Introduction
The uniform electron gas (UEG) is the reference system for the exchange–correlation functionals of density-functional theory [1 , 2 , 3 ] . At high density in three dimensions its correlation energy has the expansion
ϵ c ( r s ) = A log r s + B + C r s log r s + D r s + R ( r s ) , (1) \epsilon_c(r_s)=A\log r_s+B+C r_s\log r_s+D r_s+R(r_s), \tag*{(1)} ϵ c ( r s ) = A log r s + B + C r s log r s + D r s + R ( r s ) , (1)
where ϵ c \epsilon_c ϵ c is the correlation energy per electron in Ry, log \log log is to base e, r s r_s r s is the Wigner–Seitz radius in Bohr units and R ( r s ) R(r_s) R ( r s ) is the remainder. The coefficients, reviewed in Ref. [3, Eq. (34)], are A = 2 ( 1 − log 2 ) / π 2 ≈ 0.0622 A=2(1-\log2)/\pi^2\approx0.0622 A = 2 ( 1 − log 2 ) / π 2 ≈ 0.0622 [4 , 5 ] , B ≈ − 0.0938 B\approx-0.0938 B ≈ − 0.0938 [5 , 6 ] and [7 Eq. (8)] , C = α L ( 7 π 2 / 6 − 12 log 2 − 1 ) / ( 2 π 3 ) ≈ 0.0185 C=\alpha_L(7\pi^2/6-12\log2-1)/(2\pi^3)\approx0.0185 C = α L ( 7 π 2 /6 − 12 log 2 − 1 ) / ( 2 π 3 ) ≈ 0.0185 [8 , 9 , 10 ] with α L = [ 4 / ( 9 π ) ] 1 / 3 \alpha_L=[4/(9\pi)]^{1/3} α L = [ 4/ ( 9 π ) ] 1/3 , and D ≈ − 0.020 D\approx-0.020 D ≈ − 0.020 [9 ] and [7 Eq. (8)] .
In two dimensions, the case of the electron layers of semiconductor heterostructures [11 ] , the expansion reads
ϵ c 2 D ( r s ) = a 0 + a ℓ r s log r s + a 1 r s + O ( r s 2 log r s ) , (2) \epsilon_c^{2\mathrm{D}}(r_s)=a_0+a_\ell r_s\log r_s+a_1r_s+O(r_s^2\log r_s), \tag*{(2)} ϵ c 2 D ( r s ) = a 0 + a ℓ r s log r s + a 1 r s + O ( r s 2 log r s ) , (2)
with a 0 = 2 ( log 2 − 1 ) + 2 G − 16 β ( 4 ) / π 2 = − 0.38499 a_0=2(\log2-1)+2G-16\beta(4)/\pi^2=-0.38499 a 0 = 2 ( log 2 − 1 ) + 2 G − 16 β ( 4 ) / π 2 = − 0.38499 Ry, where G = β ( 2 ) G=\beta(2) G = β ( 2 ) is Catalan’s constant and β \beta β the Dirichlet beta function [12 ] , and a ℓ = − 2 2 [ 10 / ( 3 π ) − 1 ] = − 0.17263 a_\ell=-2\sqrt{2}[10/(3\pi)-1]=-0.17263 a ℓ = − 2 2 [ 10/ ( 3 π ) − 1 ] = − 0.17263 Ry [13 , 12 ] . The coefficient a 1 a_1 a 1 is not known [3 Table I] , and neither is the next log r s \log r_s log r s term. The three-dimensional expansion has been tested against quantum Monte Carlo energies [14 ] . In the analogous mean-field (semiclassical) scaling limit, its leading term has been proved by matching upper and lower bounds [15 , 16 ] . The expansion has also been applied to exchange contributions to the self-energy and the momentum distribution [17 , 18 ] .
Perturbation theory appears to give a single log r s \log r_s log r s at each order in three dimensions [3 Eq. (21)] , [7 Sec. I] and [4 ] , but this form extrapolates the known orders. Which diagrams contribute at the next orders has been determined order by order, each order with its own resummation. The dimension changes the form of the log r s \log r_s log r s terms. For the ring diagrams, Schlesier, Benavides-Riveros and Marques found that the log r s \log r_s log r s term of three dimensions becomes a power of r s r_s r s for d > 3 d > 3 d > 3 [19 ] . In two dimensions, the first log r s \log r_s log r s term of the ring diagrams moves to r s log r s r_s \log r_s r s log r s [13 ] . The mechanism behind a single log r s \log r_s log r s in three dimensions is well known [5 , 20 , 7 ] : screening cuts off the small-momentum Coulomb divergence at the Thomas–Fermi wave number k T F ∝ r s 1 / 2 k F k_{\mathrm{TF}} \propto r_s^{1/2} k_{\mathrm{F}} k TF ∝ r s 1/2 k F , with k F k_{\mathrm{F}} k F the Fermi wave number. The integral ∫ k T F k F d q / q \int_{k_{\mathrm{TF}}}^{k_{\mathrm{F}}} \mathrm{d}q/q ∫ k TF k F d q / q over the momentum transfer q q q thus leaves log r s \log r_s log r s . What is missing is a criterion that predicts, before a diagram is evaluated and in any dimension, at which order a log r s \log r_s log r s term appears, how many factors of log r s \log r_s log r s it carries, and when such terms cancel between diagrams.
A second question concerns correlation-energy methods based, like the G W GW G W approximation [21 ] , on the screened interaction of the random-phase approximation (RPA). Their exchange corrections, such as second-order screened exchange (SOSEX), are widely used in quantum chemistry and materials science [22 , 23 , 24 ] and are built from the diagrams studied here. The screened second-order exchange diagram is called the kite after its shape. Benites, Rosado and Manousakis (BRM) distinguish the kite from SOSEX (§ 3.2 ). They computed the energy of the kite over a wide range of densities and fitted the high-density part with terms linear in log r s \log r_s log r s [25 ] . Numerical values of a single diagram cannot show whether a screened vertex correction evaluated alone produces terms that the accompanying self-energy corrections would remove.
In this paper we give the missing criterion for the ring series and for diagrams linear in one screened line, in d d d dimensions with the three-dimensional Coulomb interaction e 2 / r e^2/r e 2 / r . With q q q the momentum transfer of the line in units of k F k_{\mathrm{F}} k F , the energy has at high density the form ∫ 0 ∞ k ( q ) g ( q / ϵ ) d q \int_0^\infty k(q)\,g(q/\epsilon)\,\mathrm{d}q ∫ 0 ∞ k ( q ) g ( q / ϵ ) d q . This integral is a Mellin convolution of a radial kernel k k k , the contribution of the diagram per unit q q q , with a scaling function g g g , which depends on q q q only through q / ϵ q/\epsilon q / ϵ . At high density, for a diagram linear in the line, the scaling function is the relative change of the line caused by screening. The scale ϵ = k T F / k F \epsilon= k_{\mathrm{TF}}/k_{\mathrm{F}} ϵ = k TF / k F depends on the density. Each term of the small-q q q and large-q q q expansions of k k k and g g g , an endpoint term , gives a pole of the Mellin transform. The density expansion is read off these poles [26 , 27 , 28 , 29 ] (§ 2 ). The approach adapts the Mellin analysis of Feynman integrals in relativistic field theory [30 , 31 ] to many-body diagrams. No step requires a resummation tailored to a particular diagram.
In the Mellin variable s s s , long-range screening places the poles of the Mellin transform of the scaling function at the integer multiples s = N ( d − 1 ) s = N(d - 1) s = N ( d − 1 ) , N = 1 , 2 , … N = 1, 2, \ldots N = 1 , 2 , … , and the phase space of the diagram places the kernel pole at s = d + 1 s = d + 1 s = d + 1 for the rings and at s = 2 s = 2 s = 2 for a polarization insertion (§ 2.8 and Figure 2 ). Once the Fermi-surface factors log ( 1 / q ) \log(1/q) log ( 1/ q ) of the kernel (§ 2.1 ) cancel between the vertex and self-energy corrections, as the Ward identity requires, a log r s \log r_s log r s term appears only where two poles coincide, a resonance . This rule reproduces the Gell-Mann–Brueckner (A A A ) and Carr–Maraududin (C C C ) terms in three dimensions, the Rajagopal–Kimball term (a ℓ a_\ell a ℓ ) in two, and the power law of the rings for d > 3 d > 3 d > 3 found by Schlesier et al. [19 ] .
Figure 2. Positions s s s of the Mellin poles against the dimension d d d . Gray lines: poles s = N ( d − 1 ) s = N(d - 1) s = N ( d − 1 ) of the Thomas–Fermi tail of the RPA scaling function, labeled by N N N . Colored lines: kernel poles of a polarization insertion (s = 2 s = 2 s = 2 , Secs. 2.9 and 3.1), of the ring series (s = d + 1 s = d+1 s = d + 1 , Sec. 2.8) and of a polarization insertion of order q 2 q^2 q 2 , such as the q 2 q^2 q 2 term of the Lindhard function (s = 4 s = 4 s = 4 , dashed). Open circles mark the coincidences that give the log r s \log r_s log r s terms of (1) and (2), the term a 2 ℓ r s 2 log r s a_{2\ell} r_s^2 \log r_s a 2 ℓ r s 2 log r s obtained in Sec. 3.1, and the term r s 3 log r s r_s^3 \log r_s r s 3 log r s from the q 2 q^2 q 2 term of the two-dimensional Lindhard function. Filled dots mark coincidences at noninteger d d d .
