Introduction: origins, benchmarks, and the two correlation scales
Reduced density-matrix functionals aim to retain the exact one-body kinetic energy while replacing the two-body interaction by an explicit expression in the one-particle density matrix.
The work has been supported by the Deutsche Forschungsgemeinschaft (DFG TRR 352), grant 470903074.
The author acknowledges the use of OpenAI’s ChatGPT and Google’s Gemini for mathematical discussions, the development and checking of arguments, and assistance in preparing the manuscript. The author takes full responsibility for the final text and proofs. This places fractional natural-orbital occupations, rather than an idempotent Hartree–Fock projector, among the variational degrees of freedom. The foundational background includes Coleman’s characterization of ensemble-representable one-particle density matrices [7 ] and Gilbert’s extension of the density-functional framework to nonlocal external potentials [12 ] . The conditions 0 ≤ γ ≤ 1 0 \le\gamma\le1 0 ≤ γ ≤ 1 and tr γ = N \operatorname{tr}\gamma= N tr γ = N concern one-particle representability; they must not be confused with representability of an approximate two-particle density matrix.
From the square-root ansatz to corrected Hartree–Fock
In 1984 Müller [18 ] proposed an approximation to the two-particle density matrix that replaces the Hartree–Fock occupation factor n i n j n_i n_j n i n j in the exchange term by n i n j \sqrt{n_i n_j} n i n j . The square-root choice is tied to the contraction relation between reduced density matrices and to Pauli-type constraints. Buijse and Baerends [2 ] later obtained the same square-root expression from an exchange-correlation hole construction. These are related routes to the functional now usually called the Müller functional.
Csányi and Arias [8 ] developed a tensor-product approximation to the two-particle density matrix. Their “corrected Hartree” (CH) functional coincides with the Müller functional. Their distinct “corrected Hartree–Fock” (CHF) functional retains ordinary exchange and adds an exchange-type term involving γ ( 1 − γ ) \sqrt{\gamma(1-\gamma)} γ ( 1 − γ ) . For homogeneous spin-symmetric occupations the two interaction factors are, respectively,
n ( p ) n ( p ′ ) , n ( p ) n ( p ′ ) + n ( p ) ( 1 − n ( p ) ) n ( p ′ ) ( 1 − n ( p ′ ) ) . \sqrt{n(p)n(p')}, \qquad n(p)n(p')+\sqrt{n(p)(1-n(p))n(p')(1-n(p'))}. n ( p ) n ( p ′ ) , n ( p ) n ( p ′ ) + n ( p ) ( 1 − n ( p )) n ( p ′ ) ( 1 − n ( p ′ )) .
Throughout this paper, CHF denotes this original corrected Hartree–Fock functional, not the later Csányi–Goedecker–Arias functional [9 ] . Staroverov and Scuseria [22 ] related CHF optimization to Hartree–Fock–Bogoliubov methods, with a reversal of the pairing-energy sign. The associated gap equation is a useful variational structure; it is not, by itself, a statement about superconductivity of the repulsive many-electron Coulomb Hamiltonian.
What was known about these functionals
Cioslowski and Pernal [5 ] analyzed homogeneous momentum occupations and the regime in which the occupation cap is inactive. Csányi and Arias [8 ] numerically observed excess correlation for CH and near-zero correlation for CHF at high density, reporting a fitted logarithmic slope for CH in a study covering 0.01 < r s < 30 0.01 < r_s < 30 0.01 < r s < 30 . Lathiotakis, Helbig, and Gross [15 ] subsequently compared these and other density-matrix functionals on the homogeneous electron gas, confirming both occupation pinning for Müller and the rapid collapse of the CHF correlation energy as r s ↓ 0 r_s \downarrow0 r s ↓ 0 , while Cioslowski, Ziesche, and Pernal [6 ] derived high-density momentum constraints on simple natural-spin-orbital functionals. The earlier finite-density calculations reveal this qualitative contrast, but they do not determine the exact high-density scales and coefficients with controlled asymptotic errors. For Müller, a finite-range logarithmic fit does not identify the limiting coefficient. For CHF, the stretched-exponential gain poses a severe resolution problem, particularly when it is obtained by subtracting total energies. The calculations in [8 ] discuss discretization, momentum-cutoff, regularization, and minimization errors; [15 ] also discusses numerical false pinning of occupations. These issues are not limitations of every possible numerical method: higher precision, suitable rescaling, and a direct calculation of energy differences can improve numerical access. The present analysis determines the exact asymptotic scales and sharp limiting coefficients without relying on finite-range extrapolations.
The mathematical theory of Müller’s functional was developed by Frank, Lieb, Seiringer, and Siedentop [10 ] : among other results, they established convexity, uniqueness of the minimizing spatial density, and existence in the atomic and molecular binding regime N ≤ Z N \le Z N ≤ Z . These are results for the general density-matrix problem, not a claim of uniqueness of every minimizing density matrix. Siedentop [20 ] showed that the neutral-atom Müller energy agrees with the quantum energy through the Thomas–Fermi, Scott, and Dirac–Schwinger orders. More recently, Siedentop [21 ] proved the general comparison E M ≤ E C H F ≤ E H F \mathcal{E}_{\mathrm{M}} \le\mathcal{E}_{\mathrm{CHF}} \le\mathcal{E}_{\mathrm{HF}} E M ≤ E CHF ≤ E HF and reduced the same atomic accuracy for CHF. Agreement through these large atomic energy orders leaves smaller correlation terms unresolved. It does not imply agreement of homogeneous correlation energies.
Why the high-density comparison is discriminating
In the homogeneous setting, the common filled-Fermi-sea energy per particle is
e H F h o m ( k ) = 3 10 k 2 − 3 4 π k . e_{\mathrm{HF}}^{\mathrm{hom}}(k)=\frac{3}{10}k^2-\frac{3}{4\pi}k. e HF hom ( k ) = 10 3 k 2 − 4 π 3 k .
Both approximations recover these kinetic and exchange terms. Our results show, however, that their next energy scales are radically different. The Müller gain is 3 ( 4 π 2 ) − 1 log k + O ( 1 ) 3(4\pi^2)^{-1}\log k+O(1) 3 ( 4 π 2 ) − 1 log k + O ( 1 ) , whereas the CHF gain is
12 e − 1 / 2 − π 2 / 8 k 2 e − π 3 / 2 / k ( 1 + o ( 1 ) ) . 12e^{-1/2-\pi^2/8}k^2e^{-\pi^{3/2}/\sqrt{k}}(1+o(1)). 12 e − 1/2 − π 2 /8 k 2 e − π 3/2 / k ( 1 + o ( 1 )) .
In particular, the CHF correction is positive at every finite density but smaller than every inverse power of k k k at high density. This resolves the scale hidden behind a numerical observation of “essentially zero” correlation. In the Müller case, identifying the coefficient, rather than only logarithmic growth, tests the quantitative quality of its correlation mechanism.
For physical, unpolarized electrons, the high-density quantum benchmark is the Macke–Gell-Mann–Brueckner logarithm. Macke [17 ] and Pines [19 ] obtained the leading logarithmic behavior; Gell-Mann and Brueckner [11 ] developed the systematic resummation and the next constant term. In Hartree units the logarithmic coefficient is ( 1 − log 2 ) / π 2 (1-\log2)/\pi^2 ( 1 − log 2 ) / π 2 ; see also Loos and Gill [16 ] for the normalization and the role of the Hartree–Fock reference. Consequently Müller has the right logarithmic scale but a different coefficient, while CHF has a different scale altogether. Neither the functional ordering nor agreement of leading total energies can detect this distinction.
A complementary motivation is the conjecture that the Müller ground-state energy bounds the exact quantum ground-state energy from below for Coulomb systems [10 , 21 ] . From this perspective, the larger logarithmic correlation coefficient is not merely an inaccuracy of approximation: relative to the quantum benchmark it places the Müller energy on the conjectured side, with a strictly positive leading separation. § 10.2 makes this comparison quantitative. It supplies evidence for the conjecture, without proving it for general Coulomb systems.
The mathematical status of this comparison requires care. The results below are theorems about two specified homogeneous variational functionals. Christiansen, Hainzl, and Nam [3 , 4 ] proved a Coulomb Gell-Mann–Brueckner analogue in a mean-field regime. That result must not be identified with the unscaled thermodynamic jellium asymptotic. We keep this distinction explicit in § 10 ; the latter formula is used as the physical benchmark, not as an input to either functional proof.
A direct variational proof
The CHF upper bound is a comparison for every admissible occupation, not a consequence of an asymptotic solution of the gap equation. We choose an explicit comparison energy E ∗ E_* E ∗ whose Coulomb Birman–Schwinger operator has norm at most one. The norms restricted to either side of the Fermi ball tend to one half; this additional fact controls the ordinary exchange of the signed occupation defect. Completion of the square then leaves a positive weighted L 2 L^2 L 2 remainder. Thus the energy estimate also controls profiles of approximate minimizers. A trial state on the scale selected by the same deterministic spectrum supplies the matching lower bound.
There is also an important structural difference regarding symmetry restrictions. In Hartree–Fock theory, the energy improvement due to translation-symmetry breaking is exponentially small at high density [13 ] . Our CHF analysis remains in the homogeneous spin-symmetric class: we neither remove this restriction nor prove symmetry breaking for CHF. For the Müller functional, by contrast, convexity with respect to the one-particle density matrix [10 ] implies that averaging over translations and U ( q ) U(q) U ( q ) spin rotations cannot increase the periodic energy and eliminates the direct Coulomb term. Proposition 4.4 proves existence of an invariant global periodic minimizer and identifies the unrestricted thermodynamic energy with the homogeneous spin-symmetric value. This energy reduction does not assert that every minimizing density matrix is itself invariant.
After the common setting and Coulomb estimates, § 4 proves the Müller theorem and its periodic global optimality. The CHF uniqueness result in § 5 is independent of its high-density energy proof. § 6 –§ 8 give the deterministic comparison, its spectral input, and the matching trial calculation; § 9 deduces the defect profile. § 10 discusses the quantum benchmark and the lower-bound motivation. No pointwise high-density asymptotic for the self-consistent CHF gap, or atomic local-density transfer, is asserted.
Common setting and main results
We first formulate both variational problems in the translation-invariant, spin-symmetric class. For Müller, Proposition 4.4 shows that this entails no loss in the periodic thermodynamic energy. For CHF, the symmetry restriction remains part of the problem studied. All units are Hartree units. Fix a spin multiplicity q ∈ N q \in\mathbb{N} q ∈ N and a positive total electron density ρ \rho ρ . We use the notation d p ˉ = d p / ( 2 π ) \mathrm{d}\bar{p} = \mathrm{d}p/(2\pi) d p ˉ = d p / ( 2 π ) for the one-dimensional normalized measure, and correspondingly
d 3 p ˉ = d 3 p ( 2 π ) 3 , ( V f ) ( p ) = ∫ R 3 4 π f ( p ′ ) ∣ p − p ′ ∣ 2 d 3 p ˉ ′ , X ( f ) = q 2 ∫ f ( V f ) d 3 p ˉ . \mathrm{d}^{3}\bar{p} = \frac{\mathrm{d}^{3}p}{(2\pi)^{3}}, \qquad(Vf)(p) = \int_{\mathbb{R}^{3}} \frac{4\pi f(p')}{\lvert p-p'\rvert^{2}}\,\mathrm{d}^{3}\bar{p}', \qquad\mathcal{X}(f) = \frac{q}{2}\int f(Vf)\,\mathrm{d}^{3}\bar{p}. d 3 p ˉ = ( 2 π ) 3 d 3 p , ( V f ) ( p ) = ∫ R 3 ∣ p − p ′ ∣ 2 4 π f ( p ′ ) d 3 p ˉ ′ , X ( f ) = 2 q ∫ f ( V f ) d 3 p ˉ .
All other differentials are written with an upright d \mathrm{d} d . Unless indicated otherwise, L p L^{p} L p norms are taken on ( R 3 , d 3 p ˉ ) (\mathbb{R}^{3},\mathrm{d}^{3}\bar{p}) ( R 3 , d 3 p ˉ ) , whereas ∥ ⋅ ∥ \|\cdot\| ∥ ⋅ ∥ on linear operators denotes the operator norm. Fix once and for all 0 < θ < 1 / 12 0<\theta<1/12 0 < θ < 1/12 ; the shell parameter is independent of k k k , and no limit θ ↓ 0 \theta\downarrow0 θ ↓ 0 is needed. Constants C θ C_{\theta} C θ may depend on this fixed choice but not on k k k . The Coulomb quadratic form X \mathcal{X} X is nonnegative even for real sign-changing functions. Define
E H F ( n ) = q 2 ∫ R 3 ∣ p ∣ 2 n ( p ) d 3 p ˉ − X ( n ) , (1) \mathcal{E}_{\mathrm{HF}}(n)=\frac{q}{2}\int_{\mathbb{R}^{3}}\lvert p\rvert^{2}n(p)\,\mathrm{d}^{3}\bar{p}-\mathcal{X}(n), \tag*{(1)} E HF ( n ) = 2 q ∫ R 3 ∣ p ∣ 2 n ( p ) d 3 p ˉ − X ( n ) , (1)
E M ( n ) = q 2 ∫ R 3 ∣ p ∣ 2 n ( p ) d 3 p ˉ − X ( n ) , (2) \mathcal{E}_{\mathrm{M}}(n)=\frac{q}{2}\int_{\mathbb{R}^{3}}\lvert p\rvert^{2}n(p)\,\mathrm{d}^{3}\bar{p}-\mathcal{X}(\sqrt{n}), \tag*{(2)} E M ( n ) = 2 q ∫ R 3 ∣ p ∣ 2 n ( p ) d 3 p ˉ − X ( n ) , (2)
E ( n ) = E C H F ( n ) = q 2 ∫ R 3 ∣ p ∣ 2 n ( p ) d 3 p ˉ − X ( n ) − X ( w n ) , w n = n ( 1 − n ) . (3) \mathcal{E}(n)=\mathcal{E}_{\mathrm{CHF}}(n)=\frac{q}{2}\int_{\mathbb{R}^{3}}\lvert p\rvert^{2}n(p)\,\mathrm{d}^{3}\bar{p}-\mathcal{X}(n)-\mathcal{X}(w_n), \qquad w_n=\sqrt{n(1-n)}. \tag*{(3)} E ( n ) = E CHF ( n ) = 2 q ∫ R 3 ∣ p ∣ 2 n ( p ) d 3 p ˉ − X ( n ) − X ( w n ) , w n = n ( 1 − n ) . (3)
The abbreviation E \mathcal{E} E without a subscript always denotes CHF. The Müller energy in (2) is the unshifted functional; adding a density-linear self-energy correction would only change its per-particle remainder by a constant, not the logarithmic coefficient. The admissible set is
A ρ = { n : R 3 → [ 0 , 1 ] : q ∫ n d 3 p ˉ = ρ , ∫ ∣ p ∣ 2 n d 3 p ˉ < ∞ } . \mathcal{A}_{\rho}=\left\{n:\mathbb{R}^{3}\to[0,1]:q\int n\,\mathrm{d}^{3}\bar{p}=\rho,\quad\int\lvert p\rvert^{2}n\,\mathrm{d}^{3}\bar{p}<\infty\right\}. A ρ = { n : R 3 → [ 0 , 1 ] : q ∫ n d 3 p ˉ = ρ , ∫ ∣ p ∣ 2 n d 3 p ˉ < ∞ } .
For the general density-matrix comparison see [21 ] . In the present scalar class the pointwise ordering can be checked directly: for s , t ∈ [ 0 , 1 ] s,t\in[0,1] s , t ∈ [ 0 , 1 ] , Schwarz’s inequality gives
s t + s ( 1 − s ) t ( 1 − t ) = s t ( s t + ( 1 − s ) ( 1 − t ) ) ≤ s t . st+\sqrt{s(1-s)t(1-t)}=\sqrt{st}\left(\sqrt{st}+\sqrt{(1-s)(1-t)}\right)\leq\sqrt{st}. s t + s ( 1 − s ) t ( 1 − t ) = s t ( s t + ( 1 − s ) ( 1 − t ) ) ≤ s t .
Therefore, on the same admissible set,
E M ( n ) ≤ E C H F ( n ) ≤ E H F ( n ) . (4) \mathcal{E}_{\mathrm{M}}(n)\leq\mathcal{E}_{\mathrm{CHF}}(n)\leq\mathcal{E}_{\mathrm{HF}}(n). \tag*{(4)} E M ( n ) ≤ E CHF ( n ) ≤ E HF ( n ) . (4)
This is an ordering of approximate functionals, not an assertion about where the exact many-body quantum energy lies.
The homogeneous direct Coulomb term is canceled by the neutralizing background. This variational problem is not unrestricted Hartree–Fock; in particular we do not allow general spatially varying one-body density matrices or spin-symmetry breaking in the CHF analysis. For the Müller functional, however, convexity permits an invariant global minimizer for periodic jellium, and restricting to invariant states does not change the minimum (Proposition 4.4 ). Set
k = ( 6 π 2 ρ q ) 1 / 3 , P k ( p ) = 1 { ∣ p ∣ < k } , D ( k ) = E ( P k ) − inf n ∈ A ρ E ( n ) ρ . (5) k=\left(\frac{6\pi^{2}\rho}{q}\right)^{1/3}, \qquad P_k(p)=1_{\{\lvert p\rvert<k\}}, \qquad D(k)=\frac{\mathcal{E}(P_k)-\inf_{n\in\mathcal{A}_{\rho}}\mathcal{E}(n)}{\rho}. \tag*{(5)} k = ( q 6 π 2 ρ ) 1/3 , P k ( p ) = 1 {∣ p ∣ < k } , D ( k ) = ρ E ( P k ) − inf n ∈ A ρ E ( n ) . (5)
For ν ∈ { M , C H F } \nu\in\{\mathrm{M},\mathrm{CHF}\} ν ∈ { M , CHF } define the energy per particle and the nonnegative gain by
e ν h o m ( k ) = ρ − 1 inf n ∈ A ρ E ν ( n ) , D ν ( k ) = e H F h o m ( k ) − e ν h o m ( k ) . (6) e_{\nu}^{\mathrm{hom}}(k)=\rho^{-1}\inf_{n\in\mathcal{A}_{\rho}}\mathcal{E}_{\nu}(n),\qquad D_{\nu}(k)=e_{\mathrm{HF}}^{\mathrm{hom}}(k)-e_{\nu}^{\mathrm{hom}}(k). \tag*{(6)} e ν hom ( k ) = ρ − 1 n ∈ A ρ inf E ν ( n ) , D ν ( k ) = e HF hom ( k ) − e ν hom ( k ) . (6)
At fixed k k k , the constraint ∫ n d 3 p = k 3 / ( 6 π 2 ) \int n\,\mathrm{d}^{3}p=k^{3}/(6\pi^{2}) ∫ n d 3 p = k 3 / ( 6 π 2 ) is independent of q q q . Both the energy functionals and ρ = q k 3 / ( 6 π 2 ) \rho=qk^{3}/(6\pi^{2}) ρ = q k 3 / ( 6 π 2 ) carry the same factor q q q . Consequently, the entire energies per particle and their gains are independent of q q q when expressed as functions of k k k :
e ν h o m ( k ; q ) = e ν h o m ( k ; 1 ) , D ν ( k ; q ) = D ν ( k ; 1 ) , ν ∈ { M , C H F } . e_{\nu}^{\mathrm{hom}}(k;q)=e_{\nu}^{\mathrm{hom}}(k;1),\qquad D_{\nu}(k;q)=D_{\nu}(k;1),\qquad\nu\in\{\mathrm{M},\mathrm{CHF}\}. e ν hom ( k ; q ) = e ν hom ( k ; 1 ) , D ν ( k ; q ) = D ν ( k ; 1 ) , ν ∈ { M , CHF } .
The same observation applies to the Hartree–Fock reference energy; it does not concern a change of q q q at fixed total density. The reference energy is common to all three functionals:
ρ − 1 E M ( P k ) = ρ − 1 E C H F ( P k ) = ρ − 1 E H F ( P k ) = e H F h o m ( k ) = 3 k 2 10 − 3 k 4 π . (7) \rho^{-1}\mathcal{E}_{\mathrm{M}}(P_{k})=\rho^{-1}\mathcal{E}_{\mathrm{CHF}}(P_{k})=\rho^{-1}\mathcal{E}_{\mathrm{HF}}(P_{k})=e_{\mathrm{HF}}^{\mathrm{hom}}(k)=\frac{3k^{2}}{10}-\frac{3k}{4\pi}. \tag*{(7)} ρ − 1 E M ( P k ) = ρ − 1 E CHF ( P k ) = ρ − 1 E HF ( P k ) = e HF hom ( k ) = 10 3 k 2 − 4 π 3 k . (7)
For completeness, we spell out the exchange comparison. Let n ∗ n^{*} n ∗ be the symmetric decreasing rearrangement of n ∈ A ρ n\in\mathcal{A}_{\rho} n ∈ A ρ and set
Φ = V ( n ∗ + P k ) . \Phi=V(n^{*}+P_{k}). Φ = V ( n ∗ + P k ) .
The function Φ \Phi Φ is bounded and radially nonincreasing, since n ∗ + P k ∈ L 1 ∩ L ∞ n^{*}+P_{k}\in L^{1}\cap L^{\infty} n ∗ + P k ∈ L 1 ∩ L ∞ and both n ∗ + P k n^{*}+P_{k} n ∗ + P k and the Coulomb kernel are nonnegative and radially nonincreasing. Riesz rearrangement, symmetry of the kernel, and ∫ ( n ∗ − P k ) d 3 p = 0 \int(n^{*}-P_{k})\,\mathrm{d}^{3}p=0 ∫ ( n ∗ − P k ) d 3 p = 0 give
X ( n ) − X ( P k ) ≤ X ( n ∗ ) − X ( P k ) = − q 2 ∫ R 3 ( n ∗ ( p ) − P k ( p ) ) [ Φ ( ∣ p ∣ ) − Φ ( k ) ] d 3 p \begin{aligned}
\mathcal{X}(n)-\mathcal{X}(P_{k})\leq\mathcal{X}(n^{*})-\mathcal{X}(P_{k}) \\
&=-\frac{q}{2}\int_{\mathbb{R}^{3}}(n^{*}(p)-P_{k}(p))[\Phi(\lvert p\rvert)-\Phi(k)]\,\mathrm{d}^{3}p
\end{aligned} X ( n ) − X ( P k ) ≤ X ( n ∗ ) − X ( P k ) = − 2 q ∫ R 3 ( n ∗ ( p ) − P k ( p )) [ Φ (∣ p ∣) − Φ ( k )] d 3 p
≤ 0. (8) \leq0. \tag*{(8)} ≤ 0. (8)
Indeed, the two factors in the integrand have opposite signs both inside and outside the Fermi ball. All integrals are absolutely convergent because Φ \Phi Φ is bounded and n ∗ − P k ∈ L 1 n^{*}-P_{k}\in L^{1} n ∗ − P k ∈ L 1 . Together with the bathtub principle, (8) shows that P k P_{k} P k minimizes E H F \mathcal{E}_{\mathrm{HF}} E HF over the whole admissible class A ρ \mathcal{A}_{\rho} A ρ . In the CHF sections we retain the shorter notation D ( k ) = D C H F ( k ) D(k)=D_{\mathrm{CHF}}(k) D ( k ) = D CHF ( k ) . By (4) , 0 ≤ D ( k ) ≤ D M ( k ) 0\leq D(k)\leq D_{\mathrm{M}}(k) 0 ≤ D ( k ) ≤ D M ( k ) .
Theorem 2.1 (The Müller logarithm). For every fixed spin multiplicity q ∈ N q\in\mathbb{N} q ∈ N , the Müller energy gain in (6) satisfies
D M ( k ) = 3 4 π 2 log k + O ( 1 ) ( k → ∞ ) . (9) D_{\mathrm{M}}(k)=\frac{3}{4\pi^{2}}\log k+O(1)\qquad(k\to\infty). \tag*{(9)} D M ( k ) = 4 π 2 3 log k + O ( 1 ) ( k → ∞ ) . (9) The bound on the remainder is independent of q q q . The statement controls a bounded remainder but does not assert that this remainder has a limit.
For the CHF asymptotics write
A = π 3 / 2 2 , y 0 = 4 e − 1 / 4 − π 2 / 16 , C ∗ = 3 4 y 0 2 = 12 e − 1 / 2 − π 2 / 8 . (10) A=\frac{\pi^{3/2}}{2},\qquad y_{0}=4e^{-1/4-\pi^{2}/16},\qquad C_{*}=\frac{3}{4}y_{0}^{2}=12e^{-1/2-\pi^{2}/8}. \tag*{(10)} A = 2 π 3/2 , y 0 = 4 e − 1/4 − π 2 /16 , C ∗ = 4 3 y 0 2 = 12 e − 1/2 − π 2 /8 . (10)
Theorem 2.2 (Sharp CHF high-density asymptotics). For every fixed spin multiplicity q ∈ N q\in\mathbb{N} q ∈ N , the energy gain D ( k ) D(k) D ( k ) defined in (5) satisfies
lim k → ∞ D ( k ) k 2 e − π 3 / 2 k = 12 exp ( − 1 2 − π 2 8 ) . (11) \lim_{k\to\infty}\frac{D(k)}{k^{2}e^{-\pi^{3/2}\sqrt{k}}}=12\exp\left(-\frac{1}{2}-\frac{\pi^{2}}{8}\right). \tag*{(11)} k → ∞ lim k 2 e − π 3/2 k D ( k ) = 12 exp ( − 2 1 − 8 π 2 ) . (11) The reference energy is that of the filled Fermi sea in the translation-invariant, spin-symmetric class. The limiting constant is independent of q q q and equals 2.1195548699 … 2.1195548699\ldots 2.1195548699 … .
