2 ∈ Ω) (bosonic and absolute ground state coincide). In the following we shall simply q write u := u1. Let u ∈ H1(R3 × R3) be the extension of u by zero, and let us consider the Steiner rearrangements S1u and S2u of u respectively with respect to the first and the sec- onde three-dimensional variable (as defined in (2.3)), as well as, correspondingly, the further rearrangementse eS1Se2u and S2S1u with respect to the other variable. By construction, e e supp(S1u) ⊂ B × Ω

(6.12) supp(S2u) ⊂ Ω × B e supp(S1S2u) ⊂ B × B . e Moreover, as a consequence of the exchange symmetry (6.11) (Proposition 2.4 (iii)), e (6.13) S1S2u = S2S1u .

1 R3 R3 Owing to Proposition 2.4, S1u, S2u, Se1S2u, S2S1eu ∈ H ( × ) with

kS1S2ukL2(R3×R3) = kukL2(R3×R3) (6.14) e e e e k∇(S1S2u)kL2(R3×R3) ≤ k∇ukL2(R3×R3) . e e B Next, we observe that for (almoste every) fixedex1 ∈ (and trivially for x1 ∈ 3 3 R \ B), the function (S1u)(x1, ·) is square integrable on the whole R and its sym- metric decreasing rearrangement is nothing but (S2S1u)(x1, ·). Thus, Proposition e e FABER-KRAHN INEQUALITIES WITH INTERACTIONS 23

2.2 (i) gives

2 ∗ 2 dx2 (S2S1u)(x1,x2) = dx2 (S1u)(x1, ·) (x2) R3 R3 (6.15) Z Z  2 e = dx2 (S1ue)(x1,x2) , R3 Z

e and analogously,

2 2 (6.16) dx1 (S1S2u)(x1,x2) = dx2 (S2u)(x1,x2) . R3 R3 Z Z

e e In turn, (6.15)-(6.16) (together with (6.13)) imply

x 2 d d |(S1S2u)( )| V q ,0(x1)+ V q ,0(x2) x1 x2 R3 R3 2 2 ZZ ×   x 2 d d x 2 d d = |(eS1u)( )| V q ,0(x1) x1 x2 + |(S2u)( )| V q ,0(x2) x1 x2 , R3 R3 2 R3 R3 2 ZZ × ZZ × e e having used the shorthand x = (x1,x2). For each summand of the r.h.s. above, the Hardy-Littlewood inequality (Proposition 2.2 (ii)) and (6.9) yield

x 2 q d x 2 q d |(S1u)( )| V ,0(x1) x1 ≥ |u( )| V ,x2 (x1) x1 R3 2 R3 2 Z Z

x 2 q d x 2 q d |(S2ue)( )| V ,0(x2) x2 ≥ |ue( )| V ,x1 (x2) x2 , R3 2 R3 2 Z Z e e and obviously V q (x1)+ V q (x2)= Vq,x (x2), whence finally 2 ,x2 2 ,x1 1

x 2 d d |(S1S2u)( )| V q ,0(x1)+ V q ,0(x2) x1 x2 R3 R3 2 2 ZZ × (6.17)   x 2 dx ≥ |ue( )| Vq,x1 (x2) . R3 R3 Z × e Now, by construction (see (6.12) and (6.14) above),

1 B B (6.18) v := (S1S2u)|B×B ∈ H0 ( × ) .

Using v as a trial function for the Rayleighe quotient for the self-adjoint operator

B B ½ T q 0 ⊗ ½ + ⊗ T q 0 2 , 2 , 24 V. LOTOREICHIK AND A. MICHELANGELI acting in the Hilbert space L2(B × B) and we finally find

2 x 2 q q dx k∇vkL2(B×B) − |v( )| V ,0(x1)+ V ,0(x2) B×B 2 2 2µB ≤ Z 2 h i kvkL2(B×B)

2 x 2 q q dx k∇(S1S2u)kL2(R3×R3) − |(S1S2u)( )| V ,0(x1)+ V ,0(x2) R3 R3 2 2 = Z × k(S S u)k2 h i e 1 2 L2(eR3×R3) 2 x 2dx k∇ukL2(R3×R3) − Vq,x1 (x2)|u( )| R3 R3 e ≤ Z × kuk2 e L2(R3×R3) e Ω tq [u] = 2 = νeΩ , kukL2(Ω×Ω) where we used (twice) the min-max principle in the first step, (6.18) in the second, (6.14) and (6.17) in the third, (6.8) in the fourth, and (6.10) in the last. 

Acknowledgement. We are most grateful to the International Centre for Mathe- matical Research FBK/CIRM Trento, under the auspices of which a large part of this project was carried on within a research in pairs programme. This work is also partially supported by the Alexander von Humboldt foundation.

