We consider N bosons in two dimensions interacting via point interactions, modeled by the TMS Hamiltonian. We prove that their ground state energy has the asymptotic behavior −eCN for large N, for an explicit constant C related to the sharp constant in the Ladyzhenskaya-Gagliardo-Nirenberg inequality. Such asymptotic behavior was predicted in the physics literature, based on a variant of the mean-field approximation.
Introduction and Main Results
Introduction
Point interactions play a prominent role in quantum physics, due to their universal nature. They have been predicted to be relevant effective models for cold atomic gases [6, 13]. In this paper, we shall consider point-interacting bosons in the two-dimensional plane. These are described by the TMS Hamiltonian [12, 3], which will be recalled in the next section. It is well known that this Hamiltonian is bounded from below for any particle number N, but the best known lower bounds on its ground state energy behave as −eCN2 for a suitable constant C>0[3, 4]. On the other hand, a mean-field approximation predicts the asymptotic behavior −eCN[10, 6] for an explicit C involving the sharp constant a∗ in the Ladyzhenskaya–Gagliardo–Nirenberg inequality
∥∇u∥22∥u∥22≥2a∗∥u∥44,u∈H1(R2).(1)
For more on this inequality, including the numerical value a∗≈11.70, see [14]. In this paper, we shall prove that the prediction from [10, 6] is indeed correct.
Asymptotics of ground state energy
Let eN denote the ground state energy of the TMS Hamiltonian defined in (12) below. The latter depends on a parameter α∈R determining the two-body binding energy e2, but the ratio eN/e2 is independent of α, and only depends on N. It has the following asymptotic behavior.
Theorem 1. Let eN denote the ground state energy of the TMS Hamiltonian defined in (12), and a∗ the sharp constant in (1). Then
N→∞limN1lne2eN=a∗8π.(2)
The proof of this theorem will be given in § 3 (upper bound) and 4 (lower bound).
These bounds will be carried out in the charge formulation of the ground state energy problem, defined on functions ξ on R2(N−1) that are permutation symmetric in the last N−2 variables. For the upper bound, following [10], we take ξ to be a product state in these last N−2 variables, multiplied by an appropriately chosen function of the remaining variable. This choice of product states is reminiscent of the Hartree approximation in the mean field limit, which was studied, in a rather general setting, in [8]. In the charge formulation, however, the effective Hamiltonian does not consist only of one-body and two-body terms, and consequently the approach from [8] is not applicable. For the lower bound we instead proceed by an ad-hoc method, where at a crucial point a weak form of a result of [7] enters. Interestingly, the proof in [7] relies on the De Finetti theorem for product observables with logarithmic dependence on the dimension of the one-particle space [2, 9], which is a variant of the De Finetti theorem lying at the basis of [8].
Universality for weakly interacting bosons
The point interaction model arises naturally for systems with genuine short range interactions in the weak coupling limit. Consider N bosons in R2, with Hamiltonian
H=j=1∑N(−Δj)−λ1≤i<j≤N∑V(xi−xj)(3)
in Lsym2(R2N), where V∈L1(R2) with ∫V>0 is even and sufficiently regular (e.g. V∈L2(R2) and ∫R2∣V(x)∣(1+∣x∣)pdx<∞ for p>0 suffice). Let E0(N,λ) denote the ground state energy of H for λ>0. The following theorem shows universality in the weak coupling limit, and the appearance of point interactions as an effective model.
Theorem 2. With the definitions above,
λ→0limE0(2,λ)E0(N,λ)=ϵ2ϵN,(4)
independently of V.
As we shall explain momentarily, this theorem can be deduced in a straightforward way from the analysis in [5]. We remark that it follows from [11, 1] that
converges, as ε→0+, in norm resolvent sense to the TMS Hamiltonian with an explicit parameter α∈R; see [5 Eq. (4.34)]. While the value of α is irrelevant for us, what is crucial is that it is independent of N. Thus, the bottom of the spectrum of the operator (1.6) is given by eN+o(1) as ε→0. By scaling, it follows that the bottom of the spectrum of the operator
j=1∑N(−Δj)−∫V4πln(1/ε)11≤i<j≤N∑V(xi−xj)(7)
is given by ε2eN+o(ε2). Applying this fact with ε=exp(−4π/(λ∫V)) proves Theorem 2.