The comparison argument also gives the weighted defect-profile convergence in Corollary 9.1 , including an assertion for approximate minimizers. Its scale is deterministic and is not identified with the self-consistent gap.
Coulomb estimates, existence, and radial coordinates
Coulomb estimates
We shall use
0 ≤ X ( f ) ≤ C q ∥ f ∥ 2 ∥ ∣ p ∣ f ∥ 2 (12) 0 \le\mathcal{X}(f) \le Cq\|f\|_2\||p|f\|_2 \tag*{(12)} 0 ≤ X ( f ) ≤ C q ∥ f ∥ 2 ∥∣ p ∣ f ∥ 2 (12)
∥ V f ∥ ∞ ≤ C ∥ f ∥ ∞ 1 / 3 ∥ f ∥ 2 2 / 3 (13) \|Vf\|_\infty\le C\|f\|_\infty^{1/3}\|f\|_2^{2/3} \tag*{(13)} ∥ V f ∥ ∞ ≤ C ∥ f ∥ ∞ 1/3 ∥ f ∥ 2 2/3 (13)
∫ R 3 ∣ V f ( p ) ∣ 2 ∣ p ∣ 2 d 3 p ≤ 4 ∥ f ∥ 2 2 (14) \int_{\mathbb{R}^3}\frac{|Vf(p)|^2}{|p|^2}\,\mathrm{d}^3p \le4\|f\|_2^2 \tag*{(14)} ∫ R 3 ∣ p ∣ 2 ∣ V f ( p ) ∣ 2 d 3 p ≤ 4∥ f ∥ 2 2 (14)
The first estimate follows from Schwarz and Hardy after Fourier transformation, where V V V is multiplication by ∣ x ∣ − 1 |x|^{-1} ∣ x ∣ − 1 . For the second, split the convolution at radius s > 0 s>0 s > 0 : ∥ V f ∥ ∞ ≤ C ( s ∥ f ∥ ∞ + s − 1 / 2 ∥ f ∥ 2 ) \|Vf\|_\infty\le C(s\|f\|_\infty+s^{-1/2}\|f\|_2) ∥ V f ∥ ∞ ≤ C ( s ∥ f ∥ ∞ + s − 1/2 ∥ f ∥ 2 ) , and optimize s s s . The same bound applies to signed functions by using ∣ f ∣ |f| ∣ f ∣ . For the third, ∥ ∇ p V f ∥ 2 = ∥ f ∥ 2 \|\nabla_pVf\|_2=\|f\|_2 ∥ ∇ p V f ∥ 2 = ∥ f ∥ 2 , and Hardy in the p p p variable applies. This is a homogeneous Sobolev assertion; it does not require V f ∈ L 2 Vf\in L^2 V f ∈ L 2 . Approximation justifies these estimates in the stated norm classes. The convolution of a nonnegative L 2 ∩ L ∞ L^2\cap L^\infty L 2 ∩ L ∞ function is bounded and continuous, and strictly positive if the function is not identically zero.
Lemma 3.1 (Continuity of the Coulomb form). Suppose f j ⇀ f f_j\rightharpoonup f f j ⇀ f weakly in L 2 ( R 3 , d 3 p ) L^2(\mathbb{R}^3,\mathrm{d}^3p) L 2 ( R 3 , d 3 p ) and sup j ( ∥ f j ∥ 2 + ∥ ∣ p ∣ f j ∥ 2 ) < ∞ \sup_j(\|f_j\|_2+\||p|f_j\|_2)<\infty sup j ( ∥ f j ∥ 2 + ∥∣ p ∣ f j ∥ 2 ) < ∞ . Then X ( f j ) → X ( f ) \mathcal{X}(f_j)\to\mathcal{X}(f) X ( f j ) → X ( f ) .
Proof. For f j o u t = 1 { ∣ p ∣ > s } f j f_j^{\mathrm{out}}=\mathbf{1}_{\{|p|>s\}}f_j f j out = 1 { ∣ p ∣ > s } f j , ∥ f j o u t ∥ 2 = O ( s − 1 ) \|f_j^{\mathrm{out}}\|_2=O(s^{-1}) ∥ f j out ∥ 2 = O ( s − 1 ) and ∥ ∣ p ∣ f j o u t ∥ 2 = O ( 1 ) \||p|f_j^{\mathrm{out}}\|_2=O(1) ∥∣ p ∣ f j out ∥ 2 = O ( 1 ) . Thus (3.1) bounds its Coulomb form by C / s C/s C / s . Schwarz for the positive Coulomb form controls the mixed term uniformly. On a fixed ball the Coulomb operator is compact: cutting out ∣ p − p ′ ∣ < δ |p-p'|<\delta ∣ p − p ′ ∣ < δ leaves a Hilbert–Schmidt kernel, and Schur’s test bounds the removed operator by C δ C\delta C δ . Weak convergence proves convergence of the interior quadratic forms. Now let s → ∞ s\to\infty s → ∞ . The weak limit satisfies the same weighted bounds.
Existence for both functionals
Theorem 3.2 (Existence). For every q ∈ N q\in\mathbb{N} q ∈ N and ρ > 0 \rho>0 ρ > 0 , both E M \mathcal{E}_{\mathrm{M}} E M and E C H F \mathcal{E}_{\mathrm{CHF}} E CHF attain their infima on A ρ \mathcal{A}_\rho A ρ .
Proof. Put N = ρ / q N=\rho/q N = ρ / q and T ( n ) = ∫ ∣ p ∣ 2 n d 3 p T(n)=\int|p|^2n\,\mathrm{d}^3p T ( n ) = ∫ ∣ p ∣ 2 n d 3 p . For either functional, n 2 ≤ n n^2\le n n 2 ≤ n and w n 2 ≤ n w_n^2\le n w n 2 ≤ n give
E ν ( n ) ≥ q 2 T ( n ) − C q N T ( n ) , ν ∈ { M , C H F } . (15) \mathcal{E}_\nu(n)\ge\frac{q}{2}T(n)-Cq\sqrt{NT(n)},\qquad\nu\in\{\mathrm{M},\mathrm{CHF}\}. \tag*{(15)} E ν ( n ) ≥ 2 q T ( n ) − C q N T ( n ) , ν ∈ { M , CHF } . (15)
For a minimizing sequence n j n_j n j the kinetic moments are bounded; in particular,
∫ ∣ p ∣ > s n j d 3 p ≤ C / s 2 . (16) \int_{|p|>s}n_j\,\mathrm{d}^3p\le C/s^2. \tag*{(16)} ∫ ∣ p ∣ > s n j d 3 p ≤ C / s 2 . (16)
Set z j = n j z_j=\sqrt{n_j} z j = n j for Müller and z j = n j ( 1 − n j ) z_j=\sqrt{n_j(1-n_j)} z j = n j ( 1 − n j ) for CHF. Extract weak L 2 L^2 L 2 limits n j ⇀ n n_j\rightharpoonup n n j ⇀ n , z j ⇀ z z_j\rightharpoonup z z j ⇀ z . The hypographs of the two concave functions φ ( s ) = s \varphi(s)=\sqrt{s} φ ( s ) = s and φ ( s ) = s ( 1 − s ) \varphi(s)=\sqrt{s(1-s)} φ ( s ) = s ( 1 − s ) on [ 0 , 1 ] [0,1] [ 0 , 1 ] are closed and convex. Weak closedness therefore implies
0 ≤ n ≤ 1 , 0 ≤ z ≤ φ ( n ) a.e. (17) 0\le n\le1,\qquad0\le z\le\varphi(n)\quad\text{a.e.} \tag*{(17)} 0 ≤ n ≤ 1 , 0 ≤ z ≤ φ ( n ) a.e. (17)
Local weak convergence and (3.5) preserve ∫ n = N \int n=N ∫ n = N , and T ( n ) ≤ lim inf T ( n j ) T(n)\le\liminf T(n_j) T ( n ) ≤ lim inf T ( n j ) . Lemma 3.1 applies to z j z_j z j and, for CHF, to n j n_j n j ; all their squared weighted moments are bounded by T ( n j ) T(n_j) T ( n j ) . Replacing z z z by φ ( n ) \varphi(n) φ ( n ) increases its positive-kernel Coulomb form. This proves lower semicontinuity in the required direction for either functional and hence existence.
Explicit radial formulas
For radial real functions,
X ( f ) = q 4 π 3 ∫ 0 ∞ ∫ 0 ∞ r s f ( r ) f ( s ) log ∣ r + s r − s ∣ d r d s . (18) \mathcal{X}(f)=\frac{q}{4\pi^3}\int_0^\infty\int_0^\infty rsf(r)f(s)\log\left|\frac{r+s}{r-s}\right|\,\mathrm{d}r\,\mathrm{d}s. \tag*{(18)} X ( f ) = 4 π 3 q ∫ 0 ∞ ∫ 0 ∞ r s f ( r ) f ( s ) log r − s r + s d r d s . (18)
The potential of the Fermi ball is
( V P k ) ( r ) = 1 π [ k + k 2 − r 2 2 r log ∣ k + r k − r ∣ ] , ( V P k ) ( k ) = k π . (19) (VP_k)(r)=\frac{1}{\pi}\left[k+\frac{k^2-r^2}{2r}\log\left|\frac{k+r}{k-r}\right|\right],\qquad(VP_k)(k)=\frac{k}{\pi}. \tag*{(19)} ( V P k ) ( r ) = π 1 [ k + 2 r k 2 − r 2 log k − r k + r ] , ( V P k ) ( k ) = π k . (19)
The values at r = 0 , k r=0,k r = 0 , k are taken by continuity. To preserve radial particle number exactly, use the volume coordinate
t = r 3 − k 3 3 k 2 , r ( t ) = ( k 3 + 3 k 2 t ) 1 / 3 , d 3 p ( radially ) = k 2 2 π 2 d t , t ≥ − k / 3. (20) t=\frac{r^3-k^3}{3k^2},\qquad r(t)=\left(k^3+3k^2t\right)^{1/3},\qquad\mathrm{d}^3p\ (\text{radially})=\frac{k^2}{2\pi^2}\,\mathrm{d}t,\qquad t\ge-k/3. \tag*{(20)} t = 3 k 2 r 3 − k 3 , r ( t ) = ( k 3 + 3 k 2 t ) 1/3 , d 3 p ( radially ) = 2 π 2 k 2 d t , t ≥ − k /3. (20)
We identify a radial momentum-space function with its radial profile; a tilde denotes its expression in the volume coordinate. In particular, h k ( p ) = n k ( p ) − P k ( p ) h_k(p)=n_k(p)-P_k(p) h k ( p ) = n k ( p ) − P k ( p ) and h ~ k ( t ) = h k ( r ( t ) ) \widetilde{h}_k(t)=h_k(r(t)) h k ( t ) = h k ( r ( t )) . For nonradial competitors h k h_k h k always remains a function of p ∈ R 3 p\in\mathbb{R}^3 p ∈ R 3 .
Lemma 3.3 (Fixed-shell Coulomb bounds). Put R = θ k R=\theta k R = θ k . For ∣ t ∣ , ∣ s ∣ ≤ R |t|,|s|\le R ∣ t ∣ , ∣ s ∣ ≤ R , the kernel
H k ( t , s ) = k 2 r ( t ) r ( s ) log r ( t ) + r ( s ) ∣ r ( t ) − r ( s ) ∣ H_k(t,s)=\frac{k^2}{r(t)r(s)}\log\frac{r(t)+r(s)}{|r(t)-r(s)|} H k ( t , s ) = r ( t ) r ( s ) k 2 log ∣ r ( t ) − r ( s ) ∣ r ( t ) + r ( s ) satisfies
0 ≤ H k ( t , s ) ≤ C θ ( 1 + log + C θ k ∣ t − s ∣ ) (21) 0\le H_k(t,s)\le C_\theta\left(1+\log_+\frac{C_\theta k}{|t-s|}\right) \tag*{(21)} 0 ≤ H k ( t , s ) ≤ C θ ( 1 + log + ∣ t − s ∣ C θ k ) (21) ∣ H k ( t , s ) − log 2 k ∣ t − s ∣ ∣ ≤ C θ ∣ t ∣ + ∣ s ∣ k ( 1 + log 2 k ∣ t − s ∣ ) . (22) \left|H_k(t,s)-\log\frac{2k}{|t-s|}\right|\le C_\theta\frac{|t|+|s|}{k}\left(1+\log\frac{2k}{|t-s|}\right). \tag*{(22)} H k ( t , s ) − log ∣ t − s ∣ 2 k ≤ C θ k ∣ t ∣ + ∣ s ∣ ( 1 + log ∣ t − s ∣ 2 k ) . (22) If 0 ≤ f ≤ 1 0\le f\le1 0 ≤ f ≤ 1 is supported in [ − R , R ] [-R,R] [ − R , R ] and m = ∫ f d t > 0 m=\int f\,\mathrm{d}t>0 m = ∫ f d t > 0 , then
sup t ∫ f ( s ) ( 1 + log + C θ k ∣ t − s ∣ ) d s ≤ C θ m ( 1 + log C θ k m ) . (23) \sup_t\int f(s)\left(1+\log_+\frac{C_\theta k}{|t-s|}\right)\,\mathrm{d}s\le C_\theta m\left(1+\log\frac{C_\theta k}{m}\right). \tag*{(23)} t sup ∫ f ( s ) ( 1 + log + ∣ t − s ∣ C θ k ) d s ≤ C θ m ( 1 + log m C θ k ) . (23)
Proof. Use r ( t ) 3 − r ( s ) 3 = 3 k 2 ( t − s ) r(t)^3-r(s)^3=3k^2(t-s) r ( t ) 3 − r ( s ) 3 = 3 k 2 ( t − s ) and r ( t ) / k = 1 + O θ ( ∣ t ∣ / k ) r(t)/k=1+O_\theta(|t|/k) r ( t ) / k = 1 + O θ ( ∣ t ∣/ k ) . In particular,
r ( t ) + r ( s ) ∣ r ( t ) − r ( s ) ∣ = ( r ( t ) + r ( s ) ) ( r ( t ) 2 + r ( t ) r ( s ) + r ( s ) 2 ) 3 k 2 ∣ t − s ∣ = 2 k ∣ t − s ∣ ( 1 + O θ ( ∣ t ∣ + ∣ s ∣ k ) ) . \frac{r(t)+r(s)}{|r(t)-r(s)|}
=\frac{(r(t)+r(s))(r(t)^2+r(t)r(s)+r(s)^2)}{3k^2|t-s|}
=\frac{2k}{|t-s|}\left(1+O_\theta\left(\frac{|t|+|s|}{k}\right)\right). ∣ r ( t ) − r ( s ) ∣ r ( t ) + r ( s ) = 3 k 2 ∣ t − s ∣ ( r ( t ) + r ( s )) ( r ( t ) 2 + r ( t ) r ( s ) + r ( s ) 2 ) = ∣ t − s ∣ 2 k ( 1 + O θ ( k ∣ t ∣ + ∣ s ∣ ) ) .
This proves the two kernel estimates. For (3.12) split at ∣ t − s ∣ = m |t-s|=m ∣ t − s ∣ = m : use f ≤ 1 f\le1 f ≤ 1 inside and total mass m m m outside. The local logarithm is integrable, uniformly in the observation point. □
The Müller logarithm with the occupation cap
This section proves Theorem 2.1 in the common normalization of § 2 , and then establishes global optimality in periodic Müller jellium. Write b = n b=\sqrt{n} b = n and
F M ( b ) : = E M ( b 2 ) = q 2 ∫ ∣ p ∣ 2 b ( p ) 2 d 3 p − X ( b ) , 0 ≤ b ≤ 1 , q ∥ b ∥ 2 2 = ρ . \mathcal{F}_{\mathrm{M}}(b):=\mathcal{E}_{\mathrm{M}}(b^2)=\frac{q}{2}\int|p|^2b(p)^2\,\mathrm{d}^3p-\mathcal{X}(b),\qquad0\le b\le1,\qquad q\|b\|_2^2=\rho. F M ( b ) := E M ( b 2 ) = 2 q ∫ ∣ p ∣ 2 b ( p ) 2 d 3 p − X ( b ) , 0 ≤ b ≤ 1 , q ∥ b ∥ 2 2 = ρ .
The field is σ b = V b \sigma_b=Vb σ b = V b . The letter b b b in this section denotes the square root of the occupation.
Existence and the constrained Euler equation.
Lemma 4.1. *The Müller variational problem has a minimizer. One may choose b M , k b_{\mathrm{M},k} b M , k radial and nonincreasing, and b M , k > 0 b_{\mathrm{M},k}>0 b M , k > 0 almost everywhere. There is a real multiplier μ M \mu_{\mathrm{M}} μ M such that, with g ( p ) = ∣ p ∣ 2 / 2 − μ M g(p)=|p|^2/2-\mu_{\mathrm{M}} g ( p ) = ∣ p ∣ 2 /2 − μ M ,
2 g b M , k = σ b M , k on { 0 < b M , k < 1 } , 2 g ≤ σ b M , k on { b M , k = 1 } . (24) 2g b_{\mathrm{M},k}=\sigma_{b_{\mathrm{M},k}}\quad\text{on }\{0<b_{\mathrm{M},k}<1\},\qquad2g\le\sigma_{b_{\mathrm{M},k}}\quad\text{on }\{b_{\mathrm{M},k}=1\}. \tag*{(24)} 2 g b M , k = σ b M , k on { 0 < b M , k < 1 } , 2 g ≤ σ b M , k on { b M , k = 1 } . (24) In particular b M , k = 1 b_{\mathrm{M},k}=1 b M , k = 1 where g ≤ 0 g\le0 g ≤ 0 ; where g > 0 g>0 g > 0 ,
b M , k = min { 1 , σ b M , k 2 g } . (25) b_{\mathrm{M},k}=\min\left\{1,\frac{\sigma_{b_{\mathrm{M},k}}}{2g}\right\}. \tag*{(25)} b M , k = min { 1 , 2 g σ b M , k } . (25)
Proof. Existence follows from Theorem 3.2 . Symmetric decreasing rearrangement of b b b preserves its cap and norm, decreases the kinetic term and increases X ( b ) \mathcal{X}(b) X ( b ) . A radial nonincreasing minimizer can consequently be selected.
The potential σ b \sigma_b σ b is positive and bounded for every nonzero admissible b b b . If n = b 2 n=b^2 n = b 2 vanished on a set of positive measure, add ε \varepsilon ε to n n n on a bounded subset and remove the same mass from a bounded set on which n n n is bounded below. The latter change has energy cost O ( ε ) O(\varepsilon) O ( ε ) , whereas the new square-root occupation produces a strictly negative energy change of order ε \sqrt{\varepsilon} ε . This contradicts minimality. The same argument rules out an indicator minimizer. Local mass-preserving variations on { 0 < n < 1 } \{0<n<1\} { 0 < n < 1 } give a common multiplier; one-sided variations on { n = 1 } \{n=1\} { n = 1 } give the indicated inequality. Exhaust the fractional set by { ∣ p ∣ ≤ j , j − 1 ≤ n ≤ 1 − j − 1 } \{|p|\le j,\ j^{-1}\le n\le1-j^{-1}\} { ∣ p ∣ ≤ j , j − 1 ≤ n ≤ 1 − j − 1 } ; overlapping positive-measure subsets give the same multiplier. This proves (4.1) and (4.2). □ \square □
The pinned core and localization
In the rest of this section b = b M , k b=b_{M,k} b = b M , k denotes the selected minimizer, and constants are independent of large k k k . Its cap and mass give
0 < σ b ( p ) ≤ C k . (26) 0<\sigma_b(p)\le Ck. \tag*{(26)} 0 < σ b ( p ) ≤ C k . (26)
For sufficiently large k k k , we claim that
0 ≤ δ k M : = k 2 2 − μ M ≤ C k . (27) 0\le\delta_k^M:=\frac{k^2}{2}-\mu_M\le Ck. \tag*{(27)} 0 ≤ δ k M := 2 k 2 − μ M ≤ C k . (27)
If μ M ≤ 0 \mu_M\le0 μ M ≤ 0 , (4.2) would give b ( p ) 2 ≤ min { 1 , C k 2 / ∣ p ∣ 4 } b(p)^2\le\min\{1,Ck^2/|p|^4\} b ( p ) 2 ≤ min { 1 , C k 2 /∣ p ∣ 4 } and therefore ∥ b ∥ 2 2 ≤ C k 3 / 2 \|b\|_2^2\le Ck^{3/2} ∥ b ∥ 2 2 ≤ C k 3/2 , contradicting ∥ b ∥ 2 2 = k 3 / ( 6 π 2 ) \|b\|_2^2=k^3/(6\pi^2) ∥ b ∥ 2 2 = k 3 / ( 6 π 2 ) for large k k k . Thus κ = 2 μ M > 0 \kappa=\sqrt{2\mu_M}>0 κ = 2 μ M > 0 . Since b = 1 b=1 b = 1 on ∣ p ∣ < κ |p|<\kappa ∣ p ∣ < κ , mass conservation gives κ ≤ k \kappa\le k κ ≤ k . Outside this ball,
b ( r ) 2 ≤ min { 1 , C k 2 / ( r 2 − κ 2 ) 2 } . b(r)^2\le\min\{1,Ck^2/(r^2-\kappa^2)^2\}. b ( r ) 2 ≤ min { 1 , C k 2 / ( r 2 − κ 2 ) 2 } .
With v = r 2 − κ 2 v=r^2-\kappa^2 v = r 2 − κ 2 and r 2 d r = 1 2 κ 2 + v d v r^2\,\mathrm{d}r=\frac{1}{2}\sqrt{\kappa^2+v}\,\mathrm{d}v r 2 d r = 2 1 κ 2 + v d v , splitting at v = k v=k v = k proves
∫ ∣ p ∣ > κ b ( p ) 2 d 3 p ≤ C ∫ 0 ∞ ( κ + v ) min { 1 , k 2 / v 2 } d v ≤ C ( k κ + k 3 / 2 ) ≤ C k 2 . \int_{|p|>\kappa}b(p)^2\,\mathrm{d}^3p\le C\int_0^\infty(\kappa+\sqrt{v})\min\{1,k^2/v^2\}\,\mathrm{d}v\le C(k\kappa+k^{3/2})\le Ck^2. ∫ ∣ p ∣ > κ b ( p ) 2 d 3 p ≤ C ∫ 0 ∞ ( κ + v ) min { 1 , k 2 / v 2 } d v ≤ C ( k κ + k 3/2 ) ≤ C k 2 .
Thus k 3 − κ 3 = O ( k 2 ) k^3-\kappa^3=O(k^2) k 3 − κ 3 = O ( k 2 ) , which implies k − κ = O ( 1 ) k-\kappa=O(1) k − κ = O ( 1 ) and proves (4.4).
Let P = P k P=P_k P = P k and H = b − P H=b-P H = b − P . Equal masses imply
∫ ∣ b 2 − P ∣ d 3 p = O ( k 2 ) , ∥ H ∥ 2 2 ≤ ∫ ∣ b 2 − P ∣ d 3 p = O ( k 2 ) . (28) \int|b^2-P|\,\mathrm{d}^3p=O(k^2),\qquad\|H\|_2^2\le\int|b^2-P|\,\mathrm{d}^3p=O(k^2). \tag*{(28)} ∫ ∣ b 2 − P ∣ d 3 p = O ( k 2 ) , ∥ H ∥ 2 2 ≤ ∫ ∣ b 2 − P ∣ d 3 p = O ( k 2 ) . (28)
Consequently (3.2) gives
∥ σ b − σ P ∥ ∞ = O ( k 2 / 3 ) . (29) \|\sigma_b-\sigma_P\|_\infty=O(k^{2/3}). \tag*{(29)} ∥ σ b − σ P ∥ ∞ = O ( k 2/3 ) . (29)
This is an absolute error, uniform in momentum. It is sufficient because its contribution to the final energy is O ( k 3 ) O(k^3) O ( k 3 ) . Moreover, on r > 2 k r>2k r > 2 k , (4.2) and 2 μ M ≤ k 2 2\mu_M\le k^2 2 μ M ≤ k 2 give r 2 b ( r ) 2 ≤ C σ b ( r ) 2 / r 2 r^2b(r)^2\le C\sigma_b(r)^2/r^2 r 2 b ( r ) 2 ≤ C σ b ( r ) 2 / r 2 . Using (3.3) there and (4.5) inside r ≤ 2 k r\le2k r ≤ 2 k , we obtain
∥ ∣ p ∣ H ∥ 2 2 = O ( k 4 ) , 0 ≤ X ( H ) ≤ C q ∥ H ∥ 2 ∥ ∣ p ∣ H ∥ 2 = O ( q k 3 ) . (30) \||p|H\|_2^2=O(k^4),\qquad0\le\mathcal{X}(H)\le Cq\|H\|_2\||p|H\|_2=O(qk^3). \tag*{(30)} ∥∣ p ∣ H ∥ 2 2 = O ( k 4 ) , 0 ≤ X ( H ) ≤ C q ∥ H ∥ 2 ∥∣ p ∣ H ∥ 2 = O ( q k 3 ) . (30)
The logarithmic integral
Set
Q k = q ∫ ∣ p ∣ > k g ( p ) b ( p ) 2 d 3 p . Q_k=q\int_{|p|>k}g(p)b(p)^2\,\mathrm{d}^3p. Q k = q ∫ ∣ p ∣ > k g ( p ) b ( p ) 2 d 3 p .