References

[1] S. ALBEVERIO, F. GESZTESY,R. HØEGH-KROHN, AND H. HOLDEN, Solvable Models in Quan- tum Mechanics, Texts and Monographs in Physics, Springer-Verlag, New York, 1988. [2] S. ALBEVERIO AND R. HØEGH-KROHN, Point interactions as limits of short range interactions, J. , 6 (1981), pp. 313–339. [3] P. R. S. ANTUNES,R.BENGURIA, V. LOTOREICHIK, AND T. OURMIERES` -BONAFOS, A varia- tional formulation for Dirac operators in bounded domains. Applications to spectral geometric inequal- ities, arXiv.org:2003.04061 (2020). [4] I.A.BAERNSTEIN, Symmetrization in analysis, vol. 36 of New Mathematical Monographs, Cam- bridge University Press, Cambridge, 2019. With David Drasin and Richard S. Laugesen, With a foreword by Walter Hayman. [5] C.BANDLE, Estimates for the Green’s functions of elliptic operators, SIAM J. Math. Anal., 9 (1978), pp. 1126–1136. [6] , Isoperimetric inequalities and applications, vol. 7 of Monographs and Studies in Mathe- matics, Pitman (Advanced Publishing Program), Boston, Mass.-London, 1980. [7] C.BANDLE AND M.FLUCHER, Harmonic radius and concentration of energy; hyperbolic radius and − Liouville’s ∆U = eU and ∆U = U (n+2)/(n 2), SIAM Rev., 38 (1996), pp. 191–238. [8] R.D.BENGURIA,S. R.FOURNAIS,E.STOCKMEYER, AND H. VAN DEN BOSCH, Spectral gaps of Dirac operators describing graphene quantum dots, Math. Phys. Anal. Geom., 20 (2017), pp. Paper No. 11, 12. [9] R.D.BENGURIA,H.LINDE, AND B.LOEWE, Isoperimetric inequalities for eigenvalues of the Lapla- cian and the Schr¨odinger operator, Bull. Math. Sci., 2 (2012), pp. 1–56. FABER-KRAHN INEQUALITIES WITH INTERACTIONS 25

[10] H. BETHE AND R. PEIERLS, Quantum Theory of the Diplon, Proceedings of the Royal Society of London. A, Mathematical and Physical Sciences, 148 (1935), pp. 146–156. [11] T. BHATTACHARYA, A proof of the Faber-Krahn inequality for the first eigenvalue of the p-Laplacian, Ann. Mat. Pura Appl. (4), 177 (1999), pp. 225–240. [12] M. S. BIRMAN AND M. Z. SOLOMJAK, Spectral theory of selfadjoint operators in Hilbert space, Mathematics and its Applications (Soviet Series), D. Reidel Publishing Co., Dordrecht, 1987. Translated from the 1980 Russian original by S. Khrushch¨ev and V. Peller. [13] P. BLANCHARD,R.FIGARI, AND A. MANTILE, Point interaction Hamiltonians in bounded do- mains, J. Math. Phys., 48 (2007), pp. 082108, 18. [14] H. J. BRASCAMP,E.H.LIEB, AND J.M.LUTTINGER, A general rearrangement inequality for multiple , J. Analysis, 17 (1974), pp. 227–237. [15] F. BROCK, An isoperimetric inequality for eigenvalues of the Stekloff problem, ZAMM Z. Angew. Math. Mech., 81 (2001), pp. 69–71. [16] F. BROCK AND A. Y. SOLYNIN, An approach to symmetrization via polarization, Trans. Amer. Math. Soc., 352 (2000), pp. 1759–1796. [17] D. BUCUR AND D. DANERS, An alternative approach to the Faber-Krahn inequality for Robin prob- lems, Calc. Var. Partial Differential Equations, 37 (2010), pp. 75–86. [18] D.BUCUR,V. FERONE,C.NITSCH, AND C. TROMBETTI, The quantitative Faber-Krahn inequality for the Robin Laplacian, J. Differential Equations, 264 (2018), pp. 4488–4503. [19] D. BUCUR, P. FREITAS, AND J. KENNEDY, The Robin problem, in Shape optimization and spec- tral theory, De Gruyter Open, Warsaw, 2017, pp. 78–119. [20] D. BUCUR AND A. GIACOMINI, Minimization of the k-th eigenvalue of the Robin-Laplacian, J. Funct. Anal., 277 (2019), pp. 643–687. [21] G. M. CAPRIANI, The Steiner rearrangement in any codimension, Calc. Var. Partial Differential Equations, 49 (2014), pp. 517–548. [22] S. CHANILLO, D. GRIESER, M. IMAI, K. KURATA, AND I. OHNISHI, Symmetry breaking and other phenomena in the optimization of eigenvalues for composite membranes, Comm. Math. Phys., 214 (2000), pp. 315–337. [23] I. CHAVEL, Eigenvalues in Riemannian geometry, vol. 115 of Pure and , Academic Press, Inc., Orlando, FL, 1984. Including a chapter by Burton Randol, With an ap- pendix by Jozef Dodziuk. [24] Y. COLINDE VERDIERE` , Pseudo-laplaciens. I., Ann. Inst. Fourier, 32 (1982), pp. 275-286. [25] G. CUPINI AND E. VECCHI, Faber-Krahn and Lieb-type inequalities for the composite membrane problem, Commun. Pure Appl. Anal., 18 (2019), pp. 2679–2691. [26] Q.-Y. DAI AND Y.-X.FU, Faber-Krahn inequality for Robin problems involving p-Laplacian, Acta Math. Appl. Sin. Engl. Ser., 27 (2011), pp. 13–28. [27] D. DANERS, A Faber-Krahn inequality for Robin problems in any space dimension, Math. Ann., 335 (2006), pp. 767–785. [28] D.DANERS AND J. KENNEDY, Uniqueness in the Faber-Krahn inequality for Robin problems, SIAM J. Math. Anal., 39 (2007/08), pp. 1191–1207. [29] P. EXNER AND V. LOTOREICHIK, Spectral optimization for Robin Laplacian in domains without cut loci, arXiv.org:2001.02718 (2020). [30] P. EXNER AND A. MANTILE, On the optimization of the principal eigenvalue for single-centre point- interaction operators in a bounded region, J. Phys. A, 41 (2008), pp. 065305, 15. [31] G. FABER, Beweis, dass unter allen homogenen Membranen von gleicher Fl¨ache und gleicher Spannung die kreisf¨ormige den tiefsten Grundton gibt, in Sitzungberichte der mathematisch- physikalischen Klasse der Bayerischen Akademie der Wissenschaften zu M ¨unchen Jahrgang, 1923, pp. 169–172. [32] S. FOURNAIS AND B. HELFFER, Inequalities for the lowest magnetic Neumann eigenvalue, Lett. Math. Phys., 109 (2019), pp. 1683–1700. 26 V. LOTOREICHIK AND A. MICHELANGELI