TMS Hamiltonian
The purpose of this section is two-fold. First, we introduce the TMS Hamiltonian, which describes N≥2 bosons of mass 1/2 in two dimensions, interacting via zero range forces, as a self-adjoint operator in the Hilbert space Lsym2(R2N) of permutation symmetric functions. Its ground state energy eN (that is, the bottom of the spectrum of this operator) is the main quantity of interest. Second, we introduce an ‘effective’ operator, acting in the Hilbert space
H:=L2(R2)⊗Lsym2(R2(N−2))(8)
of functions on R2(N−1) that are symmetric in the last N−2 variables. The elements of H are often called charges in the physics literature. The bottom of the spectrum of the effective operator in H, denoted by tN, is directly related to the quotient eN/e2. In the following sections we will study the asymptotics of tN.
Let us recall the definition of the TMS Hamiltonian, following [12, 3]. Let μ>0. In the Hilbert space H we introduce the quadratic form
Here, for p=(p1,…,pN)∈R2N we denote pij=(p1,…,pi,…,pj,…,pN)∈R2(N−2), and z∗ denotes the complex conjugate of a number z∈C. The domain of this quadratic form, domTμ, is given by those ξ∈H for which the first term on the right side is finite. The second and third terms are bounded quadratic forms in H, as we will see later on.
Now let α∈R and define a quadratic form Eα in Lsym2(R2N) as follows. We fix a parameter μ>0 (later, we will see that the construction is independent of this parameter) and define the domain of Eα to be
One easily checks that the definition of domEα and the value of Eα(ψ) are independent of μ. (To do so, one verifies that ξ is uniquely determined by ψ, independently of μ.)
As shown in [3], the quadratic form Eα is bounded from below and closed in the Hilbert space Lsym2(R2N) and, therefore, generates a self-adjoint operator. The bottom of its spectrum is
ϵN:=0=ψ∈domEαinf∥ψ∥2Eα(ψ).(13)
One easily sees that ϵN is the largest −μ such that Tμ(ξ)+α∥ξ∥2 is non-negative. In particular, for N=2, we see that ϵ2=−exp(−2α/π). By scaling, ϵN/ϵ2 is independent of α; in fact, we have the identity
In view of these facts, Theorem 1 is equivalent to the statement
N→∞limN1tN=−a∗(2π)2.(16)
which we shall prove in the following. It will be convenient to introduce the operators S0, S1 and S2 in H, corresponding to the three terms in (9) at μ=1, in such a way that
For computing its infimum, we shall use that it clearly suffices to consider functions ξ with non-negative Fourier transform, ξ≥0.
Upper Bound
Our upper bound on tN in (15) is based on a product ansatz, following [10]. We shall pick a normalized ξ∈H such that ξ^≥0. Then the second term in (17) is non-positive,
the optimal choice of v, given u, is v=u2/∥u∥42, and, in view of (15), this gives the desired upper bound
N→∞limsupN1tN≤−a∗(2π)2.(27)
Here we used the fact that the sharp constant a∗ in (1) remains the same when restricted to functions with nonnegative Fourier transform: indeed, replacing any function u by u=F−1∣u∣ preserves its L2 norm and Dirichlet integral, while ∣u∣4≥∣u∣4 by Plancherel’s theorem and the pointwise inequality ∣u∗u∣≤∣u∣∗∣u∣. Moreover, for an optimizer u, u2∈H1(R2).
Lower bound
The proof of the lower bound is substantially more involved. We shall split it into several steps.
Step 1. Reduction to S2
Step 1. Reduction to S2. We shall first show that it suffices to consider the third term S2 in (17). For the first term, we simply have S0≥0. For the second, we use the following lemma.
Proof. By writing (∣x∣2+∣y∣2+λ2)−1=∫0∞e−t(∣x∣2+∣y∣2+λ2)dt, we obtain the left inequality in (28). By Cauchy–Schwarz, we have for any positive function h,
For the last integral we have substituted t=1+ss to arrive at a beta function, which we have simplified using Euler’s reflection formula for the gamma function.
Shifting the integration variable p1 to p1−p2−p3, we can write
In the next-to-last step, we used Jensen’s inequality and, in the last step, the bosonic symmetry of ξ in the last N−2 variables. Optimizing over the choice of β thus yields ∥S1∥≤O(N−1lnN).
Remark 3. Using a product trial state similarly as in the last section, one can show that the N-dependence in the obtained bound ∥S1∥≤O(N−1lnN) is actually optimal.