We prove
Q k = q k 3 8 π 4 log k + O ( q k 3 ) . (31) Q_k=\frac{qk^3}{8\pi^4}\log k+O(qk^3). \tag*{(31)} Q k = 8 π 4 q k 3 log k + O ( q k 3 ) . (31)
Choose a fixed s 0 > 0 s_0>0 s 0 > 0 , independent of k k k , large enough that 2 g > σ b 2g>\sigma_b 2 g > σ b on r ≥ k + s 0 r\ge k+s_0 r ≥ k + s 0 ; this is possible by (4.3). On this region b = σ b / ( 2 g ) b=\sigma_b/(2g) b = σ b / ( 2 g ) . The shell k < r < k + s 0 k<r<k+s_0 k < r < k + s 0 contributes O ( q k 3 ) O(qk^3) O ( q k 3 ) by (4.4). The tail r > 2 k r>2k r > 2 k also contributes O ( q k 3 ) O(qk^3) O ( q k 3 ) , since g b 2 = σ b 2 / ( 4 g ) ≤ C σ b 2 / r 2 gb^2=\sigma_b^2/(4g)\le C\sigma_b^2/r^2 g b 2 = σ b 2 / ( 4 g ) ≤ C σ b 2 / r 2 and (3.3) applies. Hence
Q k = q 8 π 2 ∫ k + s 0 2 k r 2 σ b ( r ) 2 g ( r ) d r + O ( q k 3 ) . Q_k=\frac{q}{8\pi^2}\int_{k+s_0}^{2k}\frac{r^2\sigma_b(r)^2}{g(r)}\,\mathrm{d}r+O(qk^3). Q k = 8 π 2 q ∫ k + s 0 2 k g ( r ) r 2 σ b ( r ) 2 d r + O ( q k 3 ) .
Because σ b , σ P = O ( k ) \sigma_b,\sigma_P=O(k) σ b , σ P = O ( k ) , (4.6) and ∫ k + s 0 2 k r 2 / g ( r ) d r = O ( k log k ) \int_{k+s_0}^{2k}r^2/g(r)\,\mathrm{d}r=O(k\log k) ∫ k + s 0 2 k r 2 / g ( r ) d r = O ( k log k ) show that replacing σ b 2 \sigma_b^2 σ b 2 by σ P 2 \sigma_P^2 σ P 2 costs O ( q k 8 / 3 log k ) = O ( q k 3 ) O(qk^{8/3}\log k)=O(qk^3) O ( q k 8/3 log k ) = O ( q k 3 ) .
Use the explicit Fermi-ball field (3.8). Writing r = k + s r=k+s r = k + s , s 0 ≤ s ≤ k s_0\le s\le k s 0 ≤ s ≤ k , gives
∣ σ P ( k + s ) − k / π ∣ ≤ C s ( 1 + log ( k / s ) ) , g ( k + s ) ≥ k s . |\sigma_P(k+s)-k/\pi|\le Cs(1+\log(k/s)),\qquad g(k+s)\ge ks. ∣ σ P ( k + s ) − k / π ∣ ≤ C s ( 1 + log ( k / s )) , g ( k + s ) ≥ k s .
It follows that replacing σ P ( r ) 2 \sigma_P(r)^2 σ P ( r ) 2 by k 2 / π 2 k^2/\pi^2 k 2 / π 2 in the preceding integral costs O ( q k 3 ) O(qk^3) O ( q k 3 ) , since ∫ s 0 k ( 1 + log ( k / s ) ) d s = O ( k ) \int_{s_0}^{k}(1+\log(k/s))\,\mathrm{d}s=O(k) ∫ s 0 k ( 1 + log ( k / s )) d s = O ( k ) . Finally, putting g 0 = ( r 2 − k 2 ) / 2 g_0=(r^2-k^2)/2 g 0 = ( r 2 − k 2 ) /2 and using g = g 0 + δ k M g=g_0+\delta_k^{\mathrm{M}} g = g 0 + δ k M ,
∣ ∫ k + s 0 2 k r 2 ( 1 g − 1 g 0 ) d r ∣ ≤ C k ∫ s 0 k s − 2 d s = O ( k ) . \left|\int_{k+s_0}^{2k}r^2\left(\frac{1}{g}-\frac{1}{g_0}\right)\,\mathrm{d}r\right|\le Ck\int_{s_0}^{k}s^{-2}\,\mathrm{d}s=O(k). ∫ k + s 0 2 k r 2 ( g 1 − g 0 1 ) d r ≤ C k ∫ s 0 k s − 2 d s = O ( k ) .
The remaining elementary integral is
∫ k + s 0 2 k r 2 g 0 ( r ) d r = 2 ∫ k + s 0 2 k r 2 r 2 − k 2 d r = k log k + O ( k ) . \int_{k+s_0}^{2k}\frac{r^2}{g_0(r)}\,\mathrm{d}r=2\int_{k+s_0}^{2k}\frac{r^2}{r^2-k^2}\,\mathrm{d}r=k\log k+O(k). ∫ k + s 0 2 k g 0 ( r ) r 2 d r = 2 ∫ k + s 0 2 k r 2 − k 2 r 2 d r = k log k + O ( k ) .
This proves (4.8), with all replacements controlled at order q k 3 qk^3 q k 3 .
Cancellation in the energy and proof of the theorem
The boundary multiplier must be retained. Define the residual
r k ( p ) = 2 g ( p ) b ( p ) − σ b ( p ) . r_k(p)=2g(p)b(p)-\sigma_b(p). r k ( p ) = 2 g ( p ) b ( p ) − σ b ( p ) .
By (4.1), it vanishes on 0 < b < 1 0<b<1 0 < b < 1 and is nonpositive on b = 1 b=1 b = 1 . A quadratic expansion, followed by the equal-mass identity ∫ ( b 2 − P ) d 3 p = 0 \int(b^2-P)\,\mathrm{d}^3p=0 ∫ ( b 2 − P ) d 3 p = 0 , gives the exact formula
F M ( b ) − F M ( P ) = − q ∫ g H 2 d 3 p + X ( H ) + q ∫ H r k d 3 p . (32) \mathcal{F}_{\mathrm{M}}(b)-\mathcal{F}_{\mathrm{M}}(P)=-q\int gH^2\,\mathrm{d}^3p+\mathcal{X}(H)+q\int Hr_k\,\mathrm{d}^3p. \tag*{(32)} F M ( b ) − F M ( P ) = − q ∫ g H 2 d 3 p + X ( H ) + q ∫ H r k d 3 p . (32)
To check it, expand the exchange form around P P P and then insert 2 g b = σ b + r k 2gb=\sigma_b+r_k 2 g b = σ b + r k ; the mass term proportional to μ M \mu_{\mathrm{M}} μ M cancels. All integrals are finite by (4.7) and the estimates below.
On r < κ r<\kappa r < κ we have H = 0 H=0 H = 0 . On κ < r < k \kappa<r<k κ < r < k , 0 ≤ g ≤ δ k M = O ( k ) 0\le g\le\delta_k^{\mathrm{M}}=O(k) 0 ≤ g ≤ δ k M = O ( k ) and H 2 ≤ 1 − b 2 H^2\le1-b^2 H 2 ≤ 1 − b 2 , so the interior part of q ∫ g H 2 q\int gH^2 q ∫ g H 2 is O ( q k 3 ) O(qk^3) O ( q k 3 ) . On r > k r>k r > k , H = b H=b H = b , hence
q ∫ g H 2 d 3 p = Q k + O ( q k 3 ) . q\int gH^2\,\mathrm{d}^3p=Q_k+O(qk^3). q ∫ g H 2 d 3 p = Q k + O ( q k 3 ) .
For the residual term, H r k Hr_k H r k can be nonzero only on pinned states outside the Fermi ball: on pinned states inside, H = 0 H=0 H = 0 , and on fractional states r k = 0 r_k=0 r k = 0 . On the former set, H = 1 H=1 H = 1 , 0 ≤ 2 g ≤ σ b ≤ C k 0\le2g\le\sigma_b\le Ck 0 ≤ 2 g ≤ σ b ≤ C k ; its measure is O ( k 2 ) O(k^2) O ( k 2 ) by (4.5). Therefore
∣ q ∫ H r k d 3 p ∣ = O ( q k 3 ) . \left|q\int Hr_k\,\mathrm{d}^3p\right|=O(qk^3). q ∫ H r k d 3 p = O ( q k 3 ) .
Combining these facts with (4.7), (4.8), and (4.9) yields
E M ( n k M ) − E M ( P k ) = − q k 3 8 π 4 log k + O ( q k 3 ) . \mathcal{E}_{\mathrm{M}}(n_k^{\mathrm{M}})-\mathcal{E}_{\mathrm{M}}(P_k)=-\frac{qk^3}{8\pi^4}\log k+O(qk^3). E M ( n k M ) − E M ( P k ) = − 8 π 4 q k 3 log k + O ( q k 3 ) .
Since ρ = q k 3 / ( 6 π 2 ) \rho=qk^3/(6\pi^2) ρ = q k 3 / ( 6 π 2 ) and E M ( P k ) = ρ e H F h o m ( k ) \mathcal{E}_{\mathrm{M}}(P_k)=\rho e_{\mathrm{HF}}^{\mathrm{hom}}(k) E M ( P k ) = ρ e HF hom ( k ) , Theorem 2.1 follows.
Corollary 4.2 (The exchange defect is twice the leading energy gain). For the Müller minimizer define the nonnegative exchange defect per unit volume by
s ρ = q ∫ b ( p ) ( 1 − b ( p ) ) σ b ( p ) d 3 p . s_\rho=q\int b(p)(1-b(p))\sigma_b(p)\,\mathrm{d}^3p. s ρ = q ∫ b ( p ) ( 1 − b ( p )) σ b ( p ) d 3 p . Then
s ρ = 3 2 π 2 ρ log k + O ( ρ ) , ρ D M ( k ) = 1 2 s ρ + O ( ρ ) . s_\rho=\frac{3}{2\pi^2}\rho\log k+O(\rho),\qquad\rho D_{\mathrm{M}}(k)=\frac{1}{2}s_\rho+O(\rho). s ρ = 2 π 2 3 ρ log k + O ( ρ ) , ρ D M ( k ) = 2 1 s ρ + O ( ρ ) . Proof. The constrained Euler equation implies s ρ = 2 q ∫ g b 2 ( 1 − b ) d 3 p s_\rho=2q\int gb^2(1-b)\,\mathrm{d}^3p s ρ = 2 q ∫ g b 2 ( 1 − b ) d 3 p ; pinned states contribute zero. Inside the Fermi ball this is O ( q k 3 ) O(qk^3) O ( q k 3 ) . Outside, subtract 2 Q k 2Q_k 2 Q k and use g b 3 ≤ σ b b 2 / 2 gb^3\le\sigma_b b^2/2 g b 3 ≤ σ b b 2 /2 and the exterior occupation mass O ( k 2 ) O(k^2) O ( k 2 ) . Thus s ρ = 2 Q k + O ( q k 3 ) s_\rho=2Q_k+O(qk^3) s ρ = 2 Q k + O ( q k 3 ) .
Symmetry reduction and the periodic thermodynamic limit
We apply the convexity-based symmetry averaging of [10 Section I.C] to neutral periodic jellium and identify its thermodynamic variational problem explicitly. Unlike in Hartree–Fock theory [13 ] , restricting Müller jellium to translation–spin invariant states does not change the infimum. For L > 0 L>0 L > 0 let Λ L = R 3 / ( L Z ) 3 \Lambda_L=\mathbb{R}^3/(L\mathbb{Z})^3 Λ L = R 3 / ( L Z ) 3 , Γ L ∗ = ( 2 π / L ) Z 3 \Gamma_L^*=(2\pi/L)\mathbb{Z}^3 Γ L ∗ = ( 2 π / L ) Z 3 , and H L = L 2 ( Λ L ; C q ) \mathcal{H}_L=L^2(\Lambda_L;\mathbb{C}^q) H L = L 2 ( Λ L ; C q ) . Denote by A ρ , L p e r \mathcal{A}^{\mathrm{per}}_{\rho,L} A ρ , L per the convex set of self-adjoint operators γ \gamma γ on H L \mathcal{H}_L H L satisfying 0 ≤ γ ≤ 1 0\leq\gamma\leq1 0 ≤ γ ≤ 1 , tr γ = ρ L 3 \operatorname{tr}\gamma=\rho L^3 tr γ = ρ L 3 , and tr ( − Δ L γ ) : = tr ( ( − Δ L ) 1 / 2 γ ( − Δ L ) 1 / 2 ) < ∞ \operatorname{tr}(-\Delta_L\gamma):=\operatorname{tr}((-\Delta_L)^{1/2}\gamma(-\Delta_L)^{1/2})<\infty tr ( − Δ L γ ) := tr (( − Δ L ) 1/2 γ ( − Δ L ) 1/2 ) < ∞ (taking L → ∞ L\to\infty L → ∞ along ρ L 3 ∈ N \rho L^3\in\mathbb{N} ρ L 3 ∈ N if integer particle number is imposed). With ( U p ψ ) ( x ) = e i p ⋅ x ψ ( x ) (U_p\psi)(x)=e^{ip\cdot x}\psi(x) ( U p ψ ) ( x ) = e i p ⋅ x ψ ( x ) , the neutral periodic Müller energy with zero-mean Coulomb interaction is
E M , L p e r ( γ ) = 1 2 tr ( − Δ L γ ) + 1 2 L 3 ∑ p ∈ Γ L ∗ ∖ { 0 } 4 π ∣ p ∣ 2 ( ∣ tr ( γ U p ∗ ) ∣ 2 − tr ( γ 1 / 2 U p γ 1 / 2 U p ∗ ) ) . (33) \mathcal{E}^{\mathrm{per}}_{\mathrm{M},L}(\gamma)=\frac{1}{2}\operatorname{tr}(-\Delta_L\gamma)+\frac{1}{2L^3}\sum_{p\in\Gamma_L^*\setminus\{0\}}\frac{4\pi}{|p|^2}\left(\left|\operatorname{tr}(\gamma U_p^*)\right|^2-\operatorname{tr}(\gamma^{1/2}U_p\gamma^{1/2}U_p^*)\right). \tag*{(33)} E M , L per ( γ ) = 2 1 tr ( − Δ L γ ) + 2 L 3 1 p ∈ Γ L ∗ ∖ { 0 } ∑ ∣ p ∣ 2 4 π ( tr ( γ U p ∗ ) 2 − tr ( γ 1/2 U p γ 1/2 U p ∗ ) ) . (33)
Adding a Madelung constant μ L = c M a d / L \mu_L=c_{\mathrm{Mad}}/L μ L = c Mad / L to the periodic potential leaves the neutral direct term unchanged and shifts the exchange term only by the state-independent amount − ( μ L / 2 ) tr γ = − c M a d ρ L 2 / 2 -(\mu_L/2)\operatorname{tr}\gamma=-c_{\mathrm{Mad}}\rho L^2/2 − ( μ L /2 ) tr γ = − c Mad ρ L 2 /2 , or O ( L − 1 ) O(L^{-1}) O ( L − 1 ) per particle.
Proposition 4.4. (Symmetry reduction and thermodynamic limit for periodic Müller jellium). For every L > 0 L>0 L > 0 and ρ > 0 \rho>0 ρ > 0 , E M , L p e r \mathcal{E}^{\mathrm{per}}_{\mathrm{M},L} E M , L per is finite and convex on A ρ , L p e r \mathcal{A}^{\mathrm{per}}_{\rho,L} A ρ , L per . For every γ ∈ A ρ , L p e r \gamma\in\mathcal{A}^{\mathrm{per}}_{\rho,L} γ ∈ A ρ , L per , its translation–spin Haar average γ ‾ ∈ A ρ , L p e r \overline{\gamma}\in\mathcal{A}^{\mathrm{per}}_{\rho,L} γ ∈ A ρ , L per has the plane-wave form γ ‾ = ∑ m ∈ Γ L ∗ n L ( m ) ∣ e m ⟩ ⟨ e m ∣ ⊗ I C q \overline{\gamma}=\sum_{m\in\Gamma_L^*}n_L(m)|e_m\rangle\langle e_m|\otimes I_{\mathbb{C}^q} γ = ∑ m ∈ Γ L ∗ n L ( m ) ∣ e m ⟩ ⟨ e m ∣ ⊗ I C q with e m ( x ) = L − 3 / 2 e i m ⋅ x e_m(x)=L^{-3/2}e^{im\cdot x} e m ( x ) = L − 3/2 e im ⋅ x and satisfies
E M , L p e r ( γ ) − E M , L p e r ( γ ‾ ) ≥ 1 2 L 3 ∑ p ∈ Γ L ∗ ∖ { 0 } 4 π ∣ p ∣ 2 ∣ ρ γ ^ ( p ) ∣ 2 ≥ 0 , (34) \mathcal{E}^{\mathrm{per}}_{\mathrm{M},L}(\gamma)-\mathcal{E}^{\mathrm{per}}_{\mathrm{M},L}(\overline{\gamma})\geq\frac{1}{2L^3}\sum_{p\in\Gamma_L^*\setminus\{0\}}\frac{4\pi}{|p|^2}|\widehat{\rho_\gamma}(p)|^2\geq0, \tag*{(34)} E M , L per ( γ ) − E M , L per ( γ ) ≥ 2 L 3 1 p ∈ Γ L ∗ ∖ { 0 } ∑ ∣ p ∣ 2 4 π ∣ ρ γ ( p ) ∣ 2 ≥ 0 , (34) In particular, the energy is strictly decreased whenever ρ γ \rho_\gamma ρ γ is not constant. The infimum is attained, and at least one minimizer is translation invariant and invariant under U ( q ) U(q) U ( q ) spin rotations. Every minimizer has the constant spatial density ρ \rho ρ ; no assertion that every minimizing density matrix is itself invariant is needed here. Moreover, with k = ( 6 π 2 ρ / q ) 1 / 3 k=(6\pi^2\rho/q)^{1/3} k = ( 6 π 2 ρ / q ) 1/3 ,
lim L → ∞ 1 ρ L 3 inf γ ∈ A ρ , L p e r E M , L p e r ( γ ) = e M h o m ( k ) . (35) \lim_{L\to\infty}\frac{1}{\rho L^3}\inf_{\gamma\in\mathcal{A}^{\mathrm{per}}_{\rho,L}}\mathcal{E}^{\mathrm{per}}_{\mathrm{M},L}(\gamma)=e^{\mathrm{hom}}_{\mathrm{M}}(k). \tag*{(35)} L → ∞ lim ρ L 3 1 γ ∈ A ρ , L per inf E M , L per ( γ ) = e M hom ( k ) . (35) The thermodynamic limit is taken at fixed ρ \rho ρ and q q q . The subsequent high-density limit is applied to the resulting energy per particle.
Proof. For γ ∈ A ρ , L p e r \gamma\in\mathcal{A}^{\mathrm{per}}_{\rho,L} γ ∈ A ρ , L per , Schwarz’s inequality on its spectral decomposition gives ∥ ∇ ρ γ ∥ L 2 ( Λ L ) 2 ≤ tr ( − Δ L γ ) < ∞ \|\nabla\sqrt{\rho_\gamma}\|_{L^2(\Lambda_L)}^2\leq\operatorname{tr}(-\Delta_L\gamma)<\infty ∥∇ ρ γ ∥ L 2 ( Λ L ) 2 ≤ tr ( − Δ L γ ) < ∞ , so ρ γ ∈ L 3 ( Λ L ) ⊂ L 2 ( Λ L ) \rho_\gamma\in L^3(\Lambda_L)\subset L^2(\Lambda_L) ρ γ ∈ L 3 ( Λ L ) ⊂ L 2 ( Λ L ) ; since tr ( γ U p ∗ ) = ρ γ ^ ( p ) \operatorname{tr}(\gamma U_p^*)=\widehat{\rho_\gamma}(p) tr ( γ U p ∗ ) = ρ γ ( p ) and ∣ p ∣ ≥ 2 π / L |p|\geq2\pi/L ∣ p ∣ ≥ 2 π / L for p ≠ 0 p\neq0 p = 0 , the direct sum is finite and convex in γ \gamma γ . For each p ≠ 0 p\neq0 p = 0 , f p ( γ ) : = tr ( γ 1 / 2 U p γ 1 / 2 U p ∗ ) = ∥ γ 1 / 4 U p γ 1 / 4 ∥ H S 2 ≥ 0 f_p(\gamma):=\operatorname{tr}(\gamma^{1/2}U_p\gamma^{1/2}U_p^*)=\|\gamma^{1/4}U_p\gamma^{1/4}\|_{\mathrm{HS}}^2\geq0 f p ( γ ) := tr ( γ 1/2 U p γ 1/2 U p ∗ ) = ∥ γ 1/4 U p γ 1/4 ∥ HS 2 ≥ 0 is finite and concave on positive trace-class operators by the Wigner–Yanase theorem, as in [10 ] . For spatial translations T a T_a T a (a ∈ Λ L a\in\Lambda_L a ∈ Λ L ) and spin rotations R U R_U R U (U ∈ U ( q ) U\in U(q) U ∈ U ( q ) ), put V a , U = T a R U V_{a,U}=T_aR_U V a , U = T a R U . Then V a , U ∗ U p V a , U = e − i p ⋅ a U p V_{a,U}^*U_pV_{a,U}=e^{-ip\cdot a}U_p V a , U ∗ U p V a , U = e − i p ⋅ a U p and hence f p ( V a , U γ V a , U ∗ ) = f p ( γ ) f_p(V_{a,U}\gamma V_{a,U}^*)=f_p(\gamma) f p ( V a , U γ V a , U ∗ ) = f p ( γ ) . The Haar Bochner average γ ‾ = L − 3 ∫ Λ L d 3 a ∫ U ( q ) d μ ( U ) V a , U γ V a , U ∗ \overline{\gamma}=L^{-3}\int_{\Lambda_L}d^3a\int_{U(q)}d\mu(U)\,V_{a,U}\gamma V_{a,U}^* γ = L − 3 ∫ Λ L d 3 a ∫ U ( q ) d μ ( U ) V a , U γ V a , U ∗ lies in A ρ , L p e r \mathcal{A}^{\mathrm{per}}_{\rho,L} A ρ , L per , satisfies tr ( − Δ L γ ‾ ) = tr ( − Δ L γ ) \operatorname{tr}(-\Delta_L\overline{\gamma})=\operatorname{tr}(-\Delta_L\gamma) tr ( − Δ L γ ) = tr ( − Δ L γ ) and ρ γ ‾ ^ ( p ) = 0 \widehat{\rho_{\overline{\gamma}}}(p)=0 ρ γ ( p ) = 0 for p ≠ 0 p\neq0 p = 0 , and has the stated plane-wave form by Schur’s lemma, with 0 ≤ n L ( m ) ≤ 1 0\leq n_L(m)\leq1 0 ≤ n L ( m ) ≤ 1 and ( q / L 3 ) ∑ m n L ( m ) = ρ (q/L^3)\sum_m n_L(m)=\rho ( q / L 3 ) ∑ m n L ( m ) = ρ . Approximate γ ‾ \overline{\gamma} γ in trace norm by Riemann-sum convex combinations γ j \gamma_j γ j of V a , U γ V a , U ∗ V_{a,U}\gamma V_{a,U}^* V a , U γ V a , U ∗ . Since ∥ γ j 1 / 2 − γ ‾ 1 / 2 ∥ → 0 \|\gamma_j^{1/2}-\overline{\gamma}^{1/2}\|\to0 ∥ γ j 1/2 − γ 1/2 ∥ → 0 and tr γ j = ρ L 3 = tr γ ‾ \operatorname{tr}\gamma_j=\rho L^3=\operatorname{tr}\overline{\gamma} tr γ j = ρ L 3 = tr γ , testing against finite-rank spectral projections of γ ‾ \overline{\gamma} γ yields tr ( γ j 1 / 2 γ ‾ 1 / 2 ) → tr γ ‾ \operatorname{tr}(\gamma_j^{1/2}\overline{\gamma}^{1/2})\to\operatorname{tr}\overline{\gamma} tr ( γ j 1/2 γ 1/2 ) → tr γ and hence ∥ γ j 1 / 2 − γ ‾ 1 / 2 ∥ H S → 0 \|\gamma_j^{1/2}-\overline{\gamma}^{1/2}\|_{\mathrm{HS}}\to0 ∥ γ j 1/2 − γ 1/2 ∥ HS → 0 . Thus f p ( γ j ) = ⟨ U p γ j 1 / 2 , γ j 1 / 2 U p ⟩ H S → f p ( γ ‾ ) f_p(\gamma_j)=\langle U_p\gamma_j^{1/2},\gamma_j^{1/2}U_p\rangle_{\mathrm{HS}}\to f_p(\overline{\gamma}) f p ( γ j ) = ⟨ U p γ j 1/2 , γ j 1/2 U p ⟩ HS → f p ( γ ) , while concavity gives f p ( γ j ) ≥ f p ( γ ) f_p(\gamma_j)\geq f_p(\gamma) f p ( γ j ) ≥ f p ( γ ) , so f p ( γ ‾ ) ≥ f p ( γ ) f_p(\overline{\gamma})\geq f_p(\gamma) f p ( γ ) ≥ f p ( γ ) for every p ≠ 0 p\neq0 p = 0 .
Associate to n L n_L n L the step function n ( L ) ∈ A ρ n^{(L)}\in\mathcal{A}_\rho n ( L ) ∈ A ρ equal to n L ( m ) n_L(m) n L ( m ) on Q m = m + [ − π / L , π / L ] 3 Q_m=m+[-\pi/L,\pi/L]^3 Q m = m + [ − π / L , π / L ] 3 . Its kinetic moment is
∫ ∣ p ∣ 2 n ( L ) ( p ) d 3 p = L − 3 ∑ m ∈ Γ L ∗ ∣ m ∣ 2 n L ( m ) + π 2 ρ q L 2 . \int|p|^2n^{(L)}(p)\,d^3p=L^{-3}\sum_{m\in\Gamma_L^*}|m|^2n_L(m)+\frac{\pi^2\rho}{qL^2}. ∫ ∣ p ∣ 2 n ( L ) ( p ) d 3 p = L − 3 m ∈ Γ L ∗ ∑ ∣ m ∣ 2 n L ( m ) + q L 2 π 2 ρ .