The new result is the coefficient of the next log r s \log r_s log r s term in two dimensions (Sec. 3.1). There the kernel pole of the polarization insertion at s = 2 s = 2 s = 2 meets the second screening pole, and the conserving first-order exchange polarization Π ( 1 ) \Pi^{(1)} Π ( 1 ) contributes a 2 ℓ r s 2 log r s a_{2\ell} r_s^2 \log r_s a 2 ℓ r s 2 log r s to (2) . We obtain the long-wavelength limit of this polarization in closed form. It interpolates between the static and high-frequency limits known from earlier work [32 , 33 , 34 , 35 ] . Diagrams in which three or more screened lines of small momentum transfer meet at one closed fermion loop enter at order r s 4 r_s^4 r s 4 in two dimensions by power counting, because such loops, summed over the orderings of their vertices, remain bounded at small momenta Q Q Q and frequencies of order v F Q v_F Q v F Q [36 , 37 ] (Sec. 3.1).
In Secs. 3.2–3.4 we follow the Ward-identity cancellation for the screened kite in three dimensions. The kite alone acquires a term r s log 2 r s r_s \log^2 r_s r s log 2 r s at high density and a term r s − 1 / 4 r_s^{-1/4} r s − 1/4 at low density. The fit of BRM absorbs the first of these into its r s log r s r_s \log r_s r s log r s coefficient. The Fock self-energy corrections [38 , 39 , 40 , 41 ] [42] cancel both terms, with remainder bounds. This cancellation is consistent with the Ward identity [43, 44] and with conserving approximations [45, 46, 47, 48, 49].
The spin dependence of the isolated kite is the subject of a companion paper [50]. The paramagnetic leading coefficients of that paper agree with the present ones. Low-density limits of single diagrams with free propagators are formal, since the physical ground state at low density is the Wigner crystal [51, 52]. They nevertheless test the method, and their exponents can be used to check interpolating parametrizations of density functionals [53, 54, 55] fitted to quantum Monte Carlo energies [56].
The method is developed in Sec. 2. It derives the rule that a log r s \log r_s log r s term appears only at a resonance and ends with the check of this rule against the known log r s \log r_s log r s terms in two and three dimensions and against the ring series for d > 3 d>3 d > 3 (Secs. 2.8 and 2.9). Section 3 gives the results: the r s 2 log r s r_s^2 \log r_s r s 2 log r s term of the two-dimensional gas (Sec. 3.1) and the Ward-identity cancellation for the kite (Secs. 3.2–3.4). Proofs, remainder bounds and numerical details are given in the Supporting Information (SI). SI table S1 lists, for each result, the SI sections that derive it. SI table S2 lists the main symbols and where each is defined.
Theory: screened diagrams as Mellin convolutions
Overview
The method has three steps.
The first step is the spectral representation (Sec. 2.4). The screened line W W W is written as the bare line minus a superposition of Lorentzian lines. This step is needed at low density, where the line reduces to a single Lorentzian, and for the remainder bounds. At high density the next step follows directly from W W W at fixed ν / q \nu/q ν / q .
The second step is the Mellin convolution (Secs. 2.5 and 2.6). After the frequency integration the energy of a diagram linear in W W W becomes I ( ε ) = ∫ 0 ∞ k ( q ) g ( q / ε ) d q I(\varepsilon)=\int_0^\infty k(q)g(q/\varepsilon)\,dq I ( ε ) = ∫ 0 ∞ k ( q ) g ( q / ε ) d q , the Mellin convolution of a radial kernel k k k , fixed by the diagram, with a scaling function g g g , fixed by the line (at low density together with the frequency factor of the diagram at the width of the line, Sec. 2.6). This form is exact for the limiting forms of the line in each density limit, for example with the Lindhard function replaced by its small-q q q limit at high density. The corrections to these limiting forms are bounded separately. Apart from the factor r s n − 2 r_s^{n-2} r s n − 2 of a diagram with n n n bare interactions (Sec. 2.3), the density enters only through the scale ε \varepsilon ε . At high density this scale is the scaled Thomas–Fermi wave number q T F = k T F / k F q_{\mathrm{TF}}=k_{\mathrm{TF}}/k_F q TF = k TF / k F , with q T F d − 1 ∝ r s q_{\mathrm{TF}}^{d-1}\propto r_s q TF d − 1 ∝ r s (Sec. 2.2). At low density in three dimensions it is the scale q ∗ ∝ r s 1 / 4 q_* \propto r_s^{1/4} q ∗ ∝ r s 1/4 (Sec. 2.6).
The ring series is a nonlinear functional of the line, but its energy has the same form, with a scalar function of the screening ratio c c c of (8) as the scaling function (§ 2.8 ).
The third step is the endpoint–residue correspondence (§ 2.7 ). Each endpoint term of k k k or of g g g at q → 0 q \to0 q → 0 or q → ∞ q \to\infty q → ∞ gives one pole of the Mellin transform in a complex variable s s s . The position of the pole gives the power of r s r_s r s , and its order minus one gives the highest power of log r s \log r_s log r s .
Two poles at the same position add their orders. Diagrams whose kernels have equal and opposite endpoint terms cancel pole by pole.
In a Fermi sea the kernel of a polarization insertion has the small-q q q endpoint term k ≃ q log m ( 1 / q ) k \simeq q\log^{m}(1/q) k ≃ q log m ( 1/ q ) . Its factor q q q comes from phase space, and each factor log ( 1 / q ) \log(1/q) log ( 1/ q ) , a Fermi-surface factor , comes from the vertex or self-energy at the Fermi surface. The kernel of the ring series has the small-q q q endpoint term k ∝ q d k \propto q^d k ∝ q d .
The RPA line contributes universal poles , which do not depend on the diagram. At high density these poles come from its Thomas–Fermi tail , the decay of the screening correction in powers ( q T F / q ) N ( d − 1 ) (q_{\mathrm{TF}}/q)^{N(d-1)} ( q TF / q ) N ( d − 1 ) , N = 1 , 2 , … N=1,2,\ldots N = 1 , 2 , … , for q ≫ q T F q \gg q_{\mathrm{TF}} q ≫ q TF . They lie at s = N ( d − 1 ) s=N(d-1) s = N ( d − 1 ) . In three dimensions the first of them is at s = 2 s=2 s = 2 . At low density in three dimensions, the pole at s = − 2 s=-2 s = − 2 comes from the large-width term , the first term of the expansion of the low-density scaling function in 1 / Λ λ 1/\Lambda_\lambda 1/ Λ λ , of order q / Λ λ ∝ t 2 q/\Lambda_\lambda\propto t^2 q / Λ λ ∝ t 2 (§ 2.6 ). At low density the screened line is strongly reduced below the plasma frequency and close to the bare line above it. The energy of the large-width term comes from this high-frequency window, where only the tail ν − 2 \nu^{-2} ν − 2 of the diagram in the imaginary frequency ν \nu ν (§ 2.2 ) enters. This energy is the leading low-density term only when the kernel decays faster than q − 3 q^{-3} q − 3 at large q q q (Corollary 1 ).
Sections 2.8, 2.9 and 3 apply the three steps to particular diagrams. In each case the scaling functions are those of § 2.6 , and the work consists of finding the endpoint terms of the radial kernel and the poles they produce.
Units and RPA screening
We consider the paramagnetic UEG in d d d dimensions at zero temperature, with free Green functions G 0 G_0 G 0 for all propagators and the Coulomb interaction e 2 / r e^2/r e 2 / r of three-dimensional electrostatics. The d d d -dimensional Fourier transform of this interaction is C d e 2 / q d − 1 C_d e^2/q^{d-1} C d e 2 / q d − 1 with C d = ( 4 π ) ( d − 1 ) / 2 Γ ( ( d − 1 ) / 2 ) C_d=(4\pi)^{(d-1)/2}\Gamma((d-1)/2) C d = ( 4 π ) ( d − 1 ) /2 Γ (( d − 1 ) /2 ) [19 ] , so that C 3 = 4 π C_3=4\pi C 3 = 4 π and C 2 = 2 π C_2=2\pi C 2 = 2 π . For d = 2 d=2 d = 2 the model is the strictly two-dimensional electron gas [13 , 12 ] . Noninteger d d d is used only to follow how the poles move with the dimension (§ 2.8 ). With n e n_e n e the electron density, a B a_B a B the Bohr radius and V d = π d / 2 / Γ ( d / 2 + 1 ) V_d=\pi^{d/2}/\Gamma(d/2+1) V d = π d /2 /Γ ( d /2 + 1 ) the volume of the unit ball, the density parameter is defined by n e V d ( r s a B ) d = 1 n_eV_d(r_sa_B)^d=1 n e V d ( r s a B ) d = 1 . With two spin states, n e = 2 V d k F d / ( 2 π ) d n_e=2V_d k_F^d/(2\pi)^d n e = 2 V d k F d / ( 2 π ) d , so that
k F a B = 1 α d r s , α d = [ 2 d − 1 Γ ( d 2 + 1 ) ] − 1 / d , α 3 = α L = ( 4 9 π ) 1 / 3 , α 2 = 1 2 . (3) k_Fa_B=\frac{1}{\alpha_d r_s},\qquad\alpha_d=\left[2^{d-1}\Gamma\left(\frac{d}{2}+1\right)\right]^{-1/d},\qquad\alpha_3=\alpha_L=\left(\frac{4}{9\pi}\right)^{1/3},\qquad\alpha_2=\frac{1}{\sqrt{2}}. \tag*{(3)} k F a B = α d r s 1 , α d = [ 2 d − 1 Γ ( 2 d + 1 ) ] − 1/ d , α 3 = α L = ( 9 π 4 ) 1/3 , α 2 = 2 1 . (3)
The coupling constant of the gas is e 2 / ( ℏ v F ) = 1 / ( k F a B ) = α d r s e^2/(\hbar v_F)=1/(k_Fa_B)=\alpha_d r_s e 2 / ( ℏ v F ) = 1/ ( k F a B ) = α d r s , with v F v_F v F the Fermi velocity. With m e m_e m e the electron mass and e e e the elementary charge, we use the units (called Fermi units below)
E 0 = ℏ 2 k F 2 m e = 2 E F , ω 0 = E 0 / ℏ , v 0 = C d e 2 k F d − 1 . (4) E_0=\frac{\hbar^2k_F^2}{m_e}=2E_F,\qquad\omega_0=E_0/\hbar,\qquad v_0=\frac{C_de^2}{k_F^{d-1}}. \tag*{(4)} E 0 = m e ℏ 2 k F 2 = 2 E F , ω 0 = E 0 /ℏ , v 0 = k F d − 1 C d e 2 . (4)
Here E F E_F E F is the Fermi energy, q \mathbf{q} q the momentum transfer with magnitude q q q , and i ν i\nu i ν , ν > 0 \nu> 0 ν > 0 , the imaginary frequency. Wave vectors are measured in k F k_F k F , frequencies in ω 0 \omega_0 ω 0 , interactions v v v and W W W in v 0 v_0 v 0 , responses Π \Pi Π and χ \chi χ in v 0 − 1 v_0^{-1} v 0 − 1 [all introduced with (5) below], and Green functions and self-energies in E 0 − 1 E_0^{-1} E 0 − 1 and E 0 E_0 E 0 . In these units v ( q ) = q 1 − d v(q)=q^{1-d} v ( q ) = q 1 − d and the free dispersion is k 2 / 2 k^2/2 k 2 /2 . At small momentum transfer the excitation energy p ⋅ q + q 2 / 2 \mathbf{p}\cdot\mathbf{q}+q^2/2 p ⋅ q + q 2 /2 of a pair with hole momentum p \mathbf{p} p at the Fermi surface is of order q q q , since the Fermi velocity is 1 in these units. We therefore use the scaled frequency z = ν / q z=\nu/q z = ν / q . Diagrams are evaluated in these units, and the energies are converted at the end to Rydbergs per electron, with E 0 = 2 ( k F a B ) 2 R y = 2 / ( α d r s ) 2 R y E_0=2(k_Fa_B)^2\ \mathrm{Ry}=2/(\alpha_dr_s)^2\ \mathrm{Ry} E 0 = 2 ( k F a B ) 2 Ry = 2/ ( α d r s ) 2 Ry .