Step 2. Energy localization
The operator K=eπ2S0 in H is given by the Fourier multiplier K(q,p3,…,pN)=21q2+∣p12∣2+1. Let us write
with 1R the characteristic function of ∣ln(K(p1+p2,p12)/K(p3+p4,p34))∣≤R for some R>0. The term involving 1R⊥=1−1R can be bounded with the aid of the following lemma.
Proof. Without loss of generality, we can restrict our attention to non-negative radial functions ψ. Introducing logarithmic coordinates ∣x∣2+λ2=eu, ∣y∣2+λ2=ev, and bounding the denominator by ∣x∣2+∣y∣2+λ2≥21(∣x∣2+∣y∣2+2λ2), the desired bound follows from
To bound the term involving 1R in (34), we pick a nonnegative χ∈C0∞((−1,1)) with ∫χ2=1 (independent of R). For L>0 and s∈R, define the Fourier multipliers
χs=L−1/2χ((lnK−s)/L)(38)
satisfying ∫Rχs2ds=1. For p in the support of 1R, we shall use that
For fixed P, Q, p5,…,pN, a simple Cauchy–Schwarz inequality (using the positivity of the operator on L2(R2) with integral kernel 1/(∣x∣2+∣y∣2+λ2); see Lemma 1) implies that
We integrate this inequality with respect to P and Q and apply another Cauchy–Schwarz inequality with the two square roots on the right side. After integrating with respect to p5,…,pN, we arrive at the bound
where on the right side we undid the change of variables (4.15). The resulting expression is thus given by
(2π)2∫0∞⟨ξ∣e−tK/2δ34e−tK/2∣ξ⟩dt,(47)
where δ34 denotes multiplication by δ(x3−x4). That is, we have shown that
⟨ξ∣S2∣ξ⟩≤(2π)2∫0∞⟨ξ∣e−tK/2δ34e−tK/2∣ξ⟩dt(48)
if ξ≥0.
Step 4. Mollifying the interaction.
We shall now replace the interaction δ(x3−x4) in (48) by a mollified interaction w(x3−x4). Assume that w is a smooth, nonnegative, compactly supported, even function on R2, with ∫R2w=1 and
∫R2w(x)eipxdx≥1−Cw∣p∣2
for all p∈R2. For ξ supported on p satisfying (41), all momenta p3,…,pN in the support of ξ satisfy ∣pj∣≤e(s+L)/2. In particular, ∣p−q∣≤2e(s+L)/2 in the integrand on the right hand side of (45). Consequently, with wκ(x)=κ2w(κx) for some κ>0
This bound holds for all R,L>0 such that R/L is sufficiently small and for all ε>0. For the first error term to be small, we choose R/L=ε1/2. For the last one we need e−R/2=ε/N. This leaves us with showing that
ε−1(N/ε)4ε−1/2EN−2(a)(57)
is small for large N, i.e., that EN−2(a) goes to zero faster than any power of N. This will be deduced in the final step from the work in [7]. Accepting this for the moment, we deduce that
∥S2∥≤a(N−2)2(2π)2(1+o(1))(58)
as N→∞.
Combining this with the operator inequality S0≥0 and the norm bound on S1 from Step 1, we deduce that
To get a lower bound on En(a), we can use IMS localization. Let Hn(a) denote the operator on the right-hand side of (51). It is convenient to view it as an operator on all of L2(R2n), it is well-known that the infimum of its spectrum is unchanged upon restriction to the bosonic subspace. IMS localization implies that for any ℓ>0 there exist zi∈R2 and a ψ∈L2(R2n) supported in ∏i=1nBℓ(zi) such that ⟨ψ∣Hn(a)∣ψ⟩≤En(a)+nCℓ−2, with Bℓ(z)⊂R2 denoting the ball of radius ℓ centered at z∈R2. Since w is compactly supported by assumption, an interaction term w(xi−xj) will be zero if ∣zi−zj∣>2ℓ+d, with d the diameter of the support of w. We thus slightly enlarge the balls Bℓ to Bℓ+d/2, and write their union as a disjoint union of m clusters, with 1≤m≤n, each consisting of finitely many balls: ⋃i=1nBℓ+d/2(zi)=⋃j=1mSj, the latter sets being mutually disjoint. Since there are no interactions between particles in different clusters, we have
⟨ψ∣Hn(a)∣ψ⟩=j=1∑m⟨ψ∣HSj∣ψ⟩(63)
where HSj is the Hamiltonian for the particles in cluster Sj.