To make the exchange comparison explicit, set Q 0 u n i t = [ − 1 / 2 , 1 / 2 ) 3 Q^{\mathrm{unit}}_0=[-1/2,1/2)^3 Q 0 unit = [ − 1/2 , 1/2 ) 3 and, for ν ∈ Z 3 \nu\in\mathbb{Z}^3 ν ∈ Z 3 , define
a ν = ∬ Q 0 u n i t × Q 0 u n i t d 3 u d 3 v ∣ ν + u − v ∣ 2 , c ν = { ∣ ν ∣ − 2 , ν ≠ 0 , 0 , ν = 0. a_\nu=\iint_{Q^{\mathrm{unit}}_0\times Q^{\mathrm{unit}}_0}\frac{\mathrm{d}^3u\,\mathrm{d}^3v}{|\nu+u-v|^2},\qquad
c_\nu=\begin{cases}|\nu|^{-2},&\nu\ne0,\\0,&\nu=0.\end{cases} a ν = ∬ Q 0 unit × Q 0 unit ∣ ν + u − v ∣ 2 d 3 u d 3 v , c ν = { ∣ ν ∣ − 2 , 0 , ν = 0 , ν = 0.
The singularity is locally integrable, so each a ν a_\nu a ν is finite, including a 0 a_0 a 0 . For large ∣ ν ∣ |\nu| ∣ ν ∣ , the linear Taylor term vanishes by symmetry and a ν = ∣ ν ∣ − 2 + O ( ∣ ν ∣ − 4 ) a_\nu=|\nu|^{-2}+O(|\nu|^{-4}) a ν = ∣ ν ∣ − 2 + O ( ∣ ν ∣ − 4 ) . Thus
∑ ν ∈ Z 3 ∣ a ν − c ν ∣ < ∞ . \sum_{\nu\in\mathbb{Z}^3}|a_\nu-c_\nu|<\infty. ν ∈ Z 3 ∑ ∣ a ν − c ν ∣ < ∞.
Using f p ( γ ‾ ) = q ∑ m n L ( m ) n L ( m − p ) f_p(\overline{\gamma})=q\sum_m\sqrt{n_L(m)n_L(m-p)} f p ( γ ) = q ∑ m n L ( m ) n L ( m − p ) and
∑ m ∈ Γ L ∗ n L ( m ) n L ( m − 2 π ν / L ) ≤ ∑ m n L ( m ) = ρ L 3 q , \sum_{m\in\Gamma_L^*}\sqrt{n_L(m)n_L(m-2\pi\nu/L)}
\le\sum_m n_L(m)=\frac{\rho L^3}{q}, m ∈ Γ L ∗ ∑ n L ( m ) n L ( m − 2 π ν / L ) ≤ m ∑ n L ( m ) = q ρ L 3 ,
rescaling the two momentum cells gives
∣ 1 2 L 6 ∑ p ∈ Γ L ∗ ∖ { 0 } 4 π ∣ p ∣ 2 [ f p ( γ ‾ ) − X ( n ( L ) ) ] ∣ ≤ C ρ L . (36) \left|\frac{1}{2L^6}\sum_{p\in\Gamma_L^*\setminus\{0\}}\frac{4\pi}{|p|^2}\left[f_p(\overline{\gamma})-\mathcal{X}\bigl(\sqrt{n^{(L)}}\bigr)\right]\right|\le\frac{C\rho}{L}. \tag*{(36)} 2 L 6 1 p ∈ Γ L ∗ ∖ { 0 } ∑ ∣ p ∣ 2 4 π [ f p ( γ ) − X ( n ( L ) ) ] ≤ L C ρ . (36)
where C C C is absolute. By (12) , X ( n ( L ) ) < ∞ \mathcal{X}\bigl(\sqrt{n^{(L)}}\bigr)<\infty X ( n ( L ) ) < ∞ ; since 0 ≤ f p ( γ ) ≤ f p ( γ ‾ ) 0\le f_p(\gamma)\le f_p(\overline{\gamma}) 0 ≤ f p ( γ ) ≤ f p ( γ ) , the exchange sum in (33) converges for every γ ∈ A ρ , L p e r \gamma\in\mathcal{A}^{\mathrm{per}}_{\rho,L} γ ∈ A ρ , L per , proving finiteness and convexity. More precisely, the averaging argument gives
E M , L p e r ( γ ) − E M , L p e r ( γ ‾ ) = D L ( ρ γ − ρ ) + 1 2 L 3 ∑ p ∈ Γ L ∗ ∖ { 0 } 4 π ∣ p ∣ 2 [ f p ( γ ‾ ) − f p ( γ ) ] , \mathcal{E}^{\mathrm{per}}_{M,L}(\gamma)-\mathcal{E}^{\mathrm{per}}_{M,L}(\overline{\gamma})
=D_L(\rho_\gamma-\rho)+\frac{1}{2L^3}\sum_{p\in\Gamma_L^*\setminus\{0\}}\frac{4\pi}{|p|^2}\left[f_p(\overline{\gamma})-f_p(\gamma)\right], E M , L per ( γ ) − E M , L per ( γ ) = D L ( ρ γ − ρ ) + 2 L 3 1 p ∈ Γ L ∗ ∖ { 0 } ∑ ∣ p ∣ 2 4 π [ f p ( γ ) − f p ( γ ) ] ,
where D L ( ρ γ − ρ ) D_L(\rho_\gamma-\rho) D L ( ρ γ − ρ ) denotes the nonnegative direct term in (33) . This proves (34) . The kinetic identity and (36) also yield
E M , L p e r ( γ ‾ ) L 3 = E M ( n ( L ) ) − π 2 ρ 2 L 2 + O ( ρ / L ) , \frac{\mathcal{E}^{\mathrm{per}}_{M,L}(\overline{\gamma})}{L^3}
=\mathcal{E}_M\bigl(n^{(L)}\bigr)-\frac{\pi^2\rho}{2L^2}+O(\rho/L), L 3 E M , L per ( γ ) = E M ( n ( L ) ) − 2 L 2 π 2 ρ + O ( ρ / L ) ,
and hence the lower bound
( ρ L 3 ) − 1 E M , L p e r ( γ ) ≥ e M h o m ( k ) − O ( L − 1 + L − 2 ) . (\rho L^3)^{-1}\mathcal{E}^{\mathrm{per}}_{M,L}(\gamma)\ge e^{\mathrm{hom}}_M(k)-O(L^{-1}+L^{-2}). ( ρ L 3 ) − 1 E M , L per ( γ ) ≥ e M hom ( k ) − O ( L − 1 + L − 2 ) .
We next prove attainment for fixed L L L and N = ρ L 3 N=\rho L^3 N = ρ L 3 . Let v L v_L v L be the zero-mean periodic Coulomb potential from (33) , and let X L ( A ) X_L(A) X L ( A ) denote its exchange quadratic form in the kernel A A A . The bound
∣ v L ( z ) ∣ ≤ C dist Λ L ( z , 0 ) + C L |v_L(z)|\le\frac{C}{\operatorname{dist}_{\Lambda_L}(z,0)}+\frac{C}{L} ∣ v L ( z ) ∣ ≤ dist Λ L ( z , 0 ) C + L C
and the periodic Hardy inequality, applied in the first kernel variable and then integrated in the second, imply
∣ X L ( γ ) ∣ ≤ C N T L ( γ ) + C N L , T L ( γ ) = tr ( − Δ L γ ) . |X_L(\sqrt{\gamma})|\le C\sqrt{N T_L(\gamma)}+C\frac{N}{L},\qquad
T_L(\gamma)=\operatorname{tr}(-\Delta_L\gamma). ∣ X L ( γ ) ∣ ≤ C N T L ( γ ) + C L N , T L ( γ ) = tr ( − Δ L γ ) .
Here ∥ γ ∥ H S 2 = N \|\sqrt{\gamma}\|_{\mathrm{HS}}^2=N ∥ γ ∥ HS 2 = N and ∥ ∇ x γ ∥ H S 2 = T L ( γ ) \|\nabla_x\sqrt{\gamma}\|_{\mathrm{HS}}^2=T_L(\gamma) ∥ ∇ x γ ∥ HS 2 = T L ( γ ) . Since the direct term is nonnegative, a minimizing sequence γ j \gamma_j γ j has uniformly bounded kinetic energy. The compact resolvent of − Δ L -\Delta_L − Δ L then gives, after extraction, γ j → γ \gamma_j\to\gamma γ j → γ in trace norm. Indeed, the trace outside a kinetic spectral cutoff at energy M M M is bounded by T L ( γ j ) / M T_L(\gamma_j)/M T L ( γ j ) / M ; the corresponding finite-rank compressions are compact in trace norm. Thus 0 ≤ γ ≤ 1 0\le\gamma\le1 0 ≤ γ ≤ 1 , tr γ = N \operatorname{tr}\gamma=N tr γ = N , and T L ( γ ) ≤ lim inf j T L ( γ j ) T_L(\gamma)\le\liminf_j T_L(\gamma_j) T L ( γ ) ≤ lim inf j T L ( γ j ) .
Put A j = γ j A_j=\sqrt{\gamma_j} A j = γ j and A = γ A=\sqrt{\gamma} A = γ . The square-root continuity used above gives ∥ A j − A ∥ H S → 0 \|A_j-A\|_{\mathrm{HS}}\to0 ∥ A j − A ∥ HS → 0 . The periodic Hardy inequality and Schwarz’s inequality yield
∣ X L ( A j ) − X L ( A ) ∣ ≤ C ∥ A j − A ∥ H S ( ∥ ∇ x A j ∥ H S + ∥ ∇ x A ∥ H S + L − 1 N ) ⟶ 0. |X_L(A_j)-X_L(A)|
\le C\|A_j-A\|_{\mathrm{HS}}
\left(\|\nabla_xA_j\|_{\mathrm{HS}}+\|\nabla_xA\|_{\mathrm{HS}}+L^{-1}\sqrt{N}\right)\longrightarrow0. ∣ X L ( A j ) − X L ( A ) ∣ ≤ C ∥ A j − A ∥ HS ( ∥ ∇ x A j ∥ HS + ∥ ∇ x A ∥ HS + L − 1 N ) ⟶ 0.
The direct term is lower semicontinuous by its nonnegative Fourier series: each coefficient is continuous under trace-norm convergence, and Fatou’s lemma applies to the sum. Hence γ \gamma γ attains the infimum. Averaging a minimizer provides a translation–spin invariant minimizer. Moreover, (34) forces every minimizer to have constant spatial density. This does not require uniqueness of the minimizing density matrix.
For the upper thermodynamic bound, let n ∈ A ρ n \in\mathcal{A}_{\rho} n ∈ A ρ minimize E M \mathcal{E}_{M} E M and set n L ( m ) = L 3 ∫ Q m n d 3 p n_{L}(m)=L^{3}\int_{Q_{m}}n\,\mathrm{d}^{3}p n L ( m ) = L 3 ∫ Q m n d 3 p . For p ∈ Q m p \in Q_{m} p ∈ Q m , ∣ m ∣ 2 ≤ ∣ p ∣ 2 + 2 3 π L − 1 ∣ p ∣ + 3 π 2 L − 2 |m|^{2}\leq|p|^{2}+2\sqrt{3}\pi L^{-1}|p|+3\pi^{2}L^{-2} ∣ m ∣ 2 ≤ ∣ p ∣ 2 + 2 3 π L − 1 ∣ p ∣ + 3 π 2 L − 2 , so the associated plane-wave state γ ‾ L ∈ A ρ , L p e r \overline{\gamma}_{L}\in\mathcal{A}^{\mathrm{per}}_{\rho,L} γ L ∈ A ρ , L per satisfies
( 2 L 3 ) − 1 tr ( − Δ L γ ‾ L ) ≤ q 2 ∫ ∣ p ∣ 2 n d 3 p + O ( L − 1 ) . (2L^{3})^{-1}\operatorname{tr}(-\Delta_{L}\overline{\gamma}_{L})\leq\frac{q}{2}\int|p|^{2}n\,\mathrm{d}^{3}p+O(L^{-1}). ( 2 L 3 ) − 1 tr ( − Δ L γ L ) ≤ 2 q ∫ ∣ p ∣ 2 n d 3 p + O ( L − 1 ) .
Moreover, the step function n ( L ) n^{(L)} n ( L ) converges to n n n in L 1 ( R 3 , d 3 p ) L^{1}(\mathbb{R}^{3},\mathrm{d}^{3}p) L 1 ( R 3 , d 3 p ) , hence ∥ n ( L ) − n ∥ 2 2 ≤ ∥ n ( L ) − n ∥ 1 → 0 \|\sqrt{n^{(L)}}-\sqrt{n}\|_{2}^{2}\leq\|n^{(L)}-n\|_{1}\to0 ∥ n ( L ) − n ∥ 2 2 ≤ ∥ n ( L ) − n ∥ 1 → 0 with uniformly bounded kinetic moment. Lemma 3.1 and (36) give lim sup L → ∞ ( ρ L 3 ) − 1 E M , L p e r ( γ ‾ L ) ≤ e M h o m ( k ) \limsup_{L\to\infty}(\rho L^{3})^{-1}\mathcal{E}^{\mathrm{per}}_{M,L}(\overline{\gamma}_{L})\leq e^{\mathrm{hom}}_{M}(k) lim sup L → ∞ ( ρ L 3 ) − 1 E M , L per ( γ L ) ≤ e M hom ( k ) , proving (35) .
The CHF minimizer: strict occupation and uniqueness
A strict competitor for the Fermi ball
Use the volume coordinate (20) . If n ∈ A ρ n\in\mathcal{A}_{\rho} n ∈ A ρ satisfies w n = 0 w_{n}=0 w n = 0 almost everywhere, then E ( n ) = E H F ( n ) \mathcal{E}(n)=\mathcal{E}_{\mathrm{HF}}(n) E ( n ) = E HF ( n ) . The bathtub principle and (8) therefore imply
E ( n ) ≥ E ( P k ) . (37) \mathcal{E}(n)\geq\mathcal{E}(P_{k}). \tag*{(37)} E ( n ) ≥ E ( P k ) . (37)
Here w n = 0 w_{n}=0 w n = 0 is equivalent to n ∈ { 0 , 1 } n\in\{0,1\} n ∈ { 0 , 1 } almost everywhere.
Lemma 5.1 (The Fermi sea is not optimal). For every ρ > 0 \rho>0 ρ > 0 , inf E < E ( P k ) \inf\mathcal{E}<\mathcal{E}(P_{k}) inf E < E ( P k ) .
Proof. For 0 < ε < 1 / 2 0<\varepsilon<1/2 0 < ε < 1/2 and 0 < δ < k / 6 0<\delta<k/6 0 < δ < k /6 , define
n ε , δ ( t ) = { 1 − ε , − δ < t < 0 , ε , 0 < t < δ , 1 { t < 0 } , ∣ t ∣ ≥ δ . n_{\varepsilon,\delta}(t)=
\begin{cases}
1-\varepsilon, & -\delta<t<0,\\
\varepsilon, & 0<t<\delta,\\
\mathbf{1}_{\{t<0\}}, & |t|\geq\delta.
\end{cases} n ε , δ ( t ) = ⎩ ⎨ ⎧ 1 − ε , ε , 1 { t < 0 } , − δ < t < 0 , 0 < t < δ , ∣ t ∣ ≥ δ .
It has exactly the required mass, and w n ε , δ = ε ( 1 − ε ) 1 { ∣ t ∣ < δ } w_{n_{\varepsilon,\delta}}=\sqrt{\varepsilon(1-\varepsilon)}\mathbf{1}_{\{|t|<\delta\}} w n ε , δ = ε ( 1 − ε ) 1 { ∣ t ∣ < δ } . Using (18) and (19) and expanding at t = 0 t=0 t = 0 gives, for fixed k , ε k,\varepsilon k , ε , as δ ↓ 0 \delta\downarrow0 δ ↓ 0 ,
E ( n ε , δ ) − E ( P k ) ≤ q ε k 2 δ 2 2 π 3 [ π k − ( 1 − 2 ε ) log k δ + O k , ε ( 1 ) ] . \mathcal{E}(n_{\varepsilon,\delta})-\mathcal{E}(P_{k})\leq\frac{q\varepsilon k^{2}\delta^{2}}{2\pi^{3}}\left[\pi k-(1-2\varepsilon)\log\frac{k}{\delta}+O_{k,\varepsilon}(1)\right]. E ( n ε , δ ) − E ( P k ) ≤ 2 π 3 q ε k 2 δ 2 [ π k − ( 1 − 2 ε ) log δ k + O k , ε ( 1 ) ] .
The quadratic contribution − X ( n ε , δ − P k ) -\mathcal{X}(n_{\varepsilon,\delta}-P_{k}) − X ( n ε , δ − P k ) may be dropped for this upper bound. Since 1 − 2 ε > 0 1-2\varepsilon>0 1 − 2 ε > 0 , the right-hand side is negative when δ \delta δ is small.
Lemma 5.2 (Strictly fractional occupations). Every minimizer n ∗ n_{*} n ∗ satisfies 0 < n ∗ < 1 0<n_{*}<1 0 < n ∗ < 1 almost everywhere.
Proof. Lemma 5.1 and (37) show that w ∗ : = n ∗ ( 1 − n ∗ ) w_{*}:=\sqrt{n_{*}(1-n_{*})} w ∗ := n ∗ ( 1 − n ∗ ) is not identically zero. Consequently Δ ∗ : = V w ∗ > 0 \Delta_{*}:=\mathcal{V}w_{*}>0 Δ ∗ := V w ∗ > 0 at every finite momentum; it is bounded and continuous because w ∗ ≤ 1 / 2 w_{*}\leq1/2 w ∗ ≤ 1/2 and w ∗ ∈ L 2 w_{*}\in L^{2} w ∗ ∈ L 2 . If { n ∗ = 0 } \{n_{*}=0\} { n ∗ = 0 } has positive measure, choose a bounded positive-measure subset A A A of that set, and a bounded set B B B of positive measure on which η ≤ n ∗ ≤ 1 − η \eta\leq n_{*}\leq1-\eta η ≤ n ∗ ≤ 1 − η for some η > 0 \eta>0 η > 0 . For small ε > 0 \varepsilon>0 ε > 0 the mass-preserving variation
n ε = n ∗ + ε 1 A − ε ∣ A ∣ ∣ B ∣ 1 B n_{\varepsilon}=n_{*}+\varepsilon\mathbf{1}_{A}-\varepsilon\frac{|A|}{|B|}\mathbf{1}_{B} n ε = n ∗ + ε 1 A − ε ∣ B ∣ ∣ A ∣ 1 B
is admissible. Kinetic energy and ordinary exchange change by O ( ε ) O(\varepsilon) O ( ε ) . On A A A the new w w w is ε + O ( ε 3 / 2 ) \sqrt{\varepsilon}+O(\varepsilon^{3/2}) ε + O ( ε 3/2 ) , while its change on B B B is O ( ε ) O(\varepsilon) O ( ε ) . Since X \mathcal{X} X is a positive quadratic form,
X ( w n ε ) − X ( w ∗ ) ≥ q ε ∫ A Δ ∗ d 3 p + O ( ε ) . \mathcal{X}(w_{n_{\varepsilon}})-\mathcal{X}(w_{*})\geq q\sqrt{\varepsilon}\int_{A}\Delta_{*}\,\mathrm{d}^{3}p+O(\varepsilon). X ( w n ε ) − X ( w ∗ ) ≥ q ε ∫ A Δ ∗ d 3 p + O ( ε ) .
The leading coefficient is positive, contradicting minimality. The set { n ∗ = 1 } \{n_{*}=1\} { n ∗ = 1 } is excluded by transferring mass in the opposite direction.
Theorem 5.3 (Uniqueness and self-consistency). The minimizer n ρ n_\rho n ρ is unique. It has a continuous radial representative satisfying 0 < n ρ ( p ) < 1 0<n_\rho(p)<1 0 < n ρ ( p ) < 1 for every p p p . There is a unique μ ρ ∈ R \mu_\rho\in\mathbb{R} μ ρ ∈ R such that, with
ξ ( p ) = ∣ p ∣ 2 2 − ( V n ρ ) ( p ) − μ ρ , Δ ( p ) = ( V w ρ ) ( p ) , E ( p ) = ξ ( p ) 2 + Δ ( p ) 2 , (38) \xi(p)=\frac{|p|^2}{2}-(Vn_\rho)(p)-\mu_\rho,\qquad\Delta(p)=(Vw_\rho)(p),\qquad E(p)=\sqrt{\xi(p)^2+\Delta(p)^2}, \tag*{(38)} ξ ( p ) = 2 ∣ p ∣ 2 − ( V n ρ ) ( p ) − μ ρ , Δ ( p ) = ( V w ρ ) ( p ) , E ( p ) = ξ ( p ) 2 + Δ ( p ) 2 , (38) one has pointwise
n ρ = 1 2 ( 1 − ξ E ) , w ρ = Δ 2 E , Δ > 0. (39) n_\rho=\frac{1}{2}\left(1-\frac{\xi}{E}\right),\qquad w_\rho=\frac{\Delta}{2E},\qquad\Delta>0. \tag*{(39)} n ρ = 2 1 ( 1 − E ξ ) , w ρ = 2 E Δ , Δ > 0. (39)
Proof. Local variations on { p : ∣ p ∣ ≤ j , j − 1 ≤ n ρ ( p ) ≤ 1 − j − 1 } \{p:|p|\le j,\ j^{-1}\le n_\rho(p)\le1-j^{-1}\} { p : ∣ p ∣ ≤ j , j − 1 ≤ n ρ ( p ) ≤ 1 − j − 1 } , with zero integral, give the common Lagrange multiplier μ ρ \mu_\rho μ ρ and the a.e. Euler equation
∣ p ∣ 2 2 − ( V n ρ ) ( p ) − μ ρ = Δ ( p ) 1 − 2 n ρ ( p ) 2 w ρ ( p ) . (40) \frac{|p|^2}{2}-(Vn_\rho)(p)-\mu_\rho=\Delta(p)\frac{1-2n_\rho(p)}{2w_\rho(p)}. \tag*{(40)} 2 ∣ p ∣ 2 − ( V n ρ ) ( p ) − μ ρ = Δ ( p ) 2 w ρ ( p ) 1 − 2 n ρ ( p ) . (40)
The fields are continuous, bounded, and Δ > 0 \Delta>0 Δ > 0 . Algebraic inversion of (5.4) gives (5.3), defining a continuous representative strictly between zero and one. Uniqueness of μ ρ \mu_\rho μ ρ is immediate from (5.4).
Let n n n be another admissible occupation of the same mass and put h = n − n ρ h=n-n_\rho h = n − n ρ , g = w n − w ρ g=w_n-w_\rho g = w n − w ρ . Since ∫ h d 3 p = 0 \int h\,\mathrm{d}^3p=0 ∫ h d 3 p = 0 , expansion of the quadratic forms gives
E ( n ) − E ( n ρ ) = q ∫ E ( h 2 + g 2 ) d 3 p − X ( h ) − X ( g ) . (41) \mathcal{E}(n)-\mathcal{E}(n_\rho)=q\int E(h^2+g^2)\,\mathrm{d}^3p-\mathcal{X}(h)-\mathcal{X}(g). \tag*{(41)} E ( n ) − E ( n ρ ) = q ∫ E ( h 2 + g 2 ) d 3 p − X ( h ) − X ( g ) . (41)
Here one uses the exact pointwise circle identity
h 2 + g 2 = ( 1 − 2 n ρ ) h − 2 w ρ g h^2+g^2=(1-2n_\rho)h-2w_\rho g h 2 + g 2 = ( 1 − 2 n ρ ) h − 2 w ρ g
and the Euler relations ξ = E ( 1 − 2 n ρ ) \xi=E(1-2n_\rho) ξ = E ( 1 − 2 n ρ ) , Δ = 2 E w ρ \Delta=2Ew_\rho Δ = 2 E w ρ . We justify symmetrization and the use of Fubini’s theorem. The bounded self-consistent potentials imply 0 < E ( p ) ≤ C ( 1 + ∣ p ∣ 2 ) 0<E(p)\le C(1+|p|^2) 0 < E ( p ) ≤ C ( 1 + ∣ p ∣ 2 ) , with ρ \rho ρ fixed. For f = h f=h f = h or f = g f=g f = g , the inequalities
h 2 ≤ n + n ρ , g 2 ≤ 2 ( n + n ρ ) h^2\le n+n_\rho,\qquad g^2\le2(n+n_\rho) h 2 ≤ n + n ρ , g 2 ≤ 2 ( n + n ρ )
and the finite kinetic energies give
∫ R 3 E ( p ) f ( p ) 2 d 3 p < ∞ . (42) \int_{\mathbb{R}^3}E(p)f(p)^2\,\mathrm{d}^3p<\infty. \tag*{(42)} ∫ R 3 E ( p ) f ( p ) 2 d 3 p < ∞. (42)
Put K ( p , p ′ ) = 4 π ∣ p − p ′ ∣ − 2 K(p,p')=4\pi|p-p'|^{-2} K ( p , p ′ ) = 4 π ∣ p − p ′ ∣ − 2 . Since w ρ > 0 w_\rho>0 w ρ > 0 and V w ρ = 2 E w ρ Vw_\rho=2Ew_\rho V w ρ = 2 E w ρ , Tonelli’s theorem gives
∬ K ( p , p ′ ) w ρ ( p ) w ρ ( p ′ ) f ( p ) 2 w ρ ( p ) 2 d 3 p d 3 p ′ = ∫ f ( p ) 2 w ρ ( p ) ( V w ρ ) ( p ) d 3 p = 2 ∫ E ( p ) f ( p ) 2 d 3 p < ∞ . \iint K(p,p')w_\rho(p)w_\rho(p')\frac{f(p)^2}{w_\rho(p)^2}\,\mathrm{d}^3p\,\mathrm{d}^3p'
=\int\frac{f(p)^2}{w_\rho(p)}(Vw_\rho)(p)\,\mathrm{d}^3p
=2\int E(p)f(p)^2\,\mathrm{d}^3p<\infty. ∬ K ( p , p ′ ) w ρ ( p ) w ρ ( p ′ ) w ρ ( p ) 2 f ( p ) 2 d 3 p d 3 p ′ = ∫ w ρ ( p ) f ( p ) 2 ( V w ρ ) ( p ) d 3 p = 2 ∫ E ( p ) f ( p ) 2 d 3 p < ∞.