The correlation energy is the integral of the interaction energy over the coupling strength. The energy expressions of Sec. 2.3 differ in how they weight this integral. We therefore introduce the interaction-strength parameter 0 ≤ λ ≤ 1 0\leq\lambda\leq1 0 ≤ λ ≤ 1 at fixed density through H ^ λ = T ^ + λ V ^ \hat{H}_\lambda=\hat{T}+\lambda\hat{V} H ^ λ = T ^ + λ V ^ , with T ^ \hat{T} T ^ the kinetic energy and V ^ \hat{V} V ^ the Coulomb interaction including the neutralizing background. The bare and screened lines at coupling λ \lambda λ are denoted by λ v \lambda v λ v and W λ W_\lambda W λ . The density response, the dielectric function ϵ λ \epsilon_\lambda ϵ λ and the screened line are given by
χ λ = Π λ 1 − λ v Π λ , ϵ λ = 1 − λ v Π λ , W λ = λ v + ( λ v ) χ λ ( λ v ) . (5) \chi_\lambda=\frac{\Pi_\lambda}{1-\lambda v\Pi_\lambda},\qquad\epsilon_\lambda=1-\lambda v\Pi_\lambda,\qquad W_\lambda=\lambda v+(\lambda v)\chi_\lambda(\lambda v). \tag*{(5)} χ λ = 1 − λ v Π λ Π λ , ϵ λ = 1 − λ v Π λ , W λ = λ v + ( λ v ) χ λ ( λ v ) . (5)
Here Π \Pi Π is the proper (one-interaction-irreducible) response, written Π ∗ \Pi^* Π ∗ in Ref. [57, Chap. 4, Sec. 12]. These relations fix its sign and normalization. The full density response is χ \chi χ , and Σ \Sigma Σ denotes the proper self-energy.
Summing chains of free polarization bubbles joined by Coulomb lines gives the RPA screened interaction [21 ] [58], in which Π λ = Π 0 = χ 0 \Pi_\lambda=\Pi_0=\chi_0 Π λ = Π 0 = χ 0 is the free response:
W λ = λ v + λ v Π 0 W λ , W_\lambda=\lambda v+\lambda v\Pi_0W_\lambda, W λ = λ v + λ v Π 0 W λ ,
W λ ( q , i ν ) = λ v ( q ) 1 − λ v ( q ) Π 0 ( q , i ν ) = λ q d − 1 + λ q T F d − 1 L ( q , ν ) . (6) W_\lambda(q,i\nu)=\frac{\lambda v(q)}{1-\lambda v(q)\Pi_0(q,i\nu)}
=\frac{\lambda}{q^{d-1}+\lambda q_{\mathrm{TF}}^{d-1}\mathcal{L}(q,\nu)}. \tag*{(6)} W λ ( q , i ν ) = 1 − λ v ( q ) Π 0 ( q , i ν ) λ v ( q ) = q d − 1 + λ q TF d − 1 L ( q , ν ) λ . (6)
The last expression of (6) uses Π 0 = − q T F d − 1 L \Pi_0=-q_{\mathrm{TF}}^{d-1}\mathcal{L} Π 0 = − q TF d − 1 L of Eq. (7) below. With k T F k_{\mathrm{TF}} k TF the Thomas–Fermi wave number and A d = 2 π d / 2 / Γ ( d / 2 ) A_d=2\pi^{d/2}/\Gamma(d/2) A d = 2 π d /2 /Γ ( d /2 ) the area of the unit sphere, the free response and the scaled Thomas–Fermi wave number are
Π 0 = χ 0 = − q T F d − 1 L ( q , ν ) , q T F = k T F k F , q T F d − 1 = B d k F a B = a T F r s , B d = 2 A d C d ( 2 π ) d . (7) \Pi_0=\chi_0=-q_{\mathrm{TF}}^{d-1}\mathcal{L}(q,\nu),\qquad q_{\mathrm{TF}}=\frac{k_{\mathrm{TF}}}{k_F},\qquad q_{\mathrm{TF}}^{d-1}=\frac{B_d}{k_Fa_B}=a_{\mathrm{TF}}r_s,\qquad B_d=\frac{2A_dC_d}{(2\pi)^d}. \tag*{(7)} Π 0 = χ 0 = − q TF d − 1 L ( q , ν ) , q TF = k F k TF , q TF d − 1 = k F a B B d = a TF r s , B d = ( 2 π ) d 2 A d C d . (7)
where a T F = B d α d a_{\mathrm{TF}}=B_d\alpha_d a TF = B d α d . In three dimensions B 3 = 4 / π B_3=4/\pi B 3 = 4/ π , q T F 2 = 4 / ( π a B k F ) q_{\mathrm{TF}}^2=4/(\pi a_Bk_F) q TF 2 = 4/ ( π a B k F ) and a T F = 4 α L / π a_{\mathrm{TF}}=4\alpha_L/\pi a TF = 4 α L / π . In two dimensions B 2 = 2 B_2=2 B 2 = 2 and q T F = 2 r s q_{\mathrm{TF}}=\sqrt{2}r_s q TF = 2 r s . The nonnegative Lindhard function L \mathcal{L} L is written with the real argument ν \nu ν although Π 0 \Pi_0 Π 0 is evaluated at i ν i\nu i ν (the same convention is used for Y D Y_D Y D below). It satisfies L ( 0 , 0 ) = 1 \mathcal{L}(0,0)=1 L ( 0 , 0 ) = 1 . Its closed form is given in Ref. [57] for d = 3 d=3 d = 3 and in Ref. [59] for d = 2 d=2 d = 2 .