Each cluster Sj is a connected union of 1≤mj≤n balls of radius ℓ+d/2, and is therefore contained in a ball BD(wj) for some wj∈R2, where D:=n(ℓ+d/2). We now translate all the coordinates in Sj by wj, i.e., center all the balls at the origin. This leaves the interactions within clusters invariant, but may introduce additional attractive interactions between particles in different clusters. The translated wave function ψ hence satisfies ⟨ψ∣Hn(a)∣ψ⟩≤⟨ψ∣Hn(a)∣ψ⟩. Since by construction ψ is supported in ⨂i=1nBD(0), we can add one-body potential terms D−2∑k=1nV(xk/D) to the Hamiltonian Hn(a) at the energy cost of at most nD−2supB1(0)V. Using (4.35) with κn=D, i.e., choosing ℓ=κn/n−d/2, we conclude that
En(a)≥−ℓ2Cn−D2n(K+B1(0)supV)≥−Cn3κn−2(64)
in the case κn≥dn. This completes the proof.
Acknowledgments. R.S. is grateful for inspiring discussions with Artem Volosniev. R.L.F. acknowledges partial support by the German Research Foundation grants EXC-2111-390814868 and TRR 352–Project-ID 470903074, as well as through the Chat-GPT for Academic Researchers program of OpenAI. R.S. acknowledges partial support by the Austrian Science Fund (FWF) grant DOI: 10.55776/I6427 (as part of the SFB/TRR 352).
AI declaration. The entire text of the manuscript was written by the authors without assistance of AI tools. Crucial ideas for the proof of the lower bound in § 4 were obtained with the help of GPT-6 Astra. Claude Opus 5.5 was used for proofreading.
Rupert L. Frank, Mathematisches Institut, Ludwig-Maximilians Universität München, Theresienstr. 39, 80333 München, Germany, and Munich Center for Quantum Science and Technology, Schellingstr. 4, 80799 München, Germany
[1]F. Bentosela, R.M. Cavalcanti, P. Exner, V.A. Zagrebnov, Anomalous electron trapping by localized magnetic fields, J. Phys. A 32, 3029–3039 (1999).DOI
[2]F.G. Brandão, A.W. Harrow, Quantum de Finetti theorems under local measurements with applications, Commun. Math. Phys. 353, 469–506 (2017).
[3]G. Dell-Antonio, R. Figari, A. Teta, Hamiltonians for systems of N particles interacting through point interactions, Ann. Inst. Henri Poincaré 60, 253–290 (1994).
[4]J. Dimock, S.G. Rajeev, Multi-particle Schrödinger operators with point interactions in the plane, J. Phys. A: Math. Gen. 37, 9157 (2004).DOI
[5]M. Griesemer, M. Hofacker, From Short-Range to Contact Interactions in Two-dimensional Many-Body Quantum Systems, Ann. Henri Poincaré 23, 2769–2818 (2022).
[6]H.-W. Hammer, D.T. Son, Universal Properties of Two-Dimensional Boson Droplets, Phys. Rev. Lett. 92, 250408 (2004).pubmed.ncbi.nlm.nih.gov/15697882
[7]L. Junge, F.L.A. Visconti, Derivation of Hartree theory for two-dimensional attractive Bose gases in almost Gross–Pitaevskii regime, J. Math. Phys. Appl. 205, 103800 (2026).arxiv.org/abs/2410.23150
[8]M. Lewin, P.T. Nam, N. Rougerie, Derivation of Hartree’s theory for generic mean-field Bose systems, Adv. Math. 254, 570–621 (2014).arxiv.org/abs/1303.0981
[9]K. Li, G. Smith, Quantum de Finetti theorem under fully-one-way adaptive measurements, Phys. Rev. Lett. 114, 160503 (2015).pubmed.ncbi.nlm.nih.gov/25955040
[10]S.G. Rajeev, A Condensation of Interacting Bosons in Two Dimensional Space, unpublished, preprint arXiv:9905120
[11]B. Simon, The Bound State of Weakly Coupled Schrödinger Operators in One and Two Dimensions, Ann. Phys. 97, 279–288 (1976).
[12]G.V. Skorniakov, K.A. Ter-Martirosian, Three body problem for short range forces. I. Scattering of low energy neutrons by deuterons, Sov. Phys. JETP 4, 648–661 (1957).
[13]M. Suchorowski, F. Brauneis, H.-W. Hammer, M. Tomza, A.G. Volosniev, Generalized Gross–Pitaevskii Equation for 2D Bosons with Attractive Interactions, Phys. Rev. Lett. 136, 243402 (2026).DOI