The same holds with p p p and p ′ p' p ′ interchanged. Moreover,
2 ∣ f ( p ) f ( p ′ ) w ρ ( p ) w ρ ( p ′ ) ∣ ≤ f ( p ) 2 w ρ ( p ) 2 + f ( p ′ ) 2 w ρ ( p ′ ) 2 . 2\left|\frac{f(p)f(p')}{w_\rho(p)w_\rho(p')}\right|
\le\frac{f(p)^2}{w_\rho(p)^2}+\frac{f(p')^2}{w_\rho(p')^2}. 2 w ρ ( p ) w ρ ( p ′ ) f ( p ) f ( p ′ ) ≤ w ρ ( p ) 2 f ( p ) 2 + w ρ ( p ′ ) 2 f ( p ′ ) 2 .
All terms in the expansion of the squared difference are therefore absolutely integrable. Fubini’s theorem applies and yields
∬ K ( p , p ′ ) w ρ ( p ) w ρ ( p ′ ) ( f ( p ) w ρ ( p ) − f ( p ′ ) w ρ ( p ′ ) ) 2 d 3 p d 3 p ′ = 4 ∫ E ( p ) f ( p ) 2 d 3 p − 2 ∫ f ( p ) ( V f ) ( p ) d 3 p . \iint K(p,p')w_\rho(p)w_\rho(p')\left(\frac{f(p)}{w_\rho(p)}-\frac{f(p')}{w_\rho(p')}\right)^2\,\mathrm{d}^3p\,\mathrm{d}^3p'
=4\int E(p)f(p)^2\,\mathrm{d}^3p-2\int f(p)(Vf)(p)\,\mathrm{d}^3p. ∬ K ( p , p ′ ) w ρ ( p ) w ρ ( p ′ ) ( w ρ ( p ) f ( p ) − w ρ ( p ′ ) f ( p ′ ) ) 2 d 3 p d 3 p ′ = 4 ∫ E ( p ) f ( p ) 2 d 3 p − 2 ∫ f ( p ) ( V f ) ( p ) d 3 p .
Applying this identity to h h h and g g g gives
E ( n ) − E ( n ρ ) = q 4 ∬ 4 π w ρ ( p ) w ρ ( p ′ ) ∣ p − p ′ ∣ 2 [ ( h ( p ) w ρ ( p ) − h ( p ′ ) w ρ ( p ′ ) ) 2 + ( g ( p ) w ρ ( p ) − g ( p ′ ) w ρ ( p ′ ) ) 2 ] d 3 p d 3 p ′ . (43) \mathcal{E}(n)-\mathcal{E}(n_\rho)=\frac{q}{4}\iint\frac{4\pi w_\rho(p)w_\rho(p')}{\lvert p-p'\rvert^2}\left[\left(\frac{h(p)}{w_\rho(p)}-\frac{h(p')}{w_\rho(p')}\right)^2+\left(\frac{g(p)}{w_\rho(p)}-\frac{g(p')}{w_\rho(p')}\right)^2\right]\mathrm{d}^3p\,\mathrm{d}^3p'. \tag*{(43)} E ( n ) − E ( n ρ ) = 4 q ∬ ∣ p − p ′ ∣ 2 4 π w ρ ( p ) w ρ ( p ′ ) [ ( w ρ ( p ) h ( p ) − w ρ ( p ′ ) h ( p ′ ) ) 2 + ( w ρ ( p ) g ( p ) − w ρ ( p ′ ) g ( p ′ ) ) 2 ] d 3 p d 3 p ′ . (43)
The right-hand side is finite and nonnegative. If it vanishes, then h = c w ρ h=cw_\rho h = c w ρ almost everywhere, since the kernel is strictly positive. Here h ∈ L 1 h\in L^1 h ∈ L 1 and ∫ h d 3 p = 0 \int h\mathrm{d}^3p=0 ∫ h d 3 p = 0 ; positivity of w ρ w_\rho w ρ therefore forces c = 0 c=0 c = 0 . This does not require w ρ ∈ L 1 w_\rho\in L^1 w ρ ∈ L 1 : if c ≠ 0 c\ne0 c = 0 , then h = c w ρ h=cw_\rho h = c w ρ has a fixed nonzero sign, contradicting its zero integral. Hence every minimizer equals n ρ n_\rho n ρ . Rotations preserve the problem; uniqueness gives radiality.
For the connection between CHF optimization and Hartree–Fock–Bogoliubov methods, see [22 ] . The self-consistent equations and the stability identity needed here have been derived directly above.
A coercive deterministic comparison for CHF
The upper bound will follow from a comparison for every n ∈ A ρ n\in\mathcal{A}_\rho n ∈ A ρ , without using its Euler equation or radiality. Put
μ H F = k 2 2 − k π , \mu_{\mathrm{HF}}=\frac{k^2}{2}-\frac{k}{\pi}, μ HF = 2 k 2 − π k ,
ξ H F ( r ) = r 2 2 − ( V P k ) ( r ) − μ H F = r 2 − k 2 2 ( 1 + 1 π r log r + k ∣ r − k ∣ ) . (44) \xi_{\mathrm{HF}}(r)=\frac{r^2}{2}-(V P_k)(r)-\mu_{\mathrm{HF}}
=\frac{r^2-k^2}{2}\left(1+\frac{1}{\pi r}\log\frac{r+k}{\lvert r-k\rvert}\right). \tag*{(44)} ξ HF ( r ) = 2 r 2 − ( V P k ) ( r ) − μ HF = 2 r 2 − k 2 ( 1 + π r 1 log ∣ r − k ∣ r + k ) . (44)
e k ( r ) : = ∣ ξ H F ( r ) ∣ ≥ 1 2 ∣ r 2 − k 2 ∣ . (45) e_k(r):=\lvert\xi_{\mathrm{HF}}(r)\rvert\ge\frac{1}{2}\lvert r^2-k^2\rvert. \tag*{(45)} e k ( r ) := ∣ ξ HF ( r )∣ ≥ 2 1 ∣ r 2 − k 2 ∣ . (45)
The limiting values at r = 0 , k r=0,k r = 0 , k are understood by continuity. In particular ξ H F \xi_{\mathrm{HF}} ξ HF has the sign of r − k r-k r − k . For h = n − P k h=n-P_k h = n − P k , d = ∣ h ∣ d=\lvert h\rvert d = ∣ h ∣ , and s k = 1 − 2 P k s_k=1-2P_k s k = 1 − 2 P k we have h = s k d h=s_kd h = s k d and w n 2 = d − d 2 w_n^2=d-d^2 w n 2 = d − d 2 . Expanding ordinary exchange and using ∫ h d 3 p = 0 \int h\mathrm{d}^3p=0 ∫ h d 3 p = 0 gives the exact identity
E ( n ) − E ( P k ) = q ∫ e k d d 3 p − X ( h ) − X ( w n ) . (46) \mathcal{E}(n)-\mathcal{E}(P_k)=q\int e_kd\mathrm{d}^3p-\mathcal{X}(h)-\mathcal{X}(w_n). \tag*{(46)} E ( n ) − E ( P k ) = q ∫ e k d d 3 p − X ( h ) − X ( w n ) . (46)
For R = θ k R=\theta k R = θ k and Ω = { p : ∣ t ( ∣ p ∣ ) ∣ ≤ R } \Omega=\{p:\lvert t(\lvert p\rvert)\rvert\le R\} Ω = { p : ∣ t (∣ p ∣)∣ ≤ R } , choose a > 0 a>0 a > 0 and set
E a = e k 2 + k 2 a 2 1 Ω , d a = 1 2 ( 1 − e k E a ) , ∥ f ∥ E a 2 = ∫ E a ∣ f ∣ 2 d 3 p . (47) E_a=\sqrt{e_k^2+k^2a^2\mathbf{1}_\Omega},\qquad
d_a=\frac{1}{2}\left(1-\frac{e_k}{E_a}\right),\qquad
\lVert f\rVert_{E_a}^2=\int E_a\lvert f\rvert^2\mathrm{d}^3p. \tag*{(47)} E a = e k 2 + k 2 a 2 1 Ω , d a = 2 1 ( 1 − E a e k ) , ∥ f ∥ E a 2 = ∫ E a ∣ f ∣ 2 d 3 p . (47)
Thus 0 ≤ d a ≤ 1 / 2 0\le d_a\le1/2 0 ≤ d a ≤ 1/2 and d a = 0 d_a=0 d a = 0 outside the shell. This is a comparison defect, not a trial occupation; no particle-number constraint is imposed on d a d_a d a . Define
B a = 1 2 E a − 1 / 2 V E a − 1 / 2 , β ( a ) = ∥ B a ∥ , β ± ( a ) = ∥ Π ± B a Π ± ∥ , (48) \mathcal{B}_a=\frac{1}{2}E_a^{-1/2}VE_a^{-1/2},\qquad
\beta(a)=\lVert\mathcal{B}_a\rVert,\qquad
\beta_\pm(a)=\lVert\Pi_\pm\mathcal{B}_a\Pi_\pm\rVert, \tag*{(48)} B a = 2 1 E a − 1/2 V E a − 1/2 , β ( a ) = ∥ B a ∥ , β ± ( a ) = ∥ Π ± B a Π ± ∥ , (48)
where Π − = P k \Pi_-=P_k Π − = P k , Π + = 1 − P k \Pi_+=1-P_k Π + = 1 − P k are multiplication operators on L 2 ( R 3 , d 3 p ) L^2(\mathbb{R}^3,\mathrm{d}^3p) L 2 ( R 3 , d 3 p ) . The next section constructs a deterministic a a a for which the following hypotheses hold.
Proposition 6.1 (Coercive comparison). If β ( a ) ≤ 1 \beta(a)\le1 β ( a ) ≤ 1 and β ± ( a ) ≤ 3 / 5 \beta_\pm(a)\le3/5 β ± ( a ) ≤ 3/5 , then every n ∈ A ρ n\in\mathcal{A}_\rho n ∈ A ρ satisfies
E ( n ) − E ( P k ) ≥ − q ∥ d a ∥ E a 2 − 5 X ( d a ) + q 4 ∥ ∣ n − P k ∣ − d a ∥ E a 2 . (49) \mathcal{E}(n)-\mathcal{E}(P_k)\ge-q\lVert d_a\rVert_{E_a}^2-5\mathcal{X}(d_a)+\frac{q}{4}\lVert\lvert n-P_k\rvert-d_a\rVert_{E_a}^2. \tag*{(49)} E ( n ) − E ( P k ) ≥ − q ∥ d a ∥ E a 2 − 5 X ( d a ) + 4 q ∥∣ n − P k ∣ − d a ∥ E a 2 . (49) No symmetry or minimizing property of n n n is assumed.
Proof. The full norm bound gives X ( w n ) ≤ q ∫ E a w n 2 d 3 p \mathcal{X}(w_n) \le q\int E_a w_n^2\,\mathrm{d}^3p X ( w n ) ≤ q ∫ E a w n 2 d 3 p . Substitute w n 2 = d − d 2 w_n^2=d-d^2 w n 2 = d − d 2 into (46) and complete the square:
E ( n ) − E ( P k ) ≥ q ∥ d − d a ∥ E a 2 − q ∥ d a ∥ E a 2 − X ( h ) . (50) \mathcal{E}(n)-\mathcal{E}(P_k)\ge q\lVert d-d_a\rVert_{E_a}^2-q\lVert d_a\rVert_{E_a}^2-\mathcal{X}(h). \tag*{(50)} E ( n ) − E ( P k ) ≥ q ∥ d − d a ∥ E a 2 − q ∥ d a ∥ E a 2 − X ( h ) . (50)
We must still control the last negative term. If f ≤ 0 f\le0 f ≤ 0 inside the Fermi ball and f ≥ 0 f\ge0 f ≥ 0 outside, its mixed Coulomb contribution between the two regions is nonpositive. The compressed norms give
X ( f ) ≤ X ( Π − f ) + X ( Π + f ) ≤ 3 5 q ∥ f ∥ E a 2 . (51) \mathcal{X}(f)\le\mathcal{X}(\Pi_-f)+\mathcal{X}(\Pi_+f)\le\frac{3}{5}q\lVert f\rVert_{E_a}^2. \tag*{(51)} X ( f ) ≤ X ( Π − f ) + X ( Π + f ) ≤ 5 3 q ∥ f ∥ E a 2 . (51)
Apply this to the positive-part decomposition
h = h 1 + h 0 , h 1 = s k ( d − d a ) + , h 0 = s k min { d , d a } . h=h_1+h_0,\qquad h_1=s_k(d-d_a)_+,\qquad h_0=s_k\min\{d,d_a\}. h = h 1 + h 0 , h 1 = s k ( d − d a ) + , h 0 = s k min { d , d a } .
Here h 1 h_1 h 1 has the required sign and ∣ h 0 ∣ ≤ d a \lvert h_0\rvert\le d_a ∣ h 0 ∣ ≤ d a , so positivity of the kernel implies X ( h 0 ) ≤ X ( d a ) \mathcal{X}(h_0)\le\mathcal{X}(d_a) X ( h 0 ) ≤ X ( d a ) . Schwarz for the positive Coulomb form and 2 u v ≤ u 2 / 4 + 4 v 2 2uv\le u^2/4+4v^2 2 uv ≤ u 2 /4 + 4 v 2 imply
X ( h ) ≤ 5 4 X ( h 1 ) + 5 X ( h 0 ) ≤ 3 4 q ∥ ( d − d a ) + ∥ E a 2 + 5 X ( d a ) ≤ 3 4 q ∥ d − d a ∥ E a 2 + 5 X ( d a ) . \mathcal{X}(h)\le\frac{5}{4}\mathcal{X}(h_1)+5\mathcal{X}(h_0)
\le\frac{3}{4}q\lVert(d-d_a)_+\rVert_{E_a}^2+5\mathcal{X}(d_a)
\le\frac{3}{4}q\lVert d-d_a\rVert_{E_a}^2+5\mathcal{X}(d_a). X ( h ) ≤ 4 5 X ( h 1 ) + 5 X ( h 0 ) ≤ 4 3 q ∥( d − d a ) + ∥ E a 2 + 5 X ( d a ) ≤ 4 3 q ∥ d − d a ∥ E a 2 + 5 X ( d a ) .
Insert this into (50) . All expressions are finite: E a ( p ) ≤ C k , a ( 1 + ∣ p ∣ 2 ) E_a(p)\le C_{k,a}(1+\lvert p\rvert^2) E a ( p ) ≤ C k , a ( 1 + ∣ p ∣ 2 ) , the competitors have finite kinetic energy, and d a d_a d a is bounded with compact support. □ \square □
The full norm bound controls the pairing exchange, whereas the one-sided bounds control ordinary exchange without changing the leading source term q ∥ d a ∥ E a 2 q\lVert d_a\rVert_{E_a}^2 q ∥ d a ∥ E a 2 . The numerical constants 3 / 5 3/5 3/5 , 5 5 5 , and 1 / 4 1/4 1/4 only provide a fixed coercive margin; they will not enter the asymptotic energy coefficient.
The deterministic spectral threshold
The logarithmic energy space and model
Put L = log ( R / a ) L=\log(R/a) L = log ( R / a ) and ℓ = log ( 2 / θ ) \ell=\log(2/\theta) ℓ = log ( 2/ θ ) . The model weight is
W L , k ( x ) = [ ( 1 + x + ℓ π k ) 2 + e 2 ( x − L ) ] − 1 / 2 . (52) W_{L,k}(x)=\left[\left(1+\frac{x+\ell}{\pi k}\right)^2+e^{2(x-L)}\right]^{-1/2}. \tag*{(52)} W L , k ( x ) = [ ( 1 + π k x + ℓ ) 2 + e 2 ( x − L ) ] − 1/2 . (52)
For ℓ > 0 \ell>0 ℓ > 0 let
H ℓ = { u ∈ A C l o c ( [ 0 , ∞ ) ) : u ′ ∈ L 2 ( 0 , ∞ ) } , \mathcal{H}_{\ell}=\{u\in AC_{\mathrm{loc}}([0,\infty)):u'\in L^2(0,\infty)\}, H ℓ = { u ∈ A C loc ([ 0 , ∞ )) : u ′ ∈ L 2 ( 0 , ∞ )} ,
with inner product and squared norm
a ℓ [ u , v ] = ∫ 0 ∞ u ′ v ′ ‾ d x + ℓ − 1 u ( 0 ) v ( 0 ) ‾ , a ℓ [ u ] = a ℓ [ u , u ] . a_{\ell}[u,v]=\int_0^\infty u'\overline{v'}\,\mathrm{d}x+\ell^{-1}u(0)\overline{v(0)},\qquad a_{\ell}[u]=a_{\ell}[u,u]. a ℓ [ u , v ] = ∫ 0 ∞ u ′ v ′ d x + ℓ − 1 u ( 0 ) v ( 0 ) , a ℓ [ u ] = a ℓ [ u , u ] .
The map u ↦ ( ℓ − 1 / 2 u ( 0 ) , u ′ ) u\mapsto(\ell^{-1/2}u(0),u') u ↦ ( ℓ − 1/2 u ( 0 ) , u ′ ) is an isometric isomorphism onto C ⊕ L 2 ( 0 , ∞ ) \mathbb{C}\oplus L^2(0,\infty) C ⊕ L 2 ( 0 , ∞ ) . No L 2 L^2 L 2 condition is imposed on u u u itself. The reproducing kernel is G ℓ ( x , y ) = ℓ + min ( x , y ) G_{\ell}(x,y)=\ell+\min(x,y) G ℓ ( x , y ) = ℓ + min ( x , y ) , since a ℓ [ u , G ℓ ( ⋅ , y ) ] = u ( y ) a_{\ell}[u,G_{\ell}(\cdot,y)]=u(y) a ℓ [ u , G ℓ ( ⋅ , y )] = u ( y ) . In particular,
∣ u ( x ) ∣ 2 ≤ ( x + ℓ ) a ℓ [ u ] . (53) \lvert u(x)\rvert^2\le(x+\ell)a_{\ell}[u]. \tag*{(53)} ∣ u ( x ) ∣ 2 ≤ ( x + ℓ ) a ℓ [ u ] . (53)
For W ≥ 0 W\ge0 W ≥ 0 with ∫ 0 ∞ ( x + ℓ ) W ( x ) d x < ∞ \int_0^\infty(x+\ell)W(x)\,\mathrm{d}x<\infty ∫ 0 ∞ ( x + ℓ ) W ( x ) d x < ∞ , put
Λ ℓ ( W ) = sup 0 ≠ u ∈ H ℓ ∫ 0 ∞ W ∣ u ∣ 2 d x a ℓ [ u ] . (54) \Lambda_{\ell}(W)=\sup_{0\ne u\in\mathcal{H}_{\ell}}\frac{\int_0^\infty W\lvert u\rvert^2\,\mathrm{d}x}{a_{\ell}[u]}. \tag*{(54)} Λ ℓ ( W ) = 0 = u ∈ H ℓ sup a ℓ [ u ] ∫ 0 ∞ W ∣ u ∣ 2 d x . (54)
The associated compact positive operator S W S_W S W on H ℓ \mathcal{H}_{\ell} H ℓ is defined by a ℓ [ S W u , v ] = ∫ W u v ‾ a_{\ell}[S_Wu,v]=\int W u\overline{v} a ℓ [ S W u , v ] = ∫ W u v . It has norm Λ ℓ ( W ) \Lambda_{\ell}(W) Λ ℓ ( W ) ; (53) also gives
∥ S W 1 − S W 2 ∥ ≤ ∫ 0 ∞ ( x + ℓ ) ∣ W 1 − W 2 ∣ d x . (55) \lVert S_{W_1}-S_{W_2}\rVert\le\int_0^\infty(x+\ell)\lvert W_1-W_2\rvert\,\mathrm{d}x. \tag*{(55)} ∥ S W 1 − S W 2 ∥ ≤ ∫ 0 ∞ ( x + ℓ ) ∣ W 1 − W 2 ∣ d x . (55)
Compactness follows, for example, because the embedding u ↦ W u u\mapsto\sqrt{W}u u ↦ W u is Hilbert–Schmidt, with squared norm ∫ ( x + ℓ ) W \int(x+\ell)W ∫ ( x + ℓ ) W .
Proposition 7.1 (Model eigenvalue). Fix ℓ > 0 \ell> 0 ℓ > 0 and constants 0 < c 0 ≤ C 0 < ∞ 0 < c_0 \le C_0 < \infty 0 < c 0 ≤ C 0 < ∞ . As L → ∞ L \to\infty L → ∞ , uniformly for k > 0 k > 0 k > 0 satisfying c 0 ≤ L 2 / k ≤ C 0 c_0 \le L^2/k \le C_0 c 0 ≤ L 2 / k ≤ C 0 , we have
Λ ℓ ( W L , k ) = 4 L 2 π 2 + 8 ( ℓ + log 2 ) L π 2 − 4 L 3 π 3 k ( 1 2 + 2 π 2 ) + O ℓ , c 0 , C 0 ( 1 ) . (56) \Lambda_\ell(W_{L,k}) = \frac{4L^2}{\pi^2} + \frac{8(\ell+\log2)L}{\pi^2} - \frac{4L^3}{\pi^3 k}\left(\frac{1}{2}+\frac{2}{\pi^2}\right) + O_{\ell,c_0,C_0}(1). \tag*{(56)} Λ ℓ ( W L , k ) = π 2 4 L 2 + π 2 8 ( ℓ + log 2 ) L − π 3 k 4 L 3 ( 2 1 + π 2 2 ) + O ℓ , c 0 , C 0 ( 1 ) . (56)
Proof. In this proof, constants may also depend on c 0 , C 0 c_0,C_0 c 0 , C 0 ; this dependence is suppressed in the notation. For W L 0 = 1 [ 0 , L ] W_L^0=\mathbf{1}_{[0,L]} W L 0 = 1 [ 0 , L ] , the first eigenvalue is λ 0 = ω 0 − 2 \lambda_0=\omega_0^{-2} λ 0 = ω 0 − 2 , where cot ( ω 0 L ) = ℓ ω 0 \cot(\omega_0L)=\ell\omega_0 cot ( ω 0 L ) = ℓ ω 0 . Thus
λ 0 = 4 π 2 ( L + ℓ ) 2 + O ℓ ( 1 ) , (57) \lambda_0=\frac{4}{\pi^2}(L+\ell)^2+O_\ell(1), \tag*{(57)} λ 0 = π 2 4 ( L + ℓ ) 2 + O ℓ ( 1 ) , (57)
and the gap to the next eigenvalue is of order L 2 L^2 L 2 . The eigenfunction normalized by a ℓ [ u 0 ] = 1 a_\ell[u_0]=1 a ℓ [ u 0 ] = 1 is c L cos ( ω 0 ( L − x ) ) c_L\cos(\omega_0(L-x)) c L cos ( ω 0 ( L − x )) on [ 0 , L ] [0,L] [ 0 , L ] , continued constantly after L L L . Integration by parts gives
1 = ω 0 2 ∫ 0 L u 0 2 d x , c L 2 = 2 ω 0 2 ( L + ℓ / ( 1 + ℓ 2 ω 0 2 ) ) . 1=\omega_0^2\int_0^L u_0^2\,dx,\qquad c_L^2=\frac{2}{\omega_0^2\left(L+\ell/(1+\ell^2\omega_0^2)\right)}. 1 = ω 0 2 ∫ 0 L u 0 2 d x , c L 2 = ω 0 2 ( L + ℓ / ( 1 + ℓ 2 ω 0 2 ) ) 2 .
Since ω 0 = π / ( 2 L ) + O ℓ ( L − 2 ) \omega_0=\pi/(2L)+O_\ell(L^{-2}) ω 0 = π / ( 2 L ) + O ℓ ( L − 2 ) and c L 2 = 8 L / π 2 + O ℓ ( 1 ) c_L^2=8L/\pi^2+O_\ell(1) c L 2 = 8 L / π 2 + O ℓ ( 1 ) , uniformly on [ 0 , L ] [0,L] [ 0 , L ] ,
sup 0 ≤ x ≤ L ∣ u 0 ( x ) 2 − 8 L π 2 sin 2 π x 2 L ∣ ≤ C ℓ . (58) \sup_{0\le x\le L}\left|u_0(x)^2-\frac{8L}{\pi^2}\sin^2\frac{\pi x}{2L}\right|\le C_\ell. \tag*{(58)} 0 ≤ x ≤ L sup u 0 ( x ) 2 − π 2 8 L sin 2 2 L π x ≤ C ℓ . (58)
Indeed the phases differ by O ℓ ( L − 1 ) O_\ell(L^{-1}) O ℓ ( L − 1 ) , and the squared cosine is Lipschitz. Also u 0 ( L ) 2 = 8 L / π 2 + O ℓ ( 1 ) u_0(L)^2=8L/\pi^2+O_\ell(1) u 0 ( L ) 2 = 8 L / π 2 + O ℓ ( 1 ) and ∣ ( u 0 2 ) ′ ∣ ≤ C ℓ \left|(u_0^2)'\right|\le C_\ell ( u 0 2 ) ′ ≤ C ℓ .