We write the screening ratio and the long-wavelength Lindhard shape , the small-q q q limit of L \mathcal{L} L at fixed z = ν / q z=\nu/q z = ν / q , as
c ( q , ν ) = − v ( q ) Π 0 ( q , i ν ) = q T F d − 1 L ( q , ν ) q d − 1 , (8) c(q,\nu)=-v(q)\Pi_0(q,i\nu)=\frac{q_{\mathrm{TF}}^{d-1}\mathcal{L}(q,\nu)}{q^{d-1}}, \tag*{(8)} c ( q , ν ) = − v ( q ) Π 0 ( q , i ν ) = q d − 1 q TF d − 1 L ( q , ν ) , (8)
h d ( z ) = lim q → 0 L ( q , q z ) = ⟨ μ 2 μ 2 + z 2 ⟩ S d − 1 = 1 − 2 F 1 ( 1 , 1 2 ; d 2 ; − z − 2 ) , (9) h_d(z)=\lim_{q\to0}\mathcal{L}(q,qz)=\left\langle\frac{\mu^2}{\mu^2+z^2}\right\rangle_{S^{d-1}}=1-{}_2F_1\left(1,\frac{1}{2};\frac{d}{2};-z^{-2}\right), \tag*{(9)} h d ( z ) = q → 0 lim L ( q , q z ) = ⟨ μ 2 + z 2 μ 2 ⟩ S d − 1 = 1 − 2 F 1 ( 1 , 2 1 ; 2 d ; − z − 2 ) , (9)
so that h d h_d h d gives the dependence of the long-wavelength response on the scaled frequency. Here μ = p ⋅ q ^ ^ \mu=\hat{\mathbf{p}\cdot\hat{\mathbf{q}}} μ = p ⋅ q ^ ^ and the average runs over the unit sphere S d − 1 S^{d-1} S d − 1 of hole directions p ^ \hat{\mathbf{p}} p ^ . The limit holds because at small q q q the occupation difference confines the hole momentum to a shell of width q ∣ μ ∣ q|\mu| q ∣ μ ∣ at the Fermi surface, where the excitation energy is q μ q\mu q μ (SI § 12 ). In particular
h ( z ) = h 3 ( z ) = 1 − z arctan 1 z , h 2 ( z ) = 1 − z 1 + z 2 , h 4 ( z ) = ( 1 + z 2 − z ) 2 , (10) h(z) = h_3(z) = 1 - z \arctan\frac{1}{z}, \qquad h_2(z) = 1 - \frac{z}{\sqrt{1+z^2}}, \qquad h_4(z) = \left(\sqrt{1+z^2}-z\right)^2, \tag*{(10)} h ( z ) = h 3 ( z ) = 1 − z arctan z 1 , h 2 ( z ) = 1 − 1 + z 2 z , h 4 ( z ) = ( 1 + z 2 − z ) 2 , (10)
for z ≥ 0 z \ge0 z ≥ 0 , with h d h_d h d even in z z z , h d ( 0 ) = 1 h_d(0)=1 h d ( 0 ) = 1 and h d ( z ) ≃ 1 / ( d z 2 ) h_d(z) \simeq1/(d z^2) h d ( z ) ≃ 1/ ( d z 2 ) for z → ∞ z \to\infty z → ∞ . In three dimensions the limit also follows from the integral representation of L ( q , q z ) \mathcal{L}(q,qz) L ( q , q z ) in SI Eq. (149) , which gives h ( z ) = ∫ 0 1 a 2 d a / ( a 2 + z 2 ) h(z)=\int_0^1 a^2\,\mathrm{d}a/(a^2+z^2) h ( z ) = ∫ 0 1 a 2 d a / ( a 2 + z 2 ) with the error bound of SI Eq. (150) . For d = 3 d=3 d = 3 and q ≫ 1 q\gg1 q ≫ 1 , L ≃ 4 / [ 3 q 2 ( 1 + 4 ν 2 / q 4 ) ] \mathcal{L}\simeq4/[3q^2(1+4\nu^2/q^4)] L ≃ 4/ [ 3 q 2 ( 1 + 4 ν 2 / q 4 )] (§ 2.6 ).
Diagrams linear in one screened line
We write the formulas of this and the next two subsections for d = 3 d=3 d = 3 , as needed for the kite (§ 3.2 –§ 3.4 ). In d d d dimensions d 3 q d^3q d 3 q is replaced by d d q d^d q d d q and the angular factor 4 π 4\pi 4 π by A d A_d A d . The d d d -dimensional forms are given where they are used (§ 2.8 , § 2.9 and 3.1).
Consider a connected closed diagram D D D built from free propagators G 0 G_0 G 0 and bare Coulomb lines, and select one interaction line. Replacing it by a line f f f and carrying out all other internal integrations gives an energy per electron, in Ry, that is linear in f f f :
J D [ f ] = ∫ d 3 q ∫ 0 ∞ d ν π f ( q , i ν ) Y D ( q , ν ) , (11) \mathcal{J}_D[f]=\int\mathrm{d}^3q\int_0^\infty\frac{\mathrm{d}\nu}{\pi}\,f(q,\mathrm{i}\nu)\,Y_D(q,\nu), \tag*{(11)} J D [ f ] = ∫ d 3 q ∫ 0 ∞ π d ν f ( q , i ν ) Y D ( q , ν ) , (11)
where the bare diagram is J D [ v ] \mathcal{J}_D[v] J D [ v ] . Cutting the selected line leaves a polarization insertion Π D ( q , i ν ) \Pi_D(q,\mathrm{i}\nu) Π D ( q , i ν ) with two density vertices. We say that Π D \Pi_D Π D is closed with the line f f f . For second-order exchange, Π D \Pi_D Π D is the first-order exchange vertex correction Π v t x ( 1 ) \Pi_{\mathrm{vtx}}^{(1)} Π vtx ( 1 ) (§ 3.4 ). The integrand is Y D = − 12 Π D / ( S D π q T F 4 ) = r s n − 2 Y ^ D Y_D=-12\Pi_D/(S_D\pi q_{\mathrm{TF}}^4)=r_s^{n-2}\widehat{Y}_D Y D = − 12 Π D / ( S D π q TF 4 ) = r s n − 2 Y D , with n n n the number of bare interactions of D D D . Here 1 / S D 1/S_D 1/ S D is the closed-diagram coefficient (1 / 4 1/4 1/4 for the kite, § 3.2 ). When Π D \Pi_D Π D is the sum of m m m insertions that close to the same diagram, 1 / S D 1/S_D 1/ S D is that coefficient divided by m m m , so that S D = 4 S_D=4 S D = 4 also for the self-energy pair of § 3.4 . The insertion Π D \Pi_D Π D carries the fermion signs and spin sums and is normalized like the responses of SI § 20.2 . The factor Y ^ D \widehat{Y}_D Y D is independent of density in Fermi units. For the kite this form is derived in SI § 20.1 and § 20.2 [Eqs. (75) –(80) and (S20)], and the end of SI § 20.1 states the conditions under which the same steps hold for any D D D . The factor r s n − 2 r_s^{n-2} r s n − 2 arises because each bare interaction carries v 0 k F d / E 0 = C d α d r s v_0 k_{\mathrm{F}}^d/E_0=C_d\alpha_d r_s v 0 k F d / E 0 = C d α d r s relative to the kinetic scale E 0 E_0 E 0 (4 π α L r s 4\pi\alpha_L r_s 4 π α L r s for d = 3 d=3 d = 3 ), while E 0 E_0 E 0 itself is proportional to r s − 2 r_s^{-2} r s − 2 in Rydbergs, in every dimension.
A single diagram with a screened line has no unique energy, because the chain diagrams that build the screened line can be weighted in different ways. We use three energy expressions , which differ in the weights they assign to the diagrams that contain the screened line. The density expansions of the full-coupling energy and of the coupling averages below have the same exponents, apart from a term of order r s − 2 r_s^{-2} r s − 2 at low density. Their coefficients are related by the Mellin multipliers of § 2.7 . The full-coupling energy is F D ( r s ) = J D [ W λ = 1 ( r s ) ] F_D(r_s)=\mathcal{J}_D[W_{\lambda=1}(r_s)] F D ( r s ) = J D [ W λ = 1 ( r s )] , where the argument r s r_s r s of a line records the density at which it is evaluated. Coupling-constant averages take the form
e D ( r s ) = ∫ 0 1 w D ( λ ) F D ( λ r s ) d λ , (12) e_D(r_s)=\int_0^1 w_D(\lambda)\,F_D(\lambda r_s)\,\mathrm{d}\lambda, \tag*{(12)} e D ( r s ) = ∫ 0 1 w D ( λ ) F D ( λ r s ) d λ , (12)
because coupling and density enter the screening denominator only through λ q T F \lambda q_{\mathrm{TF}} λ q TF , so that W λ ( r s ) = λ W λ = 1 ( λ r s ) W_\lambda(r_s)=\lambda W_{\lambda=1}(\lambda r_s) W λ ( r s ) = λ W λ = 1 ( λ r s ) by (6) and q T F 2 = α T F r s q_{\mathrm{TF}}^2=\alpha_{\mathrm{TF}}r_s q TF 2 = α TF r s . The weight w D w_D w D depends on the powers of λ \lambda λ carried by the bare lines and on which interactions are marked , that is, differentiated with respect to λ \lambda λ in the coupling-constant integration. For n ≠ 2 n \ne2 n = 2 these powers include the factor λ 2 − n \lambda^{2-n} λ 2 − n from Y D = r s n − 2 Y ^ D Y_{\mathrm{D}} = r_{\mathrm{s}}^{n-2}\widehat{Y}_{\mathrm{D}} Y D = r s n − 2 Y D at density λ r s \lambda r_{\mathrm{s}} λ r s . For the kite, the functional J D \mathcal{J}_{\mathrm{D}} J D is written K \mathcal{K} K and the full-coupling energy F D F_{\mathrm{D}} F D is written F F F in § 3.2 . Since K \mathcal{K} K is linear in the line and n = 2 n = 2 n = 2 , we have K [ W λ ( r s ) ] = λ F ( λ r s ) \mathcal{K}[W_{\lambda}(r_{\mathrm{s}})] = \lambda F(\lambda r_{\mathrm{s}}) K [ W λ ( r s )] = λ F ( λ r s ) . The energy e k i t e = 2 ∫ 0 1 K [ W λ ] d λ e_{\mathrm{kite}} = 2\int_{0}^{1}\mathcal{K}[W_{\lambda}]\,\mathrm{d}\lambda e kite = 2 ∫ 0 1 K [ W λ ] d λ of Eq. (43) is therefore (12) with w D = 2 λ w_{\mathrm{D}} = 2\lambda w D = 2 λ , which satisfies ∫ 0 1 w D d λ = 1 \int_{0}^{1}w_{\mathrm{D}}\,\mathrm{d}\lambda= 1 ∫ 0 1 w D d λ = 1 . The third energy expression sums closed diagrams with polarization chains on the selected line and is treated in § 3.2 . Diagrams with two or more screened lines are outside the present treatment, except for the order counting of Sec. 3.1.