[33] P. FREITAS AND D. KREJCIˇ Rˇ´IK, The first Robin eigenvalue with negative boundary parameter, Adv. Math., 280 (2015), pp. 322–339. [34] M. GALLONE AND A. MICHELANGELI, Hydrogenoid spectra withcentral perturbations, Rep. Math. Phys., 84 (2019), pp. 215–243. [35] , Self-adjoint extensions with Friedrichs lower bound, arXiv.org:2003.09631 (2020). [36] M. GALLONE, A. MICHELANGELI, AND A. OTTOLINI, Kre˘ın-Viˇsik-Birman self-adjoint ex- tension theory revisited, in Mathematical Challenges of Zero Range Physics, A. Michelan- geli, ed., INdAM-Springer series, Springer International Publishing, 2020, SISSA preprint 25/2017/MATE. [37] P. GRISVARD, Elliptic problems in nonsmooth domains, vol. 69 of Classics in Applied Mathemat- ics, Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 2011. Reprint of the 1985 original [ MR0775683], With a foreword by Susanne C. Brenner. [38] E. M. HARRELL, P. KROGER¨ , AND K. KURATA, On the placement of an obstacle or a well so as to optimize the fundamental eigenvalue, SIAM J. Math. Anal., 33 (2001), pp. 240–259. [39] A. HENROT, Extremum problems for eigenvalues of elliptic operators, Frontiers in Mathematics, Birkh¨auser Verlag, Basel, 2006. [40] H. HU AND Q. DAI, Isoperimetric inequalities for positive solution of P-Laplacian, Math. Inequal. Appl., 17 (2014), pp. 1453–1469. [41] T. KATO, Fundamental properties of Hamiltonian operators of Schr¨odinger type, Trans. Amer. Math. Soc., 70 (1951), pp. 195–211. [42] , for linear operators, Classics in Mathematics, Springer-Verlag, Berlin, 1995. Reprint of the 1980 edition. [43] G. KEADY AND B.WIWATANAPATAPHEE, Inequalities for the fundamental Robin eigenvalue for the Laplacian on N-dimensional rectangular parallelepipeds, Math. Inequal. Appl., 21 (2018), pp. 911– 930. [44] J. KENNEDY, A Faber-Krahn inequality for the Laplacian with generalised Wentzell boundary condi- tions, J. Evol. Equ., 8 (2008), pp. 557–582. [45] J. B. KENNEDY, On the isoperimetric problem for the higher eigenvalues of the Robin and Wentzell Laplacians, Z. Angew. Math. Phys., 61 (2010), pp. 781–792. [46] S. KESAVAN, Symmetrization & applications, vol. 3 of Series in Analysis, World Scientific Pub- lishing Co. Pte. Ltd., Hackensack, NJ, 2006. [47] M. KHALILE AND V. LOTOREICHIK, Spectral isoperimetric inequalities for Robin Laplacians on 2- manifolds and unbounded cones, arXiv.org:1909.10842 (2019). [48] H. KOVARˇ´IK, On the lowest eigenvalue of Laplace operators with mixed boundary conditions, J. Geom. Anal., 24 (2014), pp. 1509–1525. [49] E. KRAHN, Uber¨ eine von Rayleigh formulierte Minimaleigenschaft des Kreises, Math. Ann., 94 (1925), pp. 97–100. [50] D. KREJCIˇ Rˇ´IK AND V. LOTOREICHIK, Optimisation of the lowest Robin eigenvalue in the exterior of a compact set, J. Convex Anal., 25 (2018), pp. 319–337. [51] , Optimisation of the lowest Robin eigenvalue in the exterior of a compact set II: non-convex domains and higher dimensions, Potential Analysis, (2020). [52] R. S. LAUGESEN, The Robin Laplacian—Spectral conjectures, rectangular theorems, J. Math. Phys., 60 (2019), pp. 121507, 31. [53] G. LEONI, A first course in Sobolev spaces, vol. 181 of Graduate Studies in Mathematics, Ameri- can Mathematical Society, Providence, RI, second ed., 2017. [54] P. LI AND S. T. YAU, Estimates of eigenvalues of a compact Riemannian , in Geometry of the Laplace operator (Proc. Sympos. Pure Math., Univ. Hawaii, Honolulu, Hawaii, 1979), Proc. Sympos. Pure Math., XXXVI, Amer. Math. Soc., Providence, R.I., 1980, pp. 205–239. [55] E. H. LIEB AND M.LOSS, Analysis, vol. 14 of Graduate Studies in Mathematics, American Mathematical Society, Providence, RI, second ed., 2001. FABER-KRAHN INEQUALITIES WITH INTERACTIONS 27