Let g ( z ) = ( 1 + e 2 z ) − 1 / 2 g(z)=(1+e^{2z})^{-1/2} g ( z ) = ( 1 + e 2 z ) − 1/2 . Taylor expansion in ( x + ℓ ) / ( π k ) (x+\ell)/(\pi k) ( x + ℓ ) / ( π k ) gives
W L , k = W L 0 + [ g ( x − L ) − W L 0 ] − x + ℓ π k W L 0 + r L , k , W_{L,k}=W_L^0+[g(x-L)-W_L^0]-\frac{x+\ell}{\pi k}W_L^0+r_{L,k}, W L , k = W L 0 + [ g ( x − L ) − W L 0 ] − π k x + ℓ W L 0 + r L , k ,
where
∫ 0 ∞ ( x + ℓ ) ∣ r L , k ( x ) ∣ d x ≤ C ℓ ( 1 + L 4 k 2 + 1 + L 2 k ) = O ℓ ( 1 ) . (59) \int_0^\infty(x+\ell)|r_{L,k}(x)|\,dx\le C_\ell\left(\frac{1+L^4}{k^2}+\frac{1+L^2}{k}\right)=O_\ell(1). \tag*{(59)} ∫ 0 ∞ ( x + ℓ ) ∣ r L , k ( x ) ∣ d x ≤ C ℓ ( k 2 1 + L 4 + k 1 + L 2 ) = O ℓ ( 1 ) . (59)
To see this, the quadratic Taylor remainder is bounded by C ( ( x + ℓ ) / k ) 2 ( 1 + e 2 ( x − L ) ) − 3 / 2 C((x+\ell)/k)^2(1+e^{2(x-L)})^{-3/2} C (( x + ℓ ) / k ) 2 ( 1 + e 2 ( x − L ) ) − 3/2 . In the linear term replace ( 1 + e 2 ( x − L ) ) − 3 / 2 (1+e^{2(x-L)})^{-3/2} ( 1 + e 2 ( x − L ) ) − 3/2 by W L 0 W_L^0 W L 0 ; the difference has finite absolute moments about x = L x=L x = L , and contributes O ℓ ( ( 1 + L 2 ) / k ) O_\ell((1+L^2)/k) O ℓ (( 1 + L 2 ) / k ) .
The total perturbation norm in (7.4) is O ( L ) O(L) O ( L ) . The spectral gap therefore makes the remainder after first-order eigenvalue perturbation O ( L 2 / L 2 ) = O ( 1 ) O(L^2/L^2)=O(1) O ( L 2 / L 2 ) = O ( 1 ) . The transition contributes 8 ( log 2 ) L / π 2 + O ℓ ( 1 ) 8(\log2)L/\pi^2+O_\ell(1) 8 ( log 2 ) L / π 2 + O ℓ ( 1 ) , because
∫ R [ g ( z ) − 1 { z < 0 } ] d z = log 2 (60) \int_{\mathbb{R}}[g(z)-\mathbf{1}_{\{z<0\}}]\,dz=\log2 \tag*{(60)} ∫ R [ g ( z ) − 1 { z < 0 } ] d z = log 2 (60)
and its first absolute moment is finite. The bound on ( u 0 2 ) ′ (u_0^2)' ( u 0 2 ) ′ justifies replacing u 0 ( L + z ) 2 u_0(L+z)^2 u 0 ( L + z ) 2 by u 0 ( L ) 2 u_0(L)^2 u 0 ( L ) 2 at a cost O ( 1 ) O(1) O ( 1 ) . Finally, (7.7) gives
− 1 π k ∫ 0 L ( x + ℓ ) u 0 ( x ) 2 d x = − 4 L 3 π 3 k ( 1 2 + 2 π 2 ) + O ℓ ( 1 ) , -\frac{1}{\pi k}\int_0^L(x+\ell)u_0(x)^2\,dx=-\frac{4L^3}{\pi^3k}\left(\frac{1}{2}+\frac{2}{\pi^2}\right)+O_\ell(1), − π k 1 ∫ 0 L ( x + ℓ ) u 0 ( x ) 2 d x = − π 3 k 4 L 3 ( 2 1 + π 2 2 ) + O ℓ ( 1 ) ,
using
∫ 0 L x sin 2 π x 2 L d x = L 2 ( 1 4 + 1 π 2 ) . (61) \int_0^L x\sin^2\frac{\pi x}{2L}\,dx=L^2\left(\frac{1}{4}+\frac{1}{\pi^2}\right). \tag*{(61)} ∫ 0 L x sin 2 2 L π x d x = L 2 ( 4 1 + π 2 1 ) . (61)
The uniform O ℓ ( 1 ) O_\ell(1) O ℓ ( 1 ) error in (7.7) contributes only O ℓ ( L 2 / k ) = O ℓ ( 1 ) O_\ell(L^2/k)=O_\ell(1) O ℓ ( L 2 / k ) = O ℓ ( 1 ) . Adding the Robin term in (7.6) proves (7.5). □ \square □
Full and one-sided Coulomb norms
For the radial weight E a E_a E a , spherical-harmonic decomposition and ∣ P j ( z ) ∣ ≤ 1 |P_j(z)| \le1 ∣ P j ( z ) ∣ ≤ 1 show that the full norm and both compressed norms in (48) equal their norms in the nonnegative radial channel. The radial kernel on L 2 ( R + , d r ) L^2(\mathbb{R}_+,dr) L 2 ( R + , d r ) is
B a ( r , s ) = 1 2 π log ( ( r + s ) / ∣ r − s ∣ ) E a ( r ) E a ( s ) . (62) \mathcal{B}_a(r,s)=\frac{1}{2\pi}\frac{\log\left((r+s)/|r-s|\right)}{\sqrt{E_a(r)E_a(s)}}. \tag*{(62)} B a ( r , s ) = 2 π 1 E a ( r ) E a ( s ) log ( ( r + s ) /∣ r − s ∣ ) . (62)
It is Hilbert–Schmidt: the diagonal logarithm is locally square integrable, and E a ( r ) ≍ r 2 E_a(r)\asymp r^2 E a ( r ) ≍ r 2 at infinity for fixed k k k , a a a . Angular domination gives the corresponding bounded full-space forms. The degenerate-dispersion Coulomb analysis is related to the approach of Gontier, Hainzl, and Lewin [13 ] , whose variational problem allows translation-symmetry breaking in Hartree–Fock theory; no such enlargement of the present class is made here.
Lemma 7.2 (Deterministic spectral comparison). Fix 0 < θ < 1 / 12 0<\theta<1/12 0 < θ < 1/12 and constants 0 < c 0 ≤ C 0 < ∞ 0<c_0\le C_0<\infty 0 < c 0 ≤ C 0 < ∞ . As L = log ( R / a ) → ∞ L=\log(R/a)\to\infty L = log ( R / a ) → ∞ , uniformly for k > 0 k>0 k > 0 satisfying c 0 ≤ L 2 / k ≤ C 0 c_0\le L^2/k\le C_0 c 0 ≤ L 2 / k ≤ C 0 ,
π k β ( a ) = Λ ℓ ( W L , k ) + O θ , c 0 , C 0 ( 1 + L ) , (63) \pi k\beta(a)=\Lambda_\ell(W_{L,k})+O_{\theta,c_0,C_0}(1+\sqrt{L}), \tag*{(63)} π k β ( a ) = Λ ℓ ( W L , k ) + O θ , c 0 , C 0 ( 1 + L ) , (63) 2 π k β ± ( a ) = Λ ℓ ( W L , k ) + O θ , c 0 , C 0 ( 1 + L ) . (64) 2\pi k\beta_\pm(a)=\Lambda_\ell(W_{L,k})+O_{\theta,c_0,C_0}(1+\sqrt{L}). \tag*{(64)} 2 π k β ± ( a ) = Λ ℓ ( W L , k ) + O θ , c 0 , C 0 ( 1 + L ) . (64)
Proof. In this proof, dependence of constants on c 0 c_0 c 0 , C 0 C_0 C 0 is again suppressed. On the two sides of the shell use t σ ( x ) = σ R e − x t_\sigma(x)=\sigma R e^{-x} t σ ( x ) = σ R e − x , r σ ( x ) = r ( t σ ( x ) ) r_\sigma(x)=r(t_\sigma(x)) r σ ( x ) = r ( t σ ( x )) , x ≥ 0 x\ge0 x ≥ 0 , and set
ω a , σ ( x ) = k R e − x E a ( r σ ( x ) ) . \omega_{a,\sigma}(x)=\frac{kR e^{-x}}{E_a(r_\sigma(x))}. ω a , σ ( x ) = E a ( r σ ( x )) k R e − x .
The exact Jacobian is ∣ d r σ / d x ∣ = k 2 R e − x / r σ 2 |dr_\sigma/dx|=k^2R e^{-x}/r_\sigma^2 ∣ d r σ / d x ∣ = k 2 R e − x / r σ 2 . The shell block of 2 π k B a 2\pi k\mathcal{B}_a 2 π k B a consequently has matrix kernel
ω a , σ ( x ) ω a , τ ( y ) k 2 r σ ( x ) r τ ( y ) log r σ ( x ) + r τ ( y ) ∣ r σ ( x ) − r τ ( y ) ∣ . \sqrt{\omega_{a,\sigma}(x)\omega_{a,\tau}(y)}\frac{k^2}{r_\sigma(x)r_\tau(y)}\log\frac{r_\sigma(x)+r_\tau(y)}{|r_\sigma(x)-r_\tau(y)|}. ω a , σ ( x ) ω a , τ ( y ) r σ ( x ) r τ ( y ) k 2 log ∣ r σ ( x ) − r τ ( y ) ∣ r σ ( x ) + r τ ( y ) .
The cubic-coordinate identity gives
log r σ ( x ) + r τ ( y ) ∣ r σ ( x ) − r τ ( y ) ∣ = min ( x , y ) + ℓ − log ∣ 1 − σ τ e − ∣ x − y ∣ ∣ + G σ τ ( x , y ) , (65) \log\frac{r_\sigma(x)+r_\tau(y)}{|r_\sigma(x)-r_\tau(y)|}
=\min(x,y)+\ell-\log\left|1-\sigma\tau e^{-|x-y|}\right|+G_{\sigma\tau}(x,y), \tag*{(65)} log ∣ r σ ( x ) − r τ ( y ) ∣ r σ ( x ) + r τ ( y ) = min ( x , y ) + ℓ − log 1 − σ τ e − ∣ x − y ∣ + G σ τ ( x , y ) , (65)
∣ G σ τ ( x , y ) ∣ + ∣ k 2 r σ ( x ) r τ ( y ) − 1 ∣ ≤ C θ ( e − x + e − y ) . (66) |G_{\sigma\tau}(x,y)|+\left|\frac{k^2}{r_\sigma(x)r_\tau(y)}-1\right|\le C_\theta(e^{-x}+e^{-y}). \tag*{(66)} ∣ G σ τ ( x , y ) ∣ + r σ ( x ) r τ ( y ) k 2 − 1 ≤ C θ ( e − x + e − y ) . (66)
The explicit dispersion (44) implies
e k ( r σ ( x ) ) k R e − x = 1 + x + ℓ π k + ϵ σ ( x ) , ∣ ϵ σ ( x ) ∣ ≤ C θ e − x ( 1 + x + ℓ k ) . \frac{e_k(r_\sigma(x))}{kR e^{-x}}=1+\frac{x+\ell}{\pi k}+\epsilon_\sigma(x),\qquad|\epsilon_\sigma(x)|\le C_\theta e^{-x}\left(1+\frac{x+\ell}{k}\right). k R e − x e k ( r σ ( x )) = 1 + π k x + ℓ + ϵ σ ( x ) , ∣ ϵ σ ( x ) ∣ ≤ C θ e − x ( 1 + k x + ℓ ) .
The exact and model dispersion factors are positive and uniformly comparable. With b = e x − L b=e^{x-L} b = e x − L , the derivative of F ( A ) = ( A 2 + b 2 ) − 1 / 2 F(A)=(A^2+b^2)^{-1/2} F ( A ) = ( A 2 + b 2 ) − 1/2 satisfies ∣ F ′ ( A ) ∣ ≤ C θ / ( 1 + b 2 ) |F'(A)|\le C_\theta/(1+b^2) ∣ F ′ ( A ) ∣ ≤ C θ / ( 1 + b 2 ) along their connecting segment. Hence
∫ 0 ∞ ( x + ℓ ) ∣ ω a , σ − W L , k ∣ d x ≤ C θ . (67) \int_0^\infty(x+\ell)|\omega_{a,\sigma}-W_{L,k}|\,dx\le C_\theta. \tag*{(67)} ∫ 0 ∞ ( x + ℓ ) ∣ ω a , σ − W L , k ∣ d x ≤ C θ . (67)
Also
ω a , σ ≤ C θ min { 1 , e L − x } , ∫ 0 ∞ ω a , σ = O θ ( 1 + L ) , ∫ 0 ∞ e − 2 x ( x + ℓ ) 2 ω a , σ d x ≤ C θ . (68) \omega_{a,\sigma}\le C_\theta\min\{1,e^{L-x}\},\qquad\int_0^\infty\omega_{a,\sigma}=O_\theta(1+L),\qquad\int_0^\infty e^{-2x}(x+\ell)^2\omega_{a,\sigma}\,dx\le C_\theta. \tag*{(68)} ω a , σ ≤ C θ min { 1 , e L − x } , ∫ 0 ∞ ω a , σ = O θ ( 1 + L ) , ∫ 0 ∞ e − 2 x ( x + ℓ ) 2 ω a , σ d x ≤ C θ . (68)
The functions − log ∣ 1 ∓ e − ∣ z ∣ ∣ -\log|1\mp e^{-|z|}| − log ∣1 ∓ e − ∣ z ∣ ∣ belong to L 1 ( R ) L^1(\mathbb{R}) L 1 ( R ) . Their weighted convolution kernels have norm O θ ( 1 ) O_\theta(1) O θ ( 1 ) by Schur’s test. For the geometric terms use min ( x , y ) + ℓ ≤ x + ℓ \min(x,y)+\ell\le x+\ell min ( x , y ) + ℓ ≤ x + ℓ when accompanied by e − x e^{-x} e − x , and symmetrically for e − y e^{-y} e − y . Their Hilbert–Schmidt norms are O θ ( 1 + L ) O_\theta(1+\sqrt{L}) O θ ( 1 + L ) by (68) .
For the leading matrix define J : H ℓ → L 2 ( 0 , ∞ ) ⊕ L 2 ( 0 , ∞ ) J:\mathcal{H}_\ell\to L^2(0,\infty)\oplus L^2(0,\infty) J : H ℓ → L 2 ( 0 , ∞ ) ⊕ L 2 ( 0 , ∞ ) by J u = ( ω a , + u , ω a , − u ) Ju=(\sqrt{\omega_{a,+}}u,\sqrt{\omega_{a,-}}u) J u = ( ω a , + u , ω a , − u ) . It is Hilbert–Schmidt, with squared norm ∑ σ ∫ ( x + ℓ ) ω a , σ \sum_\sigma\int(x+\ell)\omega_{a,\sigma} ∑ σ ∫ ( x + ℓ ) ω a , σ . The reproducing identity on H ℓ \mathcal{H}_\ell H ℓ shows that J J ∗ JJ^* J J ∗ has exactly the leading matrix kernel. Its norm is 2 Λ ℓ ( ( ω a , + + ω a , − ) / 2 ) 2\Lambda_\ell((\omega_{a,+}+\omega_{a,-})/2) 2 Λ ℓ (( ω a , + + ω a , − ) /2 ) ; for a single side it is Λ ℓ ( ω a , σ ) \Lambda_\ell(\omega_{a,\sigma}) Λ ℓ ( ω a , σ ) . The adjoints here use a ℓ a_\ell a ℓ on the domain of J J J . Equations (55) and (67) therefore yield the required shell estimates, including the factor two.
Outside the shell, (45) implies E a ( r ) = e k ( r ) ≥ c θ ( k 2 + r 2 ) E_a(r)=e_k(r)\ge c_\theta(k^2+r^2) E a ( r ) = e k ( r ) ≥ c θ ( k 2 + r 2 ) . After r = k z r=kz r = k z , the exterior–exterior block of 2 π k B a 2\pi k\mathcal{B}_a 2 π k B a has Hilbert–Schmidt norm O θ ( 1 ) O_\theta(1) O θ ( 1 ) . For the mixed block use
sup z ∈ I θ ∫ I θ c log 2 ( ( z + z ′ ) / ∣ z − z ′ ∣ ) 1 + z ′ 2 d z ′ < ∞ , I θ = [ ( 1 − 3 θ ) 1 / 3 , ( 1 + 3 θ ) 1 / 3 ] . \sup_{z\in I_\theta}\int_{I_\theta^c}\frac{\log^2((z+z')/|z-z'|)}{1+z'^2}\,dz'<\infty,\qquad I_\theta=[(1-3\theta)^{1/3},(1+3\theta)^{1/3}]. z ∈ I θ sup ∫ I θ c 1 + z ′2 log 2 (( z + z ′ ) /∣ z − z ′ ∣ ) d z ′ < ∞ , I θ = [( 1 − 3 θ ) 1/3 , ( 1 + 3 θ ) 1/3 ] .
The logarithmic singularity at a shell endpoint is square integrable, and the integrand decays at infinity. Integrating the remaining shell factor with (68) gives norm O θ ( 1 + L ) O_\theta(1+\sqrt{L}) O θ ( 1 + L ) . The same estimates hold for the one-sided restrictions. Restoring the factor 1 / ( 2 π k ) 1/(2\pi k) 1/ ( 2 π k ) proves both assertions. □
An explicit subcritical comparison scale
Define, using (10) ,
a 0 ( k ) = y 0 k e − A k , a ∗ ( k ) = a 0 ( k ) e k − 1 / 8 , E ∗ = E a ∗ , d ∗ = d a ∗ . (69) a_0(k)=y_0ke^{-A\sqrt{k}},\qquad a_*(k)=a_0(k)e^{k^{-1/8}},\qquad E_*=E_{a_*},\qquad d_*=d_{a_*}. \tag*{(69)} a 0 ( k ) = y 0 k e − A k , a ∗ ( k ) = a 0 ( k ) e k − 1/8 , E ∗ = E a ∗ , d ∗ = d a ∗ . (69)
The vanishing safety margin makes the full norm strictly subcritical without changing the leading constant.
Corollary 7.3. For all sufficiently large k k k ,
β ( a ∗ ) ≤ 1 , β ± ( a ∗ ) ≤ 3 5 , β ± ( a ∗ ) ⟶ 1 2 . (70) \beta(a_*)\le1,\qquad\beta_\pm(a_*)\le\frac{3}{5},\qquad\beta_\pm(a_*)\longrightarrow\frac{1}{2}. \tag*{(70)} β ( a ∗ ) ≤ 1 , β ± ( a ∗ ) ≤ 5 3 , β ± ( a ∗ ) ⟶ 2 1 . (70) Proof. Write b 0 = log ( θ / y 0 ) = − ℓ − log 2 + 1 / 4 + π 2 / 16 b_0=\log(\theta/y_0)=-\ell-\log2+1/4+\pi^2/16 b 0 = log ( θ / y 0 ) = − ℓ − log 2 + 1/4 + π 2 /16 . Then L ∗ : = log ( R / a ∗ ) = A k + b 0 − k − 1 / 8 L_*:=\log(R/a_*)=A\sqrt{k}+b_0-k^{-1/8} L ∗ := log ( R / a ∗ ) = A k + b 0 − k − 1/8 . The coefficient b 0 b_0 b 0 cancels the order-k \sqrt{k} k terms in (56) . Equations (63) and (56) give
π k β ( a ∗ ) = π k − 8 A π 2 k 3 / 8 + O θ ( k 1 / 4 + 1 ) < π k \pi k\beta(a_*)=\pi k-\frac{8A}{\pi^2}k^{3/8}+O_\theta(k^{1/4}+1)<\pi k π k β ( a ∗ ) = π k − π 2 8 A k 3/8 + O θ ( k 1/4 + 1 ) < π k for large k k k , while (64) gives the limit 1 / 2 1/2 1/2 . □
The sharp CHF energy asymptotics
Evaluation of the deterministic upper bound
The comparison defect in (69) has compact shell support; there are no self-consistent exterior tails to estimate.
Lemma 8.1 (The explicit comparison source). For a ∗ a_* a ∗ in (69) ,
∥ d ∗ ∥ E ∗ 2 = k 3 a ∗ 2 8 π 2 ( 1 + o ( 1 ) ) , (71) \lVert d_*\rVert_{E_*}^2=\frac{k^3a_*^2}{8\pi^2}(1+o(1)), \tag*{(71)} ∥ d ∗ ∥ E ∗ 2 = 8 π 2 k 3 a ∗ 2 ( 1 + o ( 1 )) , (71) X ( d ∗ ) ≤ C θ q k 2 a ∗ 2 ( 1 + log k a ∗ ) = o ( q k 3 a ∗ 2 ) . (72) \mathcal{X}(d_*)\le C_\theta q_k^2a_*^2\left(1+\log\frac{k}{a_*}\right)=o(q_k^3a_*^2). \tag*{(72)} X ( d ∗ ) ≤ C θ q k 2 a ∗ 2 ( 1 + log a ∗ k ) = o ( q k 3 a ∗ 2 ) . (72) Proof. On the fixed shell, (45) implies e k ( r ( t ) ) ≥ c θ k ∣ t ∣ e_k(r(t))\ge c_\theta k|t| e k ( r ( t )) ≥ c θ k ∣ t ∣ . Consequently
0 ≤ d ∗ ( r ( t ) ) ≤ C θ a ∗ 2 a ∗ 2 + t 2 , ∫ − R R d ∗ ( r ( t ) ) d t ≤ C θ a ∗ . (73) 0\le d_*(r(t))\le C_\theta\frac{a_*^2}{a_*^2+t^2},\qquad\int_{-R}^{R}d_*(r(t))\,dt\le C_\theta a_*. \tag*{(73)} 0 ≤ d ∗ ( r ( t )) ≤ C θ a ∗ 2 + t 2 a ∗ 2 , ∫ − R R d ∗ ( r ( t )) d t ≤ C θ a ∗ . (73) For fixed u u u , (44) and log ( k / a ∗ ) = O ( k ) \log(k/a_*)=O(\sqrt{k}) log ( k / a ∗ ) = O ( k ) give
e k ( r ( a ∗ u ) ) k a ∗ ⟶ ∣ u ∣ , E ∗ ( r ( a ∗ u ) ) k a ∗ ⟶ 1 + u 2 , d ∗ ( r ( a ∗ u ) ) ⟶ d ∞ ( ∣ u ∣ ) , \frac{e_k(r(a_*u))}{ka_*}\longrightarrow|u|,\qquad\frac{E_*(r(a_*u))}{ka_*}\longrightarrow\sqrt{1+u^2},\qquad d_*(r(a_*u))\longrightarrow d_\infty(|u|), k a ∗ e k ( r ( a ∗ u )) ⟶ ∣ u ∣ , k a ∗ E ∗ ( r ( a ∗ u )) ⟶ 1 + u 2 , d ∗ ( r ( a ∗ u )) ⟶ d ∞ ( ∣ u ∣ ) , where
d ∞ ( v ) = 1 2 ( 1 − v 1 + v 2 ) , v ≥ 0. (74) d_\infty(v)=\frac{1}{2}\left(1-\frac{v}{\sqrt{1+v^2}}\right),\qquad v\ge0. \tag*{(74)} d ∞ ( v ) = 2 1 ( 1 − 1 + v 2 v ) , v ≥ 0. (74) The identity
E ∗ d ∗ 2 4 E ∗ ( E ∗ + e k ) 2 = k 4 a ∗ 4 4 E ∗ ( E ∗ + e k ) 2 on Ω \frac{E_*d_*^2}{4E_*(E_*+e_k)^2}=\frac{k^4a_*^4}{4E_*(E_*+e_k)^2}\quad\text{on }\Omega 4 E ∗ ( E ∗ + e k ) 2 E ∗ d ∗ 2 = 4 E ∗ ( E ∗ + e k ) 2 k 4 a ∗ 4 on Ω and the lower bound for e k e_k e k majorize the scaled energy integrand by C θ ( 1 + ∣ u ∣ ) − 3 C_\theta(1+|u|)^{-3} C θ ( 1 + ∣ u ∣ ) − 3 . Extend it by zero off the rescaled shell. Dominated convergence gives
∥ d ∗ ∥ E ∗ 2 = k 3 a ∗ 2 2 π 2 [ 2 ∫ 0 ∞ 1 + u 2 d ∞ ( u ) 2 d u + o ( 1 ) ] . \lVert d_*\rVert_{E_*}^2=\frac{k^3a_*^2}{2\pi^2}\left[2\int_0^\infty\sqrt{1+u^2}\,d_\infty(u)^2\,du+o(1)\right]. ∥ d ∗ ∥ E ∗ 2 = 2 π 2 k 3 a ∗ 2 [ 2 ∫ 0 ∞ 1 + u 2 d ∞ ( u ) 2 d u + o ( 1 ) ] .
The integral equals 1 / 8 1/8 1/8 : with u = sinh v u=\sinh v u = sinh v its integrand times d u du d u becomes e − 2 v d v / 4 e^{-2v}\,dv/4 e − 2 v d v /4 . This proves (71) . For exchange apply (18) and Lemma 3.3 to f ( t ) = d ∗ ( r ( t ) ) f(t)=d_*(r(t)) f ( t ) = d ∗ ( r ( t )) . Its mass is O ( a ∗ ) O(a_*) O ( a ∗ ) by (73) ; the increasing function m 2 ( 1 + log ( C θ k / m ) ) m^2(1+\log(C\theta k/m)) m 2 ( 1 + log ( C θ k / m )) gives (72) . The relative bound is O θ ( log ( k / a ∗ ) / k ) = o ( 1 ) O_\theta(\log(k/a_*)/k)=o(1) O θ ( log ( k / a ∗ ) / k ) = o ( 1 ) .