Spectral representation
Let W λ R W_{\lambda}^{R} W λ R be the retarded screened interaction, obtained from W λ W_{\lambda} W λ of (6) by the continuation i ν → ω + i 0 + i\nu\to\omega+ i0^{+} i ν → ω + i 0 + (likewise ϵ λ R \epsilon_{\lambda}^{R} ϵ λ R and Π 0 R \Pi_{0}^{R} Π 0 R below from ϵ λ \epsilon_{\lambda} ϵ λ and Π 0 \Pi_{0} Π 0 of (5) and (7) ). For q > 0 q > 0 q > 0 and real ω > 0 \omega> 0 ω > 0 , its spectral density (loss spectrum)
B W , λ ( q , ω ) = − 1 π Im W λ R ( q , ω ) ≥ 0 , (13) B_{W,\lambda}(q,\omega) = -\frac{1}{\pi}\operatorname{Im} W_{\lambda}^{R}(q,\omega) \ge0, \tag*{(13)} B W , λ ( q , ω ) = − π 1 Im W λ R ( q , ω ) ≥ 0 , (13)
is nonnegative because it is proportional to the power absorbed from a weak external potential, which is nonnegative in a stable ground state (passivity). For the RPA line this follows directly from B W , λ = λ 2 v 2 ∣ ϵ λ R ∣ − 2 [ − π − 1 Im Π 0 R ( q , ω ) ] B_{W,\lambda} = \lambda^{2}v^{2}\lvert\epsilon_{\lambda}^{R}\rvert^{-2}[-\pi^{-1}\operatorname{Im}\Pi_{0}^{R}(q,\omega)] B W , λ = λ 2 v 2 ∣ ϵ λ R ∣ − 2 [ − π − 1 Im Π 0 R ( q , ω )] and Im Π 0 R ( q , ω ) ≤ 0 \operatorname{Im}\Pi_{0}^{R}(q,\omega) \le0 Im Π 0 R ( q , ω ) ≤ 0 for ω > 0 \omega> 0 ω > 0 , as the closed form of the Lindhard function shows [57]. Where an undamped plasmon makes ϵ λ R \epsilon_{\lambda}^{R} ϵ λ R vanish, B W , λ B_{W,\lambda} B W , λ has the delta-function weight λ v / ∂ ω ϵ λ R \lambda v/\partial_{\omega}\epsilon_{\lambda}^{R} λ v / ∂ ω ϵ λ R . This weight is positive because ϵ λ R \epsilon_{\lambda}^{R} ϵ λ R increases with ω \omega ω above the particle–hole continuum. It increases there because the Kramers–Kronig relation and Im Π 0 R ≤ 0 \operatorname{Im}\Pi_{0}^{R} \le0 Im Π 0 R ≤ 0 imply that Re Π 0 R \operatorname{Re}\Pi_{0}^{R} Re Π 0 R decreases with ω \omega ω .
With the high-frequency limit W λ → λ v W_{\lambda} \to\lambda v W λ → λ v [SI Eq. (S54)], the Kramers–Kronig relation [60] on the imaginary axis reads W λ ( q , i ν ) − λ v ( q ) = − ∫ 0 ∞ 2 ω B W , λ ( q , ω ) / ( ω 2 + ν 2 ) d ω W_{\lambda}(q,i\nu)-\lambda v(q) = -\int_{0}^{\infty}2\omega B_{W,\lambda}(q,\omega)/(\omega^{2}+\nu^{2})\,\mathrm{d}\omega W λ ( q , i ν ) − λ v ( q ) = − ∫ 0 ∞ 2 ω B W , λ ( q , ω ) / ( ω 2 + ν 2 ) d ω . The substitution ω = q Λ \omega= q\Lambda ω = q Λ expresses W λ W_{\lambda} W λ as the bare line minus a superposition of Lorentzian lines, each equal to the bare Coulomb factor times a Lorentzian in z z z of width $\Lambda > 0 ({xref}SI Sec. S13.1` and figure 1):
W λ ( q , i ν ) = λ v ( q ) − ∫ 0 ∞ a λ ( q , Λ ) b Λ ( q , i ν ) d Λ , b Λ ( q , i ν ) = 1 q 2 Λ 2 + ( ν / q ) 2 , a λ ( q , Λ ) = 2 q 2 B W , λ ( q , q Λ ) . (14) \begin{aligned}
W_{\lambda}(q,i\nu) &= \lambda v(q) - \int_{0}^{\infty} a_{\lambda}(q,\Lambda)b_{\Lambda}(q,i\nu)\,\mathrm{d}\Lambda,\\
b_{\Lambda}(q,i\nu) &= \frac{1}{q^{2}\Lambda^{2}+(\nu/q)^{2}}, \qquad a_{\lambda}(q,\Lambda) = 2q^{2}B_{W,\lambda}(q,q\Lambda). \tag*{(14)}
\end{aligned} W λ ( q , i ν ) b Λ ( q , i ν ) = λ v ( q ) − ∫ 0 ∞ a λ ( q , Λ ) b Λ ( q , i ν ) d Λ , = q 2 Λ 2 + ( ν / q ) 2 1 , a λ ( q , Λ ) = 2 q 2 B W , λ ( q , q Λ ) . (14)
Here Λ = ω / q \Lambda= \omega/q Λ = ω / q labels the spectral frequency q Λ q\Lambda q Λ . It is the phase velocity of a spectral component in units of the Fermi velocity, not a damping rate. At small q q q the particle–hole continuum has Λ ≲ 1 \Lambda\lesssim1 Λ ≲ 1 , while the plasmon has Λ ≃ ω p / q → ∞ \Lambda\simeq\omega_{\mathrm{p}}/q \to\infty Λ ≃ ω p / q → ∞ , with ω p \omega_{\mathrm{p}} ω p the plasma frequency of (20) below. Each b Λ b_{\Lambda} b Λ is the interaction mediated by a single mode of frequency q Λ q\Lambda q Λ , normalized to the bare line at ν = 0 \nu= 0 ν = 0 , and b Λ → v b_{\Lambda} \to v b Λ → v as Λ → ∞ \Lambda\to\infty Λ → ∞ . The representation holds for any passive screened line with the same high-frequency limit. Spectral structure beyond RPA [61, 62] can therefore be included through its loss spectrum. Unlike plasmon-pole models [63, 64], the representation integrates over all Λ \Lambda Λ and retains the complete spectrum.
Inserting the spectral representation into the linear functional gives
J D [ W λ ] = λ J D [ v ] − ∫ 0 ∞ J D [ a λ ( q , Λ ) b Λ ( q , i ν ) ] d Λ . (15) \mathcal{J}_{\mathrm{D}}[W_{\lambda}] = \lambda\mathcal{J}_{\mathrm{D}}[v] - \int_{0}^{\infty}\mathcal{J}_{\mathrm{D}}[a_{\lambda}(q,\Lambda)b_{\Lambda}(q,i\nu)]\,\mathrm{d}\Lambda. \tag*{(15)} J D [ W λ ] = λ J D [ v ] − ∫ 0 ∞ J D [ a λ ( q , Λ ) b Λ ( q , i ν )] d Λ. (15)
Here the functional acts on the product a λ ( q , Λ ) b Λ ( q , i ν ) a_{\lambda}(q,\Lambda)b_{\Lambda}(q,i\nu) a λ ( q , Λ ) b Λ ( q , i ν ) as a line, a function of q q q and ν \nu ν at fixed Λ \Lambda Λ , so that J D [ a λ ( q , Λ ) b Λ ( q , i ν ) ] = ∫ 0 ∞ a λ ( q , Λ ) e D , Λ ( q ) d q \mathcal{J}_{\mathrm{D}}[a_{\lambda}(q,\Lambda)b_{\Lambda}(q,i\nu)] = \int_{0}^{\infty}a_{\lambda}(q,\Lambda)e_{\mathrm{D},\Lambda}(q)\,\mathrm{d}q J D [ a λ ( q , Λ ) b Λ ( q , i ν )] = ∫ 0 ∞ a λ ( q , Λ ) e D , Λ ( q ) d q with the radial kernel e D , Λ e_{\mathrm{D},\Lambda} e D , Λ of (16) below. The interchange of the spectral and energy integrals is justified whenever ∫ d 3 q d ν v ( q ) ∣ Y D ( q , ν ) ∣ \int\mathrm{d}^{3}q\,\mathrm{d}\nu\,v(q)\lvert Y_{\mathrm{D}}(q,\nu)\rvert ∫ d 3 q d ν v ( q ) ∣ Y D ( q , ν )∣ is finite (SI Sec. S13.4). Since the weight a λ ( q , Λ ) a_{\lambda}(q,\Lambda) a λ ( q , Λ ) depends on q q q , the energy is not J D [ b Λ ] \mathcal{J}_{\mathrm{D}}[b_{\Lambda}] J D [ b Λ ] times a spectral weight.
Figure 1. The RPA screened interaction W λ W_\lambda W λ at coupling λ \lambda λ (double wavy line) as (a) a chain of polarization bubbles and (b) the bare line λ v \lambda v λ v (thin) minus a superposition of Lorentzian lines b Λ b_\Lambda b Λ (thick) with weight a λ ( q , Λ ) ≥ 0 a_\lambda(q,\Lambda) \ge0 a λ ( q , Λ ) ≥ 0 .