[56] A. MICHELANGELI AND A. OTTOLINI, On point interactions realised as Ter-Martirosyan- Skornyakov Hamiltonians, Rep. Math. Phys., 79 (2017), pp. 215–260. [57] A. POSILICANO, On the many Dirichlet Laplacians on a non-convex polygon and their approxima- tions by point interactions, J. Funct. Anal., 265 (2013), pp. 303–323. [58] J. W. S. RAYLEIGH, The Theory of Sound, Dover Publications, New York, N. Y., 1945. 2d ed (republication of the 1894/1896 edition). [59] M. REED AND B. SIMON, Methods of Modern , vol. 1, New York Academic Press, 1972. [60] A. SAVO, Optimal eigenvalue estimates for the Robin Laplacian on Riemannian manifolds, J. Differ- ential Equations, 268 (2020), pp. 2280–2308. [61] K. SCHMUDGEN¨ , Unbounded self-adjoint operators on Hilbert space, vol. 265 of Graduate Texts in Mathematics, Springer, Dordrecht, 2012. [62] G. SZEGO¨ , Inequalities for certain eigenvalues of a membrane of given area, J. Rational Mech. Anal., 3 (1954), pp. 343–356. [63] H. F. WEINBERGER, An isoperimetric inequality for the N-dimensional free membrane problem, J. Rational Mech. Anal., 5 (1956), pp. 633–636. [64] R. WEINSTOCK, Inequalities for a classical eigenvalue problem, J. Rational Mech. Anal., 3 (1954), pp. 745–753. [65] Y. XU, The first nonzero eigenvalue of Neumann problem on Riemannian manifolds, J. Geom. Anal., 5 (1995), pp. 151–165.

DEPARTMENT OF THEORETICAL PHYSICS, NUCLEAR PHYSICS INSTITUTE, CZECH ACADEMY OF SCIENCES, 25068 Rˇ EZˇ ,CZECH REPUBLIC, E-MAIL: [email protected]

INSTITUTE FOR APPLIED MATHEMATICS, AND HAUSDORFF CENTER FOR MATHEMATICS, UNIVER- SITY OF BONN,,ENDENICHER ALLEE 60 D-53115 BONN, GERMANY E-MAIL: [email protected] BONN.DE