Take the infimum in Proposition 6.1, use Corollary 7.3, and drop the nonnegative remainder. Since ρ = q k 3 / ( 6 π 2 ) \rho=qk^3/(6\pi^2) ρ = q k 3 / ( 6 π 2 ) , Lemma 8.1 yields
D ( k ) ≤ q ∥ d ∗ ∥ E ∗ 2 + 5 X ( d ∗ ) ρ = 3 4 a ∗ 2 ( 1 + o ( 1 ) ) . (75) D(k)\le\frac{q\lVert d_*\rVert_{E_*}^2+5\mathcal{X}(d_*)}{\rho}=\frac{3}{4}a_*^2(1+o(1)). \tag*{(75)} D ( k ) ≤ ρ q ∥ d ∗ ∥ E ∗ 2 + 5 X ( d ∗ ) = 4 3 a ∗ 2 ( 1 + o ( 1 )) . (75)
Thus the upper limit in Theorem 2.2 is at most C ∗ C_* C ∗ . This argument did not use existence, radiality, or self-consistency of a minimizer.
One number-preserving trial state on the required scale
For the matching lower bound it suffices to evaluate one trial state: the scale is already selected by the deterministic spectrum. Set
a = a 0 ( k ) = y 0 k e − A k , R = θ k , L = log ( R / a ) = A k + b 0 , b 0 = log ( θ / y 0 ) . (76) a=a_0(k)=y_0ke^{-A\sqrt{k}},\qquad R=\theta k,\qquad L=\log(R/a)=A\sqrt{k}+b_0,\qquad b_0=\log(\theta/y_0). \tag*{(76)} a = a 0 ( k ) = y 0 k e − A k , R = θ k , L = log ( R / a ) = A k + b 0 , b 0 = log ( θ / y 0 ) . (76)
The parameter θ \theta θ is fixed, and the construction is used for sufficiently large k k k , so that L > 0 L>0 L > 0 . For 0 < v ≤ R 0<v\le R 0 < v ≤ R define
g L ( v ) = sin ( π 2 L min { log ( R / v ) , L } ) , w a ( v ) = a g L ( v ) 2 v 2 + a 2 g L ( v ) 2 , g_L(v)=\sin\left(\frac{\pi}{2L}\min\{\log(R/v),L\}\right),\qquad w_a(v)=\frac{ag_L(v)}{2\sqrt{v^2+a^2g_L(v)^2}}, g L ( v ) = sin ( 2 L π min { log ( R / v ) , L } ) , w a ( v ) = 2 v 2 + a 2 g L ( v ) 2 a g L ( v ) ,
with w a ( 0 ) = 1 / 2 w_a(0)=1/2 w a ( 0 ) = 1/2 and zero continuation for v > R v>R v > R . The actual momentum occupation is n k a ( p ) = ν a ( t ( ∣ p ∣ ) ) n_k^a(p)=\nu_a(t(|p|)) n k a ( p ) = ν a ( t ( ∣ p ∣ )) , where
ν a ( t ) = 1 2 ( 1 − sgn ( t ) 1 − 4 w a ( ∣ t ∣ ) 2 ) ( ∣ t ∣ ≤ R ) , (77) \nu_a(t)=\frac{1}{2}\left(1-\operatorname{sgn}(t)\sqrt{1-4w_a(|t|)^2}\right)\qquad(|t|\le R), \tag*{(77)} ν a ( t ) = 2 1 ( 1 − sgn ( t ) 1 − 4 w a ( ∣ t ∣ ) 2 ) ( ∣ t ∣ ≤ R ) , (77)
and ν a ( t ) = 1 { t < 0 } \nu_a(t)=\mathbf{1}_{\{t<0\}} ν a ( t ) = 1 { t < 0 } outside. The constant Jacobian (20) and ν a ( t ) + ν a ( − t ) = 1 \nu_a(t)+\nu_a(-t)=1 ν a ( t ) + ν a ( − t ) = 1 on [ − R , R ] [-R,R] [ − R , R ] give the particle number exactly. Write δ a ( t ) = ∣ ν a ( t ) − 1 { t < 0 } ∣ \delta_a(t)=|\nu_a(t)-\mathbf{1}_{\{t<0\}}| δ a ( t ) = ∣ ν a ( t ) − 1 { t < 0 } ∣ and introduce the reference tail
f L ( x ) = sin π x 2 L , w 0 ( v ) = a 2 v f L ( log ( R / v ) ) 1 { a ≤ v ≤ R } . f_L(x)=\sin\frac{\pi x}{2L},\qquad w_0(v)=\frac{a}{2v}f_L(\log(R/v))\mathbf{1}_{\{a\le v\le R\}}. f L ( x ) = sin 2 L π x , w 0 ( v ) = 2 v a f L ( log ( R / v )) 1 { a ≤ v ≤ R } .
Lemma 8.2 (Inner matching). For the preceding trial profile,
∫ − R R ( w a ( ∣ t ∣ ) − w 0 ( ∣ t ∣ ) ) d t = a log 2 + o ( a ) , (78) \int_{-R}^{R}\left(w_a(|t|)-w_0(|t|)\right)\,dt=a\log2+o(a), \tag*{(78)} ∫ − R R ( w a ( ∣ t ∣ ) − w 0 ( ∣ t ∣ ) ) d t = a log 2 + o ( a ) , (78) ∫ − R R ∣ t ∣ [ δ a ( t ) − w 0 ( ∣ t ∣ ) ] 2 d t = a 2 ( log 2 2 − 1 4 ) + o ( a 2 ) . (79) \int_{-R}^{R}|t|\left[\delta_a(t)-w_0(|t|)\right]^2\,dt=a^2\left(\frac{\log2}{2}-\frac{1}{4}\right)+o(a^2). \tag*{(79)} ∫ − R R ∣ t ∣ [ δ a ( t ) − w 0 ( ∣ t ∣ ) ] 2 d t = a 2 ( 2 log 2 − 4 1 ) + o ( a 2 ) . (79) The corresponding absolute integrals, even with the factor 1 + ∣ log ( ∣ t ∣ / a ) ∣ 1+|\log(|t|/a)| 1 + ∣ log ( ∣ t ∣/ a ) ∣ , are bounded by C a Ca C a and C a 2 Ca^2 C a 2 , respectively. Moreover,
∫ − R R t 2 δ a ( t ) ( 1 + log 2 k ∣ t ∣ ) d t ≤ C θ ( a 2 R L 2 + a 3 ( 1 + L ) ) . (80) \int_{-R}^{R}t^2\delta_a(t)\left(1+\log\frac{2k}{|t|}\right)\,dt\le C_\theta\left(\frac{a^2R}{L^2}+a^3(1+L)\right). \tag*{(80)} ∫ − R R t 2 δ a ( t ) ( 1 + log ∣ t ∣ 2 k ) d t ≤ C θ ( L 2 a 2 R + a 3 ( 1 + L ) ) . (80)
Proof. For a ≤ v ≤ R a\le v\le R a ≤ v ≤ R ,
∣ w a ( v ) − w 0 ( v ) ∣ ≤ C a 3 v 3 , ∣ δ a ( v ) − w 0 ( v ) ∣ 2 ≤ C a 4 v 4 . |w_a(v)-w_0(v)|\le\frac{Ca^3}{v^3},\qquad|\delta_a(v)-w_0(v)|^2\le\frac{Ca^4}{v^4}. ∣ w a ( v ) − w 0 ( v ) ∣ ≤ v 3 C a 3 , ∣ δ a ( v ) − w 0 ( v ) ∣ 2 ≤ v 4 C a 4 .
For v < a v<a v < a all profiles are bounded and w 0 = 0 w_0=0 w 0 = 0 . After t = a u t=au t = a u , these are integrable majorants, including the stated logarithmic moments. For each fixed u > 0 u>0 u > 0 , 2 w a ( a u ) → ( 1 + u 2 ) − 1 / 2 2w_a(au)\to(1+u^2)^{-1/2} 2 w a ( a u ) → ( 1 + u 2 ) − 1/2 and δ a ( a u ) → d ∞ ( u ) \delta_a(au)\to d_\infty(u) δ a ( a u ) → d ∞ ( u ) . Hence dominated convergence and
∫ 0 ∞ [ 1 1 + u 2 − 1 { u > 1 } u ] d u = log 2 , (81) \int_0^\infty\left[\frac{1}{\sqrt{1+u^2}}-\frac{\mathbf{1}_{\{u>1\}}}{u}\right]\,du=\log2, \tag*{(81)} ∫ 0 ∞ [ 1 + u 2 1 − u 1 { u > 1 } ] d u = log 2 , (81)
2 ∫ 0 ∞ u [ d ∞ ( u ) − 1 { u > 1 } 4 u 2 ] d u = log 2 2 − 1 4 (82) 2\int_0^\infty u\left[d_\infty(u)-\frac{\mathbf{1}_{\{u>1\}}}{4u^2}\right]\,du=\frac{\log2}{2}-\frac{1}{4} \tag*{(82)} 2 ∫ 0 ∞ u [ d ∞ ( u ) − 4 u 2 1 { u > 1 } ] d u = 2 log 2 − 4 1 (82)
prove the matching formulas. The antiderivatives use arsinh u \operatorname{arsinh} u arsinh u and [ u 2 − u 1 + u 2 + arsinh u ] / 2 \left[u^2-u\sqrt{1+u^2}+\operatorname{arsinh} u\right]/2 [ u 2 − u 1 + u 2 + arsinh u ] /2 . For (8.10), split at ∣ t ∣ = a |t|=a ∣ t ∣ = a . On a ≤ ∣ t ∣ ≤ R a\leq|t|\leq R a ≤ ∣ t ∣ ≤ R , use δ a ( t ) ≤ 2 w a ( ∣ t ∣ ) 2 \delta_a(t)\leq2w_a(|t|)^2 δ a ( t ) ≤ 2 w a ( ∣ t ∣ ) 2 , x = log ( R / ∣ t ∣ ) x=\log(R/|t|) x = log ( R /∣ t ∣ ) , and f L ( x ) ≤ C x / L f_L(x)\leq Cx/L f L ( x ) ≤ C x / L to obtain
∫ a ≤ ∣ t ∣ ≤ R t 2 δ a ( t ) ( 1 + log 2 k ∣ t ∣ ) d t ≤ C a 2 R L 2 ∫ 0 L e − x ( 1 + ℓ + x ) x 2 d x ≤ C θ a 2 R L 2 . \int_{a\leq|t|\leq R}t^2\delta_a(t)\left(1+\log\frac{2k}{|t|}\right)\,\mathrm{d}t
\leq\frac{Ca^2R}{L^2}\int_0^L e^{-x}(1+\ell+x)x^2\,\mathrm{d}x\leq C_\theta\frac{a^2R}{L^2}. ∫ a ≤ ∣ t ∣ ≤ R t 2 δ a ( t ) ( 1 + log ∣ t ∣ 2 k ) d t ≤ L 2 C a 2 R ∫ 0 L e − x ( 1 + ℓ + x ) x 2 d x ≤ C θ L 2 a 2 R .
On ∣ t ∣ < a |t|<a ∣ t ∣ < a , use δ a ( t ) ≤ 1 / 2 \delta_a(t)\leq1/2 δ a ( t ) ≤ 1/2 and log ( 2 k / a ) = ℓ + L \log(2k/a)=\ell+L log ( 2 k / a ) = ℓ + L :
∫ ∣ t ∣ < a t 2 δ a ( t ) ( 1 + log 2 k ∣ t ∣ ) d t ≤ ∫ 0 a t 2 ( 1 + log 2 k t ) d t = a 3 ( 1 + ℓ + L 3 + 1 9 ) ≤ C θ a 3 ( 1 + L ) . \int_{|t|<a}t^2\delta_a(t)\left(1+\log\frac{2k}{|t|}\right)\,\mathrm{d}t
\leq\int_0^a t^2\left(1+\log\frac{2k}{t}\right)\,\mathrm{d}t
=a^3\left(\frac{1+\ell+L}{3}+\frac{1}{9}\right)\leq C_\theta a^3(1+L). ∫ ∣ t ∣ < a t 2 δ a ( t ) ( 1 + log ∣ t ∣ 2 k ) d t ≤ ∫ 0 a t 2 ( 1 + log t 2 k ) d t = a 3 ( 3 1 + ℓ + L + 9 1 ) ≤ C θ a 3 ( 1 + L ) .
Lemma 8.3 (Pairing integral). For the trial profile,
∬ [ − R , R ] 2 w a ( ∣ t ∣ ) w a ( ∣ s ∣ ) log 2 k ∣ t − s ∣ d t d s = a 2 [ 2 L 3 π 2 + 4 L 2 π 2 ( ℓ + 2 log 2 ) + o ( L 2 ) ] . (83) \iint_{[-R,R]^2}w_a(|t|)w_a(|s|)\log\frac{2k}{|t-s|}\,\mathrm{d}t\,\mathrm{d}s
=a^2\left[\frac{2L^3}{\pi^2}+\frac{4L^2}{\pi^2}(\ell+2\log2)+o(L^2)\right]. \tag*{(83)} ∬ [ − R , R ] 2 w a ( ∣ t ∣ ) w a ( ∣ s ∣ ) log ∣ t − s ∣ 2 k d t d s = a 2 [ π 2 2 L 3 + π 2 4 L 2 ( ℓ + 2 log 2 ) + o ( L 2 ) ] . (83) Replacing the logarithm by H k ( t , s ) H_k(t,s) H k ( t , s ) changes this expression by o ( a 2 L 2 ) o(a^2L^2) o ( a 2 L 2 ) .
Proof. Write w a = w 0 + δ w w_a=w_0+\delta w w a = w 0 + δ w . On each sign of t t t set t = σ R e − x t=\sigma R e^{-x} t = σ R e − x , 0 ≤ x ≤ L 0\leq x\leq L 0 ≤ x ≤ L , so that w 0 ( ∣ t ∣ ) ∣ d t ∣ = ( a / 2 ) f L ( x ) d x w_0(|t|)|\mathrm{d}t|=(a/2)f_L(x)\,\mathrm{d}x w 0 ( ∣ t ∣ ) ∣ d t ∣ = ( a /2 ) f L ( x ) d x . For s = τ R e − z s=\tau R e^{-z} s = τ R e − z ,
log 2 k ∣ t − s ∣ = min ( x , z ) + ℓ − log ∣ 1 − σ τ e − ∣ x − z ∣ ∣ . \log\frac{2k}{|t-s|}=\min(x,z)+\ell-\log\left|1-\sigma\tau e^{-|x-z|}\right|. log ∣ t − s ∣ 2 k = min ( x , z ) + ℓ − log 1 − σ τ e − ∣ x − z ∣ .
The last convolution kernel is in L 1 ( R ) L^1(\mathbb{R}) L 1 ( R ) and contributes O ( a 2 L ) O(a^2L) O ( a 2 L ) . The two remaining tail–tail terms are
a 2 [ 2 L 3 π 2 + 4 ℓ L 2 π 2 ] , a^2\left[\frac{2L^3}{\pi^2}+\frac{4\ell L^2}{\pi^2}\right], a 2 [ π 2 2 L 3 + π 2 4 ℓ L 2 ] ,
because ∫ 0 L f L = 2 L / π \int_0^L f_L=2L/\pi ∫ 0 L f L = 2 L / π and ∬ f L ( x ) min ( x , z ) f L ( z ) d x d z = 2 L 3 / π 2 \iint f_L(x)\min(x,z)f_L(z)\,\mathrm{d}x\,\mathrm{d}z=2L^3/\pi^2 ∬ f L ( x ) min ( x , z ) f L ( z ) d x d z = 2 L 3 / π 2 . For fixed u u u , the potential of w 0 w_0 w 0 at t = a u t=au t = a u , divided by a L 2 aL^2 a L 2 , converges to 4 / π 2 4/\pi^2 4/ π 2 , using ∫ 0 L x f L ( x ) d x = 4 L 2 / π 2 \int_0^L xf_L(x)\,\mathrm{d}x=4L^2/\pi^2 ∫ 0 L x f L ( x ) d x = 4 L 2 / π 2 . The local logarithmic singularity is integrable uniformly on compact u u u intervals. Lemma 3.3, applied to w 0 w_0 w 0 whose mass is O ( a L ) O(aL) O ( a L ) , bounds this divided potential uniformly by a constant. Thus the majorants of Lemma 8.2 allow integration against δ w ( a u ) \delta w(au) δ w ( a u ) and removal of the compact restriction. The two cross terms contribute 8 ( log 2 ) a 2 L 2 / π 2 + o ( a 2 L 2 ) 8(\log2)a^2L^2/\pi^2+o(a^2L^2) 8 ( log 2 ) a 2 L 2 / π 2 + o ( a 2 L 2 ) . Since ∣ δ w ( a u ) ∣ ≤ C ( 1 + u ) − 3 |\delta w(au)|\leq C(1+u)^{-3} ∣ δ w ( a u ) ∣ ≤ C ( 1 + u ) − 3 for u ≥ 0 u\geq0 u ≥ 0 , the correction–correction term is O ( a 2 L ) O(a^2L) O ( a 2 L ) .
For the exact geometry use (3.11). In the tail–tail part a factor ∣ t ∣ / k = θ e − x |t|/k=\theta e^{-x} ∣ t ∣/ k = θ e − x yields
C θ a 2 ∫ 0 L e − x f L ( x ) ∫ 0 L f L ( z ) ( min ( x , z ) + 1 ) d z d x = O θ ( a 2 ) , C_\theta a^2\int_0^L e^{-x}f_L(x)\int_0^L f_L(z)(\min(x,z)+1)\,\mathrm{d}z\,\mathrm{d}x=O_\theta(a^2), C θ a 2 ∫ 0 L e − x f L ( x ) ∫ 0 L f L ( z ) ( min ( x , z ) + 1 ) d z d x = O θ ( a 2 ) ,
by f L ( x ) ≤ C x / L f_L(x)\leq Cx/L f L ( x ) ≤ C x / L and ∫ f L ( z ) min ( x , z ) d z ≤ C x L \int f_L(z)\min(x,z)\,\mathrm{d}z\leq CxL ∫ f L ( z ) min ( x , z ) d z ≤ C xL . The convolution remainder and terms involving δ w \delta w δ w are O θ ( a 2 L ) O_\theta(a^2L) O θ ( a 2 L ) , using its absolute and logarithmic moments from Lemma 8.2. All are o ( a 2 L 2 ) o(a^2L^2) o ( a 2 L 2 ) .
Proposition 8.4 (Matching trial bound). The admissible occupation n k t r n_k^{\mathrm{tr}} n k tr satisfies
E ( n k t r ) − E ( P k ) = − q k 3 a 0 ( k ) 2 8 π 2 ( 1 + o ( 1 ) ) . (84) \mathcal{E}(n_k^{\mathrm{tr}})-\mathcal{E}(P_k)=-\frac{qk^3a_0(k)^2}{8\pi^2}(1+o(1)). \tag*{(84)} E ( n k tr ) − E ( P k ) = − 8 π 2 q k 3 a 0 ( k ) 2 ( 1 + o ( 1 )) . (84)
Proof. Put h t r = n k t r − P k h^{\mathrm{tr}}=n_k^{\mathrm{tr}}-P_k h tr = n k tr − P k . Since the trial occupation has the prescribed particle number, the exact decomposition (46) gives
E ( n k t r ) − E ( P k ) = q ∫ ∣ e k ∣ h t r d 3 p − X ( h t r ) − X ( w n k t r ) . \mathcal{E}(n_k^{\mathrm{tr}})-\mathcal{E}(P_k)=q\int\left|e_k\right|h^{\mathrm{tr}}\,\mathrm{d}^{3}p-\mathcal{X}(h^{\mathrm{tr}})-\mathcal{X}(w_{n_k^{\mathrm{tr}}}). E ( n k tr ) − E ( P k ) = q ∫ ∣ e k ∣ h tr d 3 p − X ( h tr ) − X ( w n k tr ) .
Thus the kinetic contribution and the linear ordinary-exchange contribution are evaluated together. For 0 < ∣ t ∣ ≤ R = θ k 0<|t|\le R=\theta k 0 < ∣ t ∣ ≤ R = θ k , the explicit dispersion (44) yields
e k ( r ( t ) ) = k ∣ t ∣ + ∣ t ∣ π log 2 k ∣ t ∣ e_k(r(t))=k|t|+\frac{|t|}{\pi}\log\frac{2k}{|t|} e k ( r ( t )) = k ∣ t ∣ + π ∣ t ∣ log ∣ t ∣ 2 k
+ O θ ( t 2 [ 1 + 1 k log 2 k ∣ t ∣ ] ) . \qquad+O_\theta\left(t^2\left[1+\frac{1}{k}\log\frac{2k}{|t|}\right]\right). + O θ ( t 2 [ 1 + k 1 log ∣ t ∣ 2 k ] ) .
Lemma 8.2 and (61) imply
∫ − R R ∣ t ∣ δ a ( t ) d t = a 2 4 [ L + 2 log 2 − 1 + o ( 1 ) ] , \int_{-R}^{R}|t|\delta_a(t)\,\mathrm{d}t=\frac{a^2}{4}\left[L+2\log2-1+o(1)\right], ∫ − R R ∣ t ∣ δ a ( t ) d t = 4 a 2 [ L + 2 log 2 − 1 + o ( 1 ) ] ,
∫ − R R ∣ t ∣ δ a ( t ) log 2 k ∣ t ∣ d t = a 2 L 2 2 ( 1 4 + 1 π 2 ) + O θ ( a 2 L ) . \int_{-R}^{R}|t|\delta_a(t)\log\frac{2k}{|t|}\,\mathrm{d}t=\frac{a^2L^2}{2}\left(\frac{1}{4}+\frac{1}{\pi^2}\right)+O_\theta(a^2L). ∫ − R R ∣ t ∣ δ a ( t ) log ∣ t ∣ 2 k d t = 2 a 2 L 2 ( 4 1 + π 2 1 ) + O θ ( a 2 L ) .
For the second identity, replace δ a \delta_a δ a by w 0 2 w_0^2 w 0 2 and use log ( 2 k / ∣ t ∣ ) = ℓ + x \log(2k/|t|)=\ell+x log ( 2 k /∣ t ∣ ) = ℓ + x , where x = log ( R / ∣ t ∣ ) x=\log(R/|t|) x = log ( R /∣ t ∣ ) . The replacement error is O θ ( a 2 L ) O_\theta(a^2L) O θ ( a 2 L ) by the absolute logarithmic moment bound in Lemma 8.2 . By (80) , the integrated dispersion remainder is bounded by
C θ ( a 2 R L 2 + a 3 ( 1 + L ) ) = o ( k a 2 ) . C_\theta\left(\frac{a^2R}{L^2}+a^3(1+L)\right)=o(ka^2). C θ ( L 2 a 2 R + a 3 ( 1 + L ) ) = o ( k a 2 ) .
Using the exact volume Jacobian therefore gives
q ∫ e k ∣ h t r ∣ d 3 p = q k 2 2 π 2 ∫ − R R e k ( r ( t ) ) δ a ( t ) d t = q k 3 a 2 8 π 2 [ L + 2 log 2 − 1 + 2 L 2 π k ( 1 4 + 1 π 2 ) + o ( 1 ) ] . \begin{aligned}
q\int e_k|h^{\mathrm{tr}}|\,\mathrm{d}^3p=\frac{qk^2}{2\pi^2}\int_{-R}^{R}e_k(r(t))\delta_a(t)\,\mathrm{d}t \\
&=\frac{qk^3a^2}{8\pi^2}\left[L+2\log2-1+\frac{2L^2}{\pi k}\left(\frac{1}{4}+\frac{1}{\pi^2}\right)+o(1)\right].
\end{aligned} q ∫ e k ∣ h tr ∣ d 3 p = 2 π 2 q k 2 ∫ − R R e k ( r ( t )) δ a ( t ) d t = 8 π 2 q k 3 a 2 [ L + 2 log 2 − 1 + π k 2 L 2 ( 4 1 + π 2 1 ) + o ( 1 ) ] .
The volume-coordinate absolute mass of h t r h^{\mathrm{tr}} h tr is O ( a ) O(a) O ( a ) . Thus Lemma 3.3 gives
0 ≤ X ( h t r ) ≤ C θ q k 2 a 2 ( 1 + log k a ) = o ( q k 3 a 2 ) . 0\le\mathcal{X}(h^{\mathrm{tr}})\le C_\theta qk^2a^2\left(1+\log\frac{k}{a}\right)=o(qk^3a^2). 0 ≤ X ( h tr ) ≤ C θ q k 2 a 2 ( 1 + log a k ) = o ( q k 3 a 2 ) .
For the pairing term, (18) and Lemma 8.3 give
X ( w n k t r ) = q k 2 a 2 4 π 3 [ 2 L 2 π 2 + 4 L 2 π 2 ( ℓ + 2 log 2 ) + o ( L 2 ) ] . \mathcal{X}(w_{n_k^{\mathrm{tr}}})=\frac{qk^2a^2}{4\pi^3}\left[\frac{2L^2}{\pi^2}+\frac{4L^2}{\pi^2}(\ell+2\log2)+o(L^2)\right]. X ( w n k tr ) = 4 π 3 q k 2 a 2 [ π 2 2 L 2 + π 2 4 L 2 ( ℓ + 2 log 2 ) + o ( L 2 ) ] .