Radial kernels and frequency integrals
For a Lorentzian line the frequency integral can be done first. The energy per unit momentum transfer,
J D [ b Λ ] = ∫ 0 ∞ e D , Λ ( q ) d q , e D , Λ ( q ) = 4 π ∫ 0 ∞ d ν π Λ 2 Λ 2 + ν 2 / q 2 Y D ( q , ν ) , (16) \mathcal{J}_D[b_\Lambda] = \int_{0}^{\infty} e_{D,\Lambda}(q)\,\mathrm{d}q,\qquad e_{D,\Lambda}(q) = 4\pi\int_{0}^{\infty} \frac{\mathrm{d}\nu}{\pi}\frac{\Lambda^2}{\Lambda^2+\nu^2/q^2}Y_D(q,\nu), \tag*{(16)} J D [ b Λ ] = ∫ 0 ∞ e D , Λ ( q ) d q , e D , Λ ( q ) = 4 π ∫ 0 ∞ π d ν Λ 2 + ν 2 / q 2 Λ 2 Y D ( q , ν ) , (16)
defines the radial kernel of D D D , with bare-line limit e D , ∞ e_{D,\infty} e D , ∞ (Λ → ∞ \Lambda\to\infty Λ → ∞ , where b Λ → v b_\Lambda\to v b Λ → v ). Here e D , Λ ( q ) e_{D,\Lambda}(q) e D , Λ ( q ) , an energy per unit q q q , is distinct from the coupling average e D ( r s ) e_D(r_s) e D ( r s ) of (12). When Y D Y_D Y D is built from electron–hole factors ( i ν − δ ) − 1 (i\nu-\delta)^{-1} ( i ν − δ ) − 1 with signed excitation energies δ \delta δ , the frequency integrals are elementary. The integrals below take the real part of the product of these factors because the frequency integral runs over ν > 0 \nu> 0 ν > 0 only and the integrand at − ν -\nu − ν is the complex conjugate. With Ω = q Λ \Omega=q\Lambda Ω = q Λ the Lorentzian is Ω 2 / ( Ω 2 + ν 2 ) \Omega^2/(\Omega^2+\nu^2) Ω 2 / ( Ω 2 + ν 2 ) , and for a double pole in ν \nu ν
∫ 0 ∞ d ν π Ω 2 Ω 2 + ν 2 Re ( i ν − δ ) − 2 = Ω 2 ( Ω + ∣ δ ∣ ) 2 , Ω > 0 , δ ≠ 0 , (17) \int_{0}^{\infty}\frac{\mathrm{d}\nu}{\pi}\frac{\Omega^2}{\Omega^2+\nu^2}\operatorname{Re}(i\nu-\delta)^{-2}=\frac{\Omega}{2(\Omega+|\delta|)^2},\qquad\Omega>0,\quad\delta\ne0, \tag*{(17)} ∫ 0 ∞ π d ν Ω 2 + ν 2 Ω 2 Re ( i ν − δ ) − 2 = 2 ( Ω + ∣ δ ∣ ) 2 Ω , Ω > 0 , δ = 0 , (17)
which follows by differentiating the single-pole integral in δ \delta δ (SI Sec. S15.1). SI Sec. S13.2 gives the general closed form for two factors. For a frequency-independent line the frequency integral acts on the fermion denominators alone,
∫ 0 ∞ d ν π Re 1 ( i ν − a ) ( i ν − b ) = sgn ( a b ) − 1 2 ( ∣ a ∣ + ∣ b ∣ ) , (18) \int_{0}^{\infty}\frac{\mathrm{d}\nu}{\pi}\operatorname{Re}\frac{1}{(i\nu-a)(i\nu-b)}=\frac{\operatorname{sgn}(ab)-1}{2(|a|+|b|)}, \tag*{(18)} ∫ 0 ∞ π d ν Re ( i ν − a ) ( i ν − b ) 1 = 2 ( ∣ a ∣ + ∣ b ∣ ) sgn ( ab ) − 1 , (18)
which is the limit of infinite width of the Lorentzian integral (SI Sec. S13.2). Here a a a and b b b are the signed excitation energies of the two electron–hole factors, and equal signs mean that both pairs make transitions in the same direction.
The infinite-width limit vanishes when a a a and b b b have the same sign, whereas the Lorentzian integral of the same pair is positive. In imaginary time each factor ( i ν − δ ) − 1 (i\nu-\delta)^{-1} ( i ν − δ ) − 1 is nonzero for only one sign of the time argument, fixed by the sign of δ \delta δ , and the frequency integral of a product is the time integral of one factor against the other at reversed time. With a static line the two factors therefore give a nonzero integral only if their excitation energies have opposite signs. A line with its own frequency dependence inserts a time dependence between them and removes this restriction. Transitions in the same direction thus contribute to a dynamically screened diagram but not to a statically screened one. This distinction fixes the low-density exponents of § 3.3 and SI Sec. S17.3.
Scaling functions
Table 1 lists the scaling functions of the RPA line as functions of the scaled momentum transfer t = q / ε t = q/\varepsilon t = q / ε , with the density scale ε \varepsilon ε of § 2.1 . At high density ε = q T F \varepsilon= q_{\mathrm{TF}} ε = q TF . At small momentum transfer and fixed z z z the Lindhard function then tends to h d ( z ) h_d(z) h d ( z ) of (9) , so that c ≃ h d ( z ) / t d − 1 c \simeq h_d(z)/t^{d-1} c ≃ h d ( z ) / t d − 1 with t = q / q T F t = q/q_{\mathrm{TF}} t = q / q TF . The full-coupling line satisfies W λ = 1 − v = − v c / ( 1 + c ) W_{\lambda=1}-v=-vc/(1+c) W λ = 1 − v = − v c / ( 1 + c ) , whose Thomas–Fermi tail ∑ N ≥ 1 ( − h d ) N t − N ( d − 1 ) \sum_{N\geq1}(-h_d)^N t^{-N(d-1)} ∑ N ≥ 1 ( − h d ) N t − N ( d − 1 ) gives simple poles at s = N ( d − 1 ) s=N(d-1) s = N ( d − 1 ) (§ 2.7 ). In three dimensions the first of them comes from the t − 2 t^{-2} t − 2 term − h / t 2 -h/t^2 − h / t 2 of this tail. For the coupling average of BRM (§ 3.2 ) the line enters as 2 W ‾ 2\overline{W} 2 W , with W ‾ = ∫ 0 1 W λ d λ \overline{W}=\int_0^1 W_\lambda\,d\lambda W = ∫ 0 1 W λ d λ . With W λ = λ v / ( 1 + λ c ) W_\lambda=\lambda v/(1+\lambda c) W λ = λ v / ( 1 + λ c ) , from (6) and (8) , the integral gives [SI Eq. (S28)]
2 W ‾ − v = v [ ρ ( c ) − 1 ] , ρ ( c ) = 2 [ c − log ( 1 + c ) ] c 2 , (19) 2\overline{W}-v=v[\rho(c)-1],\qquad\rho(c)=\frac{2[c-\log(1+c)]}{c^2}, \tag*{(19)} 2 W − v = v [ ρ ( c ) − 1 ] , ρ ( c ) = c 2 2 [ c − log ( 1 + c )] , (19)
with ρ ( c ) ≃ 1 − 2 3 c \rho(c)\simeq1-\frac{2}{3}c ρ ( c ) ≃ 1 − 3 2 c as c → 0 c\to0 c → 0 .
The low-density analysis is restricted to d = 3 d=3 d = 3 . At low density the scale is
ε = q ∗ = ( 4 q T F 2 3 ) 1 / 4 , ω p = ( q T F 2 3 ) 1 / 2 = q ∗ 2 2 , (20) \varepsilon=q_*=\left(\frac{4q_{\mathrm{TF}}^2}{3}\right)^{1/4},\qquad\omega_p=\left(\frac{q_{\mathrm{TF}}^2}{3}\right)^{1/2}=\frac{q_*^2}{2}, \tag*{(20)} ε = q ∗ = ( 3 4 q TF 2 ) 1/4 , ω p = ( 3 q TF 2 ) 1/2 = 2 q ∗ 2 , (20)
which grows as r s 1 / 4 r_s^{1/4} r s 1/4 in Fermi units, whereas k F q ∗ k_Fq_* k F q ∗ decreases as r s − 3 / 4 r_s^{-3/4} r s − 3/4 . The scale q ∗ q_* q ∗ is the momentum transfer at which the recoil energy q 2 / 2 q^2/2 q 2 /2 equals the plasma frequency ω p \omega_p ω p . The plasma frequency ( 4 π n e e 2 / m e ) 1 / 2 (4\pi n_e e^2/m_e)^{1/2} ( 4 π n e e 2 / m e ) 1/2 is ( q T F 2 / 3 ) 1 / 2 (q_{\mathrm{TF}}^2/3)^{1/2} ( q TF 2 /3 ) 1/2 in units of ω 0 \omega_0 ω 0 . Since at low density ω p ≫ 1 / 2 \omega_p\gg1/2 ω p ≫ 1/2 , the Fermi energy in these units, the relevant momentum transfers are large compared with k F k_F k F . For q ≫ 1 q\gg1 q ≫ 1 the Fermi momenta are negligible, every pair has the energy q 2 / 2 q^2/2 q 2 /2 , and the Lindhard function is L ( q , q 2 u ) ≃ 4 / [ 3 q 2 ( 1 + 4 u 2 ) ] \mathcal{L}(q,q^2u)\simeq4/[3q^2(1+4u^2)] L ( q , q 2 u ) ≃ 4/ [ 3 q 2 ( 1 + 4 u 2 )] with u = ν / q 2 u=\nu/q^2 u = ν / q 2 [SI Eq. (S93)]. This form has a single pole in ν \nu ν , and the screening and bare terms balance at q ∼ q ∗ q\sim q_* q ∼ q ∗ . The RPA line then also has a single pole, at the frequency q Λ λ = 1 2 ( q 4 + λ q ∗ 4 ) 1 / 2 q\Lambda_\lambda=\frac{1}{2}(q^4+\lambda q_*^4)^{1/2} q Λ λ = 2 1 ( q 4 + λ q ∗ 4 ) 1/2 , which interpolates between λ ω p \sqrt{\lambda}\,\omega_p λ ω p for q ≪ q ∗ q\ll q_* q ≪ q ∗ and q 2 / 2 q^2/2 q 2 /2 for q ≫ q ∗ q\gg q_* q ≫ q ∗ . The line is therefore one Lorentzian, whose weight θ λ \theta_\lambda θ λ and scaled width Λ λ / q \Lambda_\lambda/q Λ λ / q depend on q q q only through q / q ∗ q/q_* q / q ∗ :
W λ = λ [ v − θ λ ( q ) b Λ λ ( q ) + v ς λ ] , W_\lambda=\lambda[v-\theta_\lambda(q)b_{\Lambda_\lambda}(q)+v\varsigma_\lambda], W λ = λ [ v − θ λ ( q ) b Λ λ ( q ) + v ς λ ] ,
θ λ ( q ) = λ q ∗ 4 q 4 + λ q ∗ 4 , Λ λ ( q ) = q 2 1 + λ ( q ∗ / q ) 4 , \theta_\lambda(q)=\frac{\lambda q_*^4}{q^4+\lambda q_*^4},\qquad\Lambda_\lambda(q)=\frac{q}{2}\sqrt{1+\lambda(q_*/q)^4}, θ λ ( q ) = q 4 + λ q ∗ 4 λ q ∗ 4 , Λ λ ( q ) = 2 q 1 + λ ( q ∗ / q ) 4 ,
ς λ = O ( q − 2 λ c ^ ( 1 + λ c ^ ) 2 ) , c ^ = ( q ∗ / q ) 4 1 + 4 ν 2 / q 4 (21) \varsigma_\lambda=O\left(q^{-2}\frac{\lambda\hat{c}}{(1+\lambda\hat{c})^2}\right),\qquad\hat{c}=\frac{(q_*/q)^4}{1+4\nu^2/q^4} \tag*{(21)} ς λ = O ( q − 2 ( 1 + λ c ^ ) 2 λ c ^ ) , c ^ = 1 + 4 ν 2 / q 4 ( q ∗ / q ) 4 (21)
Here b Λ λ ( q ) b_{\Lambda_\lambda}(q) b Λ λ ( q ) is the Lorentzian line of (14) with width Λ λ ( q ) \Lambda_\lambda(q) Λ λ ( q ) , c ^ \hat{c} c ^ is the leading large-q q q form of the screening ratio c c c of (8) and ς λ \varsigma_\lambda ς λ the remainder [SI Eqs. (S96) and (S97)]. This remainder is bounded on the imaginary axis, so no undamped plasmon is assumed. The corrections to the large-q q q form of the Lindhard function enter ς λ \varsigma_\lambda ς λ at relative order q − 2 q^{-2} q − 2 , uniformly on the imaginary axis [SI Eq. (S93)].