Combining the three terms yields
8 π 2 q k 3 a 2 [ E ( n k t r ) − E ( P k ) ] = L − 4 L 3 π 3 k + 2 log 2 − 1 − 8 L 2 π 3 k ( ℓ + 2 log 2 ) + 2 L 2 π k ( 1 4 + 1 π 2 ) + o ( 1 ) . (85) \begin{aligned}
\frac{8\pi^2}{qk^3a^2}\left[\mathcal{E}(n_k^{\mathrm{tr}})-\mathcal{E}(P_k)\right] \\
&=L-\frac{4L^3}{\pi^3k}+2\log2-1-\frac{8L^2}{\pi^3k}(\ell+2\log2)+\frac{2L^2}{\pi k}\left(\frac{1}{4}+\frac{1}{\pi^2}\right)+o(1). \tag*{(85)}
\end{aligned} q k 3 a 2 8 π 2 [ E ( n k tr ) − E ( P k ) ] = L − π 3 k 4 L 3 + 2 log 2 − 1 − π 3 k 8 L 2 ( ℓ + 2 log 2 ) + π k 2 L 2 ( 4 1 + π 2 1 ) + o ( 1 ) . (85)
With L = A k + b 0 L=A\sqrt{k}+b_0 L = A k + b 0 and b 0 = − ℓ − log 2 + 1 / 4 + π 2 / 16 b_0=-\ell-\log2+1/4+\pi^2/16 b 0 = − ℓ − log 2 + 1/4 + π 2 /16 , the right side tends to − 1 -1 − 1 . This proves (84) . □
Proof of Theorem 2.2. The variational principle and Proposition 8.4 give D ( k ) ≥ 3 4 a 0 ( k ) 2 ( 1 + o ( 1 ) ) D(k)\ge\frac{3}{4}a_0(k)^2(1+o(1)) D ( k ) ≥ 4 3 a 0 ( k ) 2 ( 1 + o ( 1 )) . Equation (75) gives the converse inequality with a ∗ a_\ast a ∗ in place of a 0 a_0 a 0 . Since a ∗ / a 0 → 1 a_\ast/a_0\to1 a ∗ / a 0 → 1 , the two bounds agree, and 3 y 2 / 4 = C ∗ 3y^2/4=C_\ast 3 y 2 /4 = C ∗ gives (11) . In particular the exponential rate and polynomial power follow without a preliminary two-sided-scale proposition. □
The occupation-defect profile
Corollary 9.1 (Weighted occupation-defect profile). Let n k n_k n k be the minimizer and write h ~ k ( t ) = n k ( r ( t ) ) − 1 { t < 0 } \widetilde{h}_k(t)=n_k(r(t))-\mathbf{1}_{\{t<0\}} h k ( t ) = n k ( r ( t )) − 1 { t < 0 } . Then
∫ R 3 E ∗ ( p ) ( ∣ n k ( p ) − P k ( p ) ∣ − d ∗ ( p ) ) 2 d 3 p = o ( k 3 a ∗ 2 ) . (86) \int_{\mathbb{R}^3} E_*(p)\bigl(\lvert n_k(p)-P_k(p)\rvert-d_*(p)\bigr)^2\,\mathrm{d}^3p=o(k^3a_*^2). \tag*{(86)} ∫ R 3 E ∗ ( p ) ( ∣ n k ( p ) − P k ( p )∣ − d ∗ ( p ) ) 2 d 3 p = o ( k 3 a ∗ 2 ) . (86) Equivalently, with E ^ k ( u ) = E ∗ ( r ( a ∗ u ) ) / ( k a ∗ ) \widehat{E}_k(u)=E_*(r(a_*u))/(ka_*) E k ( u ) = E ∗ ( r ( a ∗ u )) / ( k a ∗ ) , the same convergence reads
∫ − k / ( 3 a ∗ ) ∞ E ^ k ( u ) ( ∣ h ~ k ( a ∗ u ) ∣ − d ∗ ( r ( a ∗ u ) ) ) 2 d u ⟶ 0. \int_{-k/(3a_*)}^{\infty}\widehat{E}_k(u)\bigl(\lvert\widetilde{h}_k(a_*u)\rvert-d_*(r(a_*u))\bigr)^2\,\mathrm{d}u\longrightarrow0. ∫ − k / ( 3 a ∗ ) ∞ E k ( u ) ( ∣ h k ( a ∗ u )∣ − d ∗ ( r ( a ∗ u )) ) 2 d u ⟶ 0. For a fixed-weight version, define the shell-restricted profile on R \mathbb{R} R by
F k ( u ) = { ∣ h ~ k ( a ∗ u ) ∣ , ∣ u ∣ ≤ R / a ∗ , 0 , ∣ u ∣ > R / a ∗ . F_k(u)=
\begin{cases}
\lvert\widetilde{h}_k(a_*u)\rvert, & \lvert u\rvert\le R/a_*,\\
0, & \lvert u\rvert>R/a_*.
\end{cases} F k ( u ) = { ∣ h k ( a ∗ u )∣ , 0 , ∣ u ∣ ≤ R / a ∗ , ∣ u ∣ > R / a ∗ . Then
∫ R 1 + u 2 ∣ F k ( u ) − d ∞ ( ∣ u ∣ ) ∣ 2 d u ⟶ 0. (87) \int_{\mathbb{R}}\sqrt{1+u^2}\lvert F_k(u)-d_\infty(\lvert u\rvert)\rvert^2\,\mathrm{d}u\longrightarrow0. \tag*{(87)} ∫ R 1 + u 2 ∣ F k ( u ) − d ∞ (∣ u ∣) ∣ 2 d u ⟶ 0. (87) The energy-space convergence (86) also holds for arbitrary (possibly nonradial) n k ∈ A ρ n_k\in\mathcal{A}_\rho n k ∈ A ρ satisfying E ( n k ) − inf A ρ E = o ( k 3 a ∗ 2 ) \mathcal{E}(n_k)-\inf_{\mathcal{A}_\rho}\mathcal{E}=o(k^3a_*^2) E ( n k ) − inf A ρ E = o ( k 3 a ∗ 2 ) . The radial fixed-weight conclusion holds for such approximate minimizers whenever they are radial.
Proof. For the minimizer, Proposition 6.1 yields
q 4 ∥ n k − P k ∣ − d ∗ ∥ E ∗ 2 ≤ q ∥ d ∗ ∥ E ∗ 2 + 5 X ( d ∗ ) − ρ D ( k ) . \frac{q}{4}\lVert n_k-P_k\lvert-d_*\rVert_{E_*}^2\le q\lVert d_*\rVert_{E_*}^2+5\mathcal{X}(d_*)-\rho D(k). 4 q ∥ n k − P k ∣ − d ∗ ∥ E ∗ 2 ≤ q ∥ d ∗ ∥ E ∗ 2 + 5 X ( d ∗ ) − ρ D ( k ) . Lemma 8.1 and Theorem 2.2 show that the right-hand side is o ( k 3 a ∗ 2 ) o(k^3a_*^2) o ( k 3 a ∗ 2 ) . This proves the global assertion and, after the exact volume-coordinate transformation, its rescaled version. On ∣ t ∣ ≤ R \lvert t\rvert\le R ∣ t ∣ ≤ R , (45) implies
E ∗ ( r ( a ∗ u ) ) k a ∗ ≥ c θ 1 + u 2 . \frac{E_*(r(a_*u))}{ka_*}\ge c_\theta\sqrt{1+u^2}. k a ∗ E ∗ ( r ( a ∗ u )) ≥ c θ 1 + u 2 . The deterministic profiles d ∗ ( r ( a ∗ u ) ) d_*(r(a_*u)) d ∗ ( r ( a ∗ u )) , extended by zero outside ∣ u ∣ ≤ R / a ∗ \lvert u\rvert\le R/a_* ∣ u ∣ ≤ R / a ∗ , converge to d ∞ ( ∣ u ∣ ) d_\infty(\lvert u\rvert) d ∞ (∣ u ∣) in the fixed weighted space by (73) and dominated convergence. The triangle inequality proves (87) . The proof for approximate minimizers is identical. □ \square □
Remark 9.2 (Why the global weight is specified). Without a shell cutoff, one should not replace E ^ k ( u ) \widehat{E}_k(u) E k ( u ) by 1 + u 2 \sqrt{1+u^2} 1 + u 2 over the entire positive half-line without another argument. Far beyond the Fermi shell, t ( r ) ≍ r 3 / k 2 t(r)\asymp r^3/k^2 t ( r ) ≍ r 3 / k 2 , whereas E ∗ ( r ) ≍ r 2 E_*(r)\asymp r^2 E ∗ ( r ) ≍ r 2 . Thus these weights are not globally comparable in the direction needed for that stronger statement. The global energy-weighted assertion and the fixed-weight shell assertion above follow directly from completion of the square; no self-consistent exterior-tail analysis is needed for either.
Comparison with quantum-mechanical predictions
Theorems 2.1 and 2.2 concern the specified density-matrix functionals. Their physical interpretation requires a separate comparison with the interacting quantum gas. We make that comparison at the level of energy per particle and explicitly distinguish rigorous functional asymptotics, rigorous quantum results in scaled regimes, and the thermodynamic high-density prediction.
A common normalization and the quantum benchmark
For the physical unpolarized gas take q = 2 q = 2 q = 2 and introduce the Seitz radius (in Bohr radii)
r s = ( 3 4 π ρ ) 1 / 3 , k = κ 0 r s , κ 0 = ( 9 π 4 ) 1 / 3 . (88) r_s = \left(\frac{3}{4\pi\rho}\right)^{1/3}, \qquad k = \frac{\kappa_0}{r_s}, \qquad\kappa_0 = \left(\frac{9\pi}{4}\right)^{1/3}. \tag*{(88)} r s = ( 4 π ρ 3 ) 1/3 , k = r s κ 0 , κ 0 = ( 4 9 π ) 1/3 . (88)
Let e Q ( r s ) e_{\mathrm{Q}}(r_s) e Q ( r s ) denote the thermodynamic ground-state energy per particle of neutral Coulomb jellium in the unpolarized sector. The standard high-density quantum prediction is
e Q ( r s ) − e H F h o m ( r s ) = a Q log r s + B Q + o ( 1 ) , a Q = 1 − log 2 π 2 . (89) e_{\mathrm{Q}}(r_s) - e_{\mathrm{HF}}^{\mathrm{hom}}(r_s) = a_{\mathrm{Q}}\log r_s + B_{\mathrm{Q}} + o(1), \qquad a_{\mathrm{Q}} = \frac{1-\log2}{\pi^2}. \tag*{(89)} e Q ( r s ) − e HF hom ( r s ) = a Q log r s + B Q + o ( 1 ) , a Q = π 2 1 − log 2 . (89)
This is the Macke–Gell-Mann–Brueckner correlation formula; its leading logarithm is obtained by the resummation of the long-range screening contributions [17 , 19 , 11 ] . The Hartree convention and the Hartree–Fock reference are also discussed in [16 ] . The coefficient is a Q = 0.0310906908 … a_{\mathrm{Q}} = 0.0310906908\ldots a Q = 0.0310906908 … in Hartree units; in Rydberg units it is twice as large. The constant B Q B_{\mathrm{Q}} B Q is not needed below. Since log r s = log κ 0 − log k \log r_s = \log\kappa_0 - \log k log r s = log κ 0 − log k , the predicted positive energy gain is
D Q ( k ) : = e H F h o m ( k ) − e Q ( k ) = a Q log k + O ( 1 ) . D_{\mathrm{Q}}(k) := e_{\mathrm{HF}}^{\mathrm{hom}}(k) - e_{\mathrm{Q}}(k) = a_{\mathrm{Q}}\log k + O(1). D Q ( k ) := e HF hom ( k ) − e Q ( k ) = a Q log k + O ( 1 ) .
All statements involving D Q D_{\mathrm{Q}} D Q at this precision are understood relative to the physical benchmark (89) , not as additional theorems proved in this paper.
For clarity, the three results in the same units are summarized in Table 1 . No factor of density is included in any entry. A per-volume correlation is obtained by multiplication by ρ \rho ρ .
Table 1. Correlation scales in Hartree units, with C ∗ = 12 e − 1 / 2 − π 2 / 8 C_* = 12e^{-1/2-\pi^2/8} C ∗ = 12 e − 1/2 − π 2 /8 . The status of the quantum entry differs from that of the two functional theorems.
In the traditional density variable, our functional results become
e M h o m ( r s ) − e H F h o m ( r s ) = 3 4 π 2 log r s + O ( 1 ) , (90) e_{\mathrm{M}}^{\mathrm{hom}}(r_s) - e_{\mathrm{HF}}^{\mathrm{hom}}(r_s) = \frac{3}{4\pi^2}\log r_s + O(1), \tag*{(90)} e M hom ( r s ) − e HF hom ( r s ) = 4 π 2 3 log r s + O ( 1 ) , (90)
e C H F h o m ( r s ) − e H F h o m ( r s ) = − C ∗ κ 0 2 r s − 2 exp ( − π 3 / 2 κ 0 r s ) ( 1 + o ( 1 ) ) . (91) e_{\mathrm{CHF}}^{\mathrm{hom}}(r_s) - e_{\mathrm{HF}}^{\mathrm{hom}}(r_s) = -C_*\kappa_0^2r_s^{-2}\exp\left(-\frac{\pi^{3/2}\sqrt{\kappa_0}}{\sqrt{r_s}}\right)(1+o(1)). \tag*{(91)} e CHF hom ( r s ) − e HF hom ( r s ) = − C ∗ κ 0 2 r s − 2 exp ( − r s π 3/2 κ 0 ) ( 1 + o ( 1 )) . (91)
Thus the Müller coefficient is
a M = 3 4 π 2 = 0.0759908877 … , a M a Q = 3 4 ( 1 − log 2 ) = 2.4441685149 … . a_{\mathrm{M}} = \frac{3}{4\pi^2} = 0.0759908877\ldots, \qquad\frac{a_{\mathrm{M}}}{a_{\mathrm{Q}}} = \frac{3}{4(1-\log2)} = 2.4441685149\ldots. a M = 4 π 2 3 = 0.0759908877 … , a Q a M = 4 ( 1 − log 2 ) 3 = 2.4441685149 … .
Relative to the quantum prediction, Müller therefore overestimates the leading correlation magnitude by this factor. CHF, in contrast, misses the logarithmic scale: conditionally on (89) ,
D M ( k ) D Q ( k ) ⟶ 3 4 ( 1 − log 2 ) , D ( k ) D Q ( k ) ⟶ 0. (92) \frac{D_{\mathrm{M}}(k)}{D_{\mathrm{Q}}(k)} \longrightarrow\frac{3}{4(1-\log2)}, \qquad\frac{D(k)}{D_{\mathrm{Q}}(k)} \longrightarrow0. \tag*{(92)} D Q ( k ) D M ( k ) ⟶ 4 ( 1 − log 2 ) 3 , D Q ( k ) D ( k ) ⟶ 0. (92)
There is no inconsistency with their common leading total energy: both corrections are o ( k ) o(k) o ( k ) , and hence leave the kinetic k 2 k^2 k 2 and exchange k k k terms unchanged. Agreement at those orders is therefore a weak test of the correlation mechanism.
Csányi and Arias [8 ] reported a fitted logarithmic coefficient 0.0570 for CH/Müller in a numerical study covering 0.01 < r s < 30 0.01 < r_s < 30 0.01 < r s < 30 . This range describes the densities studied, not a specified fitting interval. Such a finite-range fit does not determine the limiting coefficient 3 / ( 4 π 2 ) = 0.07599 … 3/(4\pi^2)=0.07599\ldots 3/ ( 4 π 2 ) = 0.07599 … . Our theorem gives that coefficient with a bounded remainder, not a limit for the remainder; it does not attribute the discrepancy to particular finite-range or numerical errors.
To illustrate the CHF scale, denote the positive leading term in (91) by
D l e a d ( r s ) = C ∗ κ 0 2 r s − 2 exp ( − π 3 / 2 κ 0 r s ) . D_{\mathrm{lead}}(r_s)=C_*\kappa_0^2r_s^{-2}\exp\left(-\frac{\pi^{3/2}\sqrt{\kappa_0}}{\sqrt{r_s}}\right). D lead ( r s ) = C ∗ κ 0 2 r s − 2 exp ( − r s π 3/2 κ 0 ) .
Evaluating this expression gives D l e a d ( 0.04 ) ≈ 8.7 × 10 − 14 D_{\mathrm{lead}}(0.04)\approx8.7\times10^{-14} D lead ( 0.04 ) ≈ 8.7 × 1 0 − 14 Hartree, D l e a d ( 0.04 ) / e H F h o m ( 0.04 ) ≈ 1.3 × 10 − 16 D_{\mathrm{lead}}(0.04)/e_{\mathrm{HF}}^{\mathrm{hom}}(0.04)\approx1.3\times10^{-16} D lead ( 0.04 ) / e HF hom ( 0.04 ) ≈ 1.3 × 1 0 − 16 , and D l e a d ( 0.01 ) ≈ 2.5 × 10 − 29 D_{\mathrm{lead}}(0.01)\approx2.5\times10^{-29} D lead ( 0.01 ) ≈ 2.5 × 1 0 − 29 Hartree. These are leading-term evaluations, not certified finite-density minima. The first relative value is comparable to double-precision roundoff, so subtracting total energies is poorly conditioned. Higher precision or a direct calculation of the difference can improve numerical resolution; neither replaces analytic control of the asymptotic law.
Evidence for the Müller lower-bound conjecture
The comparison has a second interpretation, distinct from judging the accuracy of an approximation. A motivating question is whether the Müller ground-state energy is a lower bound to the exact quantum energy for the same Coulomb system [10 , 21 ] . For this question, a larger correlation magnitude has the favorable sign. For the unpolarized homogeneous gas (q = 2 q=2 q = 2 ), Theorem 2.1 and the quantum benchmark (89) give
e Q ( k ) − e M h o m ( k ) = log 2 − 1 4 π 2 log k + O ( 1 ) . (93) e_{\mathrm{Q}}(k)-e_{\mathrm{M}}^{\mathrm{hom}}(k)=\frac{\log2-\frac{1}{4}}{\pi^2}\log k+O(1). \tag*{(93)} e Q ( k ) − e M hom ( k ) = π 2 log 2 − 4 1 log k + O ( 1 ) . (93)
This statement is conditional on (89) .
The spin multiplicity is essential in this comparison. For fixed q q q equally populated components, counting the components in the RPA ring expression [11 ] gives the logarithmic coefficient per particle
a R P A ( q ) = q ( 1 − log 2 ) 2 π 2 , a_{\mathrm{RPA}}(q)=\frac{q(1-\log2)}{2\pi^2}, a RPA ( q ) = 2 π 2 q ( 1 − log 2 ) ,
whereas a M = 3 / ( 4 π 2 ) a_{\mathrm{M}}=3/(4\pi^2) a M = 3/ ( 4 π 2 ) is independent of q q q . Hence, under the corresponding q q q -component quantum high-density benchmark,
e Q ( q ) ( k ) − e M h o m , ( q ) ( k ) = 3 − 2 q ( 1 − log 2 ) 4 π 2 log k + O q ( 1 ) . e_{\mathrm{Q}}^{(q)}(k)-e_{\mathrm{M}}^{\mathrm{hom},(q)}(k)=\frac{3-2q(1-\log2)}{4\pi^2}\log k+O_q(1). e Q ( q ) ( k ) − e M hom , ( q ) ( k ) = 4 π 2 3 − 2 q ( 1 − log 2 ) log k + O q ( 1 ) .
The leading coefficient is positive for q ≤ 4 q\leq4 q ≤ 4 and negative for q ≥ 5 q\geq5 q ≥ 5 . Thus this benchmark supports the proposed lower-bound ordering in the physical case q = 2 q=2 q = 2 , but predicts the opposite ordering for sufficiently large k k k when q ≥ 5 q\geq5 q ≥ 5 . An extension of the lower-bound conjecture to arbitrary spin multiplicities would therefore be incompatible with this high-density benchmark.
By Proposition 4.4 , this comparison is not an artifact of restricting Müller to homogeneous spin-symmetric states: convexity and averaging leave the periodic minimum unchanged, and e M h o m , ( q ) ( k ) e_{\mathrm{M}}^{\mathrm{hom},(q)}(k) e M hom , ( q ) ( k ) equals the unrestricted periodic thermodynamic energy per particle. Under the same quantum high-density benchmark, allowing inhomogeneous or spin-polarized density matrices cannot restore a high-density lower bound for q ≥ 5 q\geq5 q ≥ 5 , whereas for q = 2 q=2 q = 2 the positive logarithmic separation (93) compares the quantum benchmark with the unrestricted periodic Müller thermodynamic energy.
Returning to q = 2 q=2 q = 2 , since log 2 − 1 / 4 > 0 \log2-1/4>0 log 2 − 1/4 > 0 , the predicted separation is strictly positive for all sufficiently large k k k and is itself of logarithmic order; a bounded remainder cannot reverse its sign in that limit. Thus the analytic Müller coefficient supplies additional quantitative evidence for the lower-bound conjecture, not merely a measure of overcorrelation. It does not establish the conjecture for arbitrary densities, particle numbers, atoms, or molecules, and it does not strengthen the mathematical status of the quantum benchmark. No such quantum statement is used in either functional proof.
What is rigorously known for the quantum gas.
The exchange-order control in the high-density quantum problem belongs to the correlation-estimate framework of Graf and Solovej [14 ] . Resolving the much smaller logarithmic correlation term requires additional arguments. For weak, sufficiently regular interactions in mean-field scaling, Benedikter, Nam, Porta, Schlein, and Seiringer [1 ] obtained the random-phase correlation energy. Christiansen, Hainzl, and Nam extended the rigorous upper and matching lower analysis to singular interactions including Coulomb [3 , 4 ] . Their result establishes the corresponding Gell-Mann–Brueckner analogue, including the next order, in a scaled many-fermion problem. It does not by itself establish (89) for unscaled thermodynamic jellium. The latter distinction, rather than the formal similarity of the logarithms, determines what can be used as a proved quantum comparison.
There is a second distinction concerning the reference state. Gontier, Hainzl, and Lewin [13 ] proved that the high-density Hartree–Fock energy improvement from allowing translation-symmetry breaking is exponentially small. Thus it cannot change the leading quantum logarithm. Our reference in both functional theorems is, however, exactly the uniform spin-symmetric Fermi sea. At the exponentially small CHF scale, their result cannot simply be discarded or identified with our exponent and prefactor. We neither identify the unrestricted Hartree–Fock minimizer with P k P_k P k nor analyze CHF in that larger variational class. By contrast, Proposition 4.4 shows that restricting periodic Müller jellium to translation–spin invariant states does not change its minimum at any density; an invariant global minimizer can be chosen.
Why the two approximate mechanisms differ.
The distinction is visible directly in the variational structure. For Müller, the pinned core has b = n = 1 b=\sqrt{n}=1 b = n = 1 . Its exchange field is nonzero at the Fermi surface, with σ P k ( k ) = k / π \sigma_{P_k}(k)=k/\pi σ P k ( k ) = k / π . The fractional tail is therefore driven by an already present core source. In the region 1 ≪ r − k ≪ k 1\ll r-k\ll k 1 ≪ r − k ≪ k , the equation 2 g b = σ b 2gb=\sigma_b 2 g b = σ b produces an inverse energy tail. Its integrated contribution is the logarithm in (31) . The energy cancellation in (32) fixes its coefficient; counting only the exchange defect would incorrectly double it.
For CHF, w n = n ( 1 − n ) w_n=\sqrt{n(1-n)} w n = n ( 1 − n ) vanishes on a filled Fermi sea, so there is no analogous nonzero core source. The deterministic comparison identifies the scale at which a fractional occupation can lower the energy. The angular Coulomb logarithm and the normal Fermi-surface coordinate produce the Green kernel min ( x , y ) \min(x,y) min ( x , y ) , whose first eigenvalue is 4 L 2 / π 2 4L^2/\pi^2 4 L 2 / π 2 . The full comparison norm reaches order one at L ∼ π 3 / 2 k / 2 L\sim\pi^{3/2}\sqrt{k}/2 L ∼ π 3/2 k /2 , whereas each one-sided norm is then close to one half. The latter fact absorbs ordinary exchange, and the completed square makes the energy gain quadratic in the shell width. The linear spectral correction and the nonlinear trial core determine C ∗ C_* C ∗ . This explanation uses no asymptotic formula for the actual self-consistent gap.
These are mechanisms of the approximate functionals, not a derivation of the quantum screening diagrams from them. In particular the auxiliary CHF quantity Δ = V w n \Delta=Vw_n Δ = V w n is not proved to be a many-body excitation gap, and the Hartree–Fock–Bogoliubov representation [22 ] is not evidence for superconductivity in the present quantum comparison.
Scope, atomic results, and functional design.
The pointwise ordering E M ≤ E C H F ≤ E H F \mathcal{E}_{\mathrm{M}}\leq\mathcal{E}_{\mathrm{CHF}}\leq\mathcal{E}_{\mathrm{HF}} E M ≤ E CHF ≤ E HF in (4) orders these approximations, not the exact quantum energy. One-particle representability does not make an approximate two-particle interaction functional variational for the many-body Hamiltonian. Neither an unconditional inequality E M ≤ E Q \mathcal{E}_{\mathrm{M}}\leq\mathcal{E}_{\mathrm{Q}} E M ≤ E Q for all densities nor E Q ≤ E C H F \mathcal{E}_{\mathrm{Q}}\leq\mathcal{E}_{\mathrm{CHF}} E Q ≤ E CHF is proved here. The overcorrelation and undercorrelation conclusions, as well as the lower-bound evidence in § 10.2 , are leading-order comparisons to (89) in the specified unpolarized sector.
Likewise, the atomic agreement through order Z 5 / 3 Z^{5/3} Z 5/3 in [20 , 21 ] does not determine terms on a smaller correlation scale such as Z log Z Z\log Z Z log Z . The homogeneous formulas cannot be inserted into an atomic local-density expression and then declared an atomic theorem: uniform semiclassical and localization errors would have to be controlled separately. No such transfer is used or proved in this paper.
The results separate three tests: leading total energy, correlation scale, and correlation coefficient. They also distinguish approximation accuracy from a favorable sign for a lower bound. A finite-density fit, a contraction constraint, or an auxiliary gap representation alone does not ensure the correct high-density correlation. Both functional limits concern nonrelativistic models. For CHF, translation invariance and spin symmetry remain part of the variational restriction. For Müller, the same energy is the unrestricted periodic thermodynamic energy by Proposition 4.4 ; the thermodynamic limit is taken at fixed density before the high-density limit.
Mathematisches Institut, Ludwig-Maximilians-Universität München, Theresienstr. 39, 80333 München, Germany
Email address: h.s@lmu.de