Mellin convolutions and the endpoint–residue correspondence
After the frequency integration, the energies considered here are built from integrals of the form
I ( ε ) = ∫ 0 ∞ k ( q ) g ( q / ε ) d q , (22) I(\varepsilon)=\int_0^\infty k(q)\,g(q/\varepsilon)\,dq, \tag*{(22)} I ( ε ) = ∫ 0 ∞ k ( q ) g ( q / ε ) d q , (22)
with the radial kernel k k k of the diagram, the scaling function g g g of the line (at low density together with a frequency factor of the diagram), the scaled momentum transfer t = q / ε t=q/\varepsilon t = q / ε , and the scale ε = q T F \varepsilon=q_{\mathrm{TF}} ε = q TF at high density or ε = q ∗ \varepsilon=q_* ε = q ∗ at low density (§ 2.6 ). In the density limits k k k and g g g depend in addition on z z z , which is integrated last. With the Mellin transforms k ~ ( s ) = ∫ 0 ∞ q s − 1 k ( q ) d q \widetilde{k}(s)=\int_0^\infty q^{s-1}k(q)\,\mathrm{d}q k ( s ) = ∫ 0 ∞ q s − 1 k ( q ) d q and g ~ ( s ) \widetilde{g}(s) g ( s ) , defined in the same way, the Mellin–Parseval formula [26 Sec. 3.1.3] gives
I ( ε ) = 1 2 π i ∫ γ − i ∞ γ + i ∞ k ~ ( 1 − s ) g ~ ( s ) ε s d s (23) I(\varepsilon)=\frac{1}{2\pi i}\int_{\gamma-i\infty}^{\gamma+i\infty}\widetilde{k}(1-s)\widetilde{g}(s)\varepsilon^s\,\mathrm{d}s \tag*{(23)} I ( ε ) = 2 π i 1 ∫ γ − i ∞ γ + i ∞ k ( 1 − s ) g ( s ) ε s d s (23)
on a vertical line Re s = γ \operatorname{Re}s=\gamma Re s = γ in the strip where both transforms converge. The argument 1 − s 1-s 1 − s follows from inserting g ( t ) = 1 2 π i ∫ γ − i ∞ γ + i ∞ g ~ ( s ) t − s d s g(t)=\frac{1}{2\pi i}\int_{\gamma-i\infty}^{\gamma+i\infty}\widetilde{g}(s)t^{-s}\,\mathrm{d}s g ( t ) = 2 π i 1 ∫ γ − i ∞ γ + i ∞ g ( s ) t − s d s into (22) .
The asymptotic expansion of I ( ε ) I(\varepsilon) I ( ε ) for small and for large ε \varepsilon ε follows from the endpoint–residue correspondence . This correspondence is the converse mapping between the singular expansion of a Mellin transform (its poles with their principal parts) and the asymptotic expansion of the function [26 , 27 , 28 ] . The approach follows Friot, Greynat and de Rafael [31 ] [65], who read expansions of relativistic Feynman integrals in a ratio of mass scales off the poles of a Mellin transform obtained from a Mellin–Barnes representation. That representation writes a power of a sum as a contour integral over gamma functions. Here the energy is already a Mellin convolution, so no such representation is needed. At high density the small- and large-momentum terms play the roles of the regions of momentum transfer of order k T F k_{\mathrm{TF}} k TF and k F k_{\mathrm{F}} k F , respectively, in the expansion by regions [66, 67]. The usual contour argument requires the analytic continuation of the transforms and their decay along vertical lines. López showed that the same expansion, with a bound on the remainder, follows from the endpoint expansions of the two factors and the convergence of the integral alone [29 ] . We use the correspondence in this form. Its conditions therefore concern only the endpoint behavior of k k k and g g g and the convergence of (22). SI Sec. S14.1 states these conditions in full and proves the correspondence as SI Proposition S3. The proof extends that of Ref. [29 ] to endpoint terms with a factor log m ( 1 / q ) \log^m(1/q) log m ( 1/ q ) .
The correspondence holds when (22) converges absolutely, k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) and g ~ ( s ) \widetilde{g}(s) g ( s ) have a common strip of absolute convergence, k k k admits finite endpoint expansions in terms of the form x α log m x x^\alpha\log^m x x α log m x at 0 0 0 and at ∞ \infty ∞ , and g g g admits one in terms of the form x α x^\alpha x α at the endpoint that generates the expansion (t → ∞ t\to\infty t → ∞ for ε → 0 \varepsilon\to0 ε → 0 and t → 0 t\to0 t → 0 for ε → ∞ \varepsilon\to\infty ε → ∞ ), with remainders of strictly lower order and bounded as in SI Proposition S3, the bounds being integrable in a parameter such as z z z that is integrated after the residues are taken. A small-q q q term q α log m ( 1 / q ) q^{\alpha}\log^{m}(1/q) q α log m ( 1/ q ) of k k k gives k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) a pole of order m + 1 m+1 m + 1 at s = 1 + α s=1+\alpha s = 1 + α , a large-q q q term q − α ′ q^{-\alpha'} q − α ′ a simple pole at s = 1 − α ′ s=1-\alpha' s = 1 − α ′ . A large-t t t term t − β t^{-\beta} t − β of g g g gives g ~ \widetilde{g} g a simple pole at s = β s=\beta s = β , a small-t t t term t β ′ t^{\beta'} t β ′ one at s = − β ′ s=-\beta' s = − β ′ . As ε → 0 \varepsilon\to0 ε → 0 , I ( ε ) I(\varepsilon) I ( ε ) is minus the sum of the residues of k ~ ( 1 − s ) g ~ ( s ) ε s \widetilde{k}(1-s)\widetilde{g}(s)\varepsilon^{s} k ( 1 − s ) g ( s ) ε s at the poles to the right of the strip, in increasing order, and as ε → ∞ \varepsilon\to\infty ε → ∞ the sum of the residues at the poles to its left, in decreasing order. A pole of total order N N N at s 0 s_0 s 0 contributes ε s 0 \varepsilon^{s_0} ε s 0 times a polynomial of degree N − 1 N-1 N − 1 in log ε \log\varepsilon log ε . The remainder after the poles up to the line Re s = σ \operatorname{Re}s=\sigma Re s = σ is bounded by ε σ \varepsilon^{\sigma} ε σ times a power of 1 + ∣ log ε ∣ 1+|\log\varepsilon| 1 + ∣ log ε ∣ , and by a constant times ε σ \varepsilon^{\sigma} ε σ if the endpoint bounds extend beyond that line in the direction of the expansion with no further pole. SI Sec. S14.6 writes out the residue at a pole of higher order, a formula used repeatedly below.
For the first-order exchange diagrams the remainder is also bounded explicitly (SI Secs. S10, S17 and S18). At low density the width Λ λ ( q ) \Lambda_{\lambda}(q) Λ λ ( q ) of the single Lorentzian that represents the screened line (§ 2.6 ) enters the kernel. The energy therefore has the form (22) only after the large-momentum kernel is separated. The low-density statements proved here rest on these direct bounds.
By the endpoint–residue correspondence, each term of a density expansion is the residue of one pole, generated by one endpoint term of the kernel or of the scaling function, or by endpoint terms of both when their poles coincide. The leading exponent therefore follows from a comparison of pole positions before any coefficient is computed. Corollaries 1 and 2 below follow from the correspondence by such comparisons. Corollary 1 makes the comparison for the RPA scaling functions of § 2.6 .
For a line f f f with relative change g = ( f − v ) / v g=(f-v)/v g = ( f − v ) / v , inserting d 3 q = 4 π q 2 d q d^{3}q=4\pi q^{2}\,dq d 3 q = 4 π q 2 d q and d ν = q d z d\nu=q\,dz d ν = q d z into (11) gives the energy change in terms of a radial kernel,
J D [ f ] − J D [ v ] = ∫ 0 ∞ d z π ∫ 0 ∞ k D ( q ; z ) g ( q , i q z ) d q , k D ( q ; z ) = 4 π q Y D ( q , q z ) . (24) \mathcal{J}_{D}[f]-\mathcal{J}_{D}[v]=\int_{0}^{\infty}\frac{dz}{\pi}\int_{0}^{\infty}k_{D}(q;z)g(q,iqz)\,dq,\qquad k_{D}(q;z)=4\pi qY_{D}(q,qz). \tag*{(24)} J D [ f ] − J D [ v ] = ∫ 0 ∞ π d z ∫ 0 ∞ k D ( q ; z ) g ( q , i q z ) d q , k D ( q ; z ) = 4 π q Y D ( q , q z ) . (24)
In the density limits g ( q , i q z ) g(q,iqz) g ( q , i q z ) depends on t = q / ε t=q/\varepsilon t = q / ε and z z z alone (Table 1 ), and at high density it is the scaling function of (22). At high density the kernel k k k of (22) is this z z z -resolved kernel, whose Lorentzian average ∫ 0 ∞ ( d z / π ) Λ 2 k D ( q ; z ) / ( Λ 2 + z 2 ) \int_{0}^{\infty}(dz/\pi)\,\Lambda^{2}k_{D}(q;z)/(\Lambda^{2}+z^{2}) ∫ 0 ∞ ( d z / π ) Λ 2 k D ( q ; z ) / ( Λ 2 + z 2 ) is e D , Λ ( q ) e_{D,\Lambda}(q) e D , Λ ( q ) of (16). At low density k k k is e D , Λ ( q ) e_{D,\Lambda}(q) e D , Λ ( q ) at fixed scaled width w Λ = Λ / q w_{\Lambda}=\Lambda/q w Λ = Λ/ q , since Λ λ / q \Lambda_{\lambda}/q Λ λ / q depends on q q q only through q / q ∗ q/q_{*} q / q ∗ (§ 2.6 ). The exponents α 0 \alpha_{0} α 0 , m m m and α \alpha α in Corollaries 1 and 2 refer to these kernels, not to the bare-line kernel e D , ∞ e_{D,\infty} e D , ∞ . For the kite, with A 0 A_{0} A 0 and X X X of (40) and (41) in § 3.2 , k D ( q ; z ) = 4 π A 0 q X ( q , q z ) ≃ κ D ( z ) q log ( 1 / q ) k_{D}(q;z)=4\pi A_{0}qX(q,qz)\simeq\kappa_{D}(z)q\log(1/q) k D ( q ; z ) = 4 π A 0 q X ( q , q z ) ≃ κ D ( z ) q log ( 1/ q ) with κ D = 4 π A 0 a x \kappa_{D}=4\pi A_{0}a_{x} κ D = 4 π A 0 a x by (47), and e D , Λ ∝ q − 2 e_{D,\Lambda}\propto q^{-2} e D , Λ ∝ q − 2 at fixed w Λ w_{\Lambda} w Λ by (48). The bare-line kernel e D , ∞ e_{D,\infty} e D , ∞ has no term q log ( 1 / q ) q\log(1/q) q log ( 1/ q ) , because ∫ 0 ∞ a x d z = 0 \int_{0}^{\infty}a_{x}\,dz=0 ∫ 0 ∞ a x d z = 0 [κ ( Λ ) \kappa(\Lambda) κ ( Λ ) , the coefficient of q log ( 1 / q ) q\log(1/q) q log ( 1/ q ) in e D , Λ e_{D,\Lambda} e D , Λ for the kite, vanishes at Λ = ∞ \Lambda=\infty Λ = ∞ , SI Eq. (S26)]. The bare-line kernel decays as q − 4 q^{-4} q − 4 at large momentum transfer. For q > 2 q>2 q > 2 the regions in which both loop momenta lie in the same one of the two balls ∣ p ∣ < 1 |p|<1 ∣ p ∣ < 1 and ∣ p + q ∣ < 1 |p+q|<1 ∣ p + q ∣ < 1 give no contribution for the bare line by (18). The other two regions are of order q − 4 q^{-4} q − 4 (SI Sec. S17.1).
Table 1. Scaling functions of the RPA line, with t = q / ε t=q/\varepsilon t = q / ε and h d ( z ) h_d(z) h d ( z ) of (9) (h = h 3 h=h_3 h = h 3 ). The first row holds in d d d dimensions, the others for d = 3 d=3 d = 3 . At low density the table gives the weight and the scaled width of the single Lorentzian. The scaling function of a diagram is then θ λ \theta_\lambda θ λ times its frequency factor evaluated at Λ λ / q \Lambda_\lambda/q Λ λ / q , e.g. θ λ p l o c \theta_\lambda p_{\mathrm{loc}} θ λ p loc for the kite, with the frequency factor p l o c p_{\mathrm{loc}} p loc of (48). At high density the Thomas–Fermi tails give the universal poles s = N ( d − 1 ) s=N(d-1) s = N ( d − 1 ) (s = 2 s=2 s = 2 for d = 3 d=3 d = 3 ). At low density the large-width term q / Λ λ ∝ t 2 q/\Lambda_\lambda\propto t^2 q / Λ λ ∝ t 2 gives the universal pole s = − 2 s=-2 s = − 2 , because the radial kernel e D , Λ e_{D,\Lambda} e D , Λ of a diagram whose Y D Y_D Y D has the high-frequency tail A D ( q ) / ν 2 A_D(q)/\nu^2 A D ( q ) / ν 2 (Corollary 1) differs from e D , ∞ e_{D,\infty} e D , ∞ by − 2 π A D / ( q Λ ) -2\pi A_D/(q\Lambda) − 2 π A D / ( q Λ ) at large width [SI Eq. (S75)].
Corollary 1 (Exponent rules for RPA screening). Let the radial kernel of D D D in d d d dimensions behave as κ 0 q α 0 log m ( 1 / q ) \kappa_{0}q^{\alpha_{0}}\log^{m}(1/q) κ 0 q α 0 log m ( 1/ q ) for q → 0 q\to0 q → 0 and as κ ∞ q − α \kappa_{\infty}q^{-\alpha} κ ∞ q − α for q → ∞ q\to\infty q → ∞ , let the integrand Y D Y_{D} Y D have the high-frequency tail Y D ≃ A D ( q ) / ν 2 Y_{D}\simeq A_{D}(q)/\nu^{2} Y D ≃ A D ( q ) / ν 2 , and assume that the residues named below are nonzero. Take the RPA scaling functions of § 2.6 , whose high-density tail starts with t − ( d − 1 ) t^{-(d-1)} t − ( d − 1 ) and whose low-density expansion for d = 3 d=3 d = 3 proceeds in powers of t 2 t^{2} t 2 , starting with the large-width term. With the overall factor r s n − 2 r_{s}^{n-2} r s n − 2 of § 2.3 multiplying all orders stated below, the following holds. Here
and in Corollary 2, W W W denotes the screened line at the coupling of the energy expression, W λ = 1 W_{\lambda=1} W λ = 1 unless stated otherwise.
At high density the screening correction J D [ W ] − J D [ v ] \mathcal{J}_{D}[W]-\mathcal{J}_{D}[v] J D [ W ] − J D [ v ] scales as q T F min ( 1 + α 0 , d − 1 ) q_{\mathrm{TF}}^{\min(1+\alpha_{0},d-1)} q TF m i n ( 1 + α 0 , d − 1 ) , up to powers of log q T F \log q_{\mathrm{TF}} log q TF . For d = 3 d=3 d = 3 and α 0 = 1 \alpha_{0}=1 α 0 = 1 the two poles coincide at s = 2 s=2 s = 2 and the correction is q T F 2 log m + 1 q T F ∝ r s log m + 1 r s q_{\mathrm{TF}}^{2}\log^{m+1}q_{\mathrm{TF}}\propto r_{s}\log^{m+1}r_{s} q TF 2 log m + 1 q TF ∝ r s log m + 1 r s .
At low density, for d = 3 d=3 d = 3 , J D [ W ] \mathcal{J}_{D}[W] J D [ W ] scales as q ∗ − min ( α − 1 , 2 ) q_{*}^{-\min(\alpha-1,2)} q ∗ − m i n ( α − 1 , 2 ) , with log q ∗ \log q_{*} log q ∗ when α = 3 \alpha=3 α = 3 . For α > 3 \alpha>3 α > 3 the leading term comes from the screening pole at s = − 2 s=-2 s = − 2 ,
J D [ W λ = 1 ] ≃ 4 π q ∗ 2 ∫ 0 ∞ A D ( q ) d q , (25) \mathcal{J}_{D}[W_{\lambda=1}]\simeq\frac{4\pi}{q_{*}^{2}}\int_{0}^{\infty} A_{D}(q)\,\mathrm{d}q, \tag*{(25)} J D [ W λ = 1 ] ≃ q ∗ 2 4 π ∫ 0 ∞ A D ( q ) d q , (25)
which depends on the diagram only through its high-frequency tail, provided ∫ 0 ∞ A D d q \int_{0}^{\infty}A_{D}\,\mathrm{d}q ∫ 0 ∞ A D d q converges and the remainder of the expansion of the radial kernel in 1 / Λ 1/\Lambda 1/Λ is integrable uniformly in q ∗ q_{*} q ∗ (SI Sec. S14.3).
The corollary follows from the endpoint–residue correspondence (SI Proposition S3) by comparing the positions of the first poles of the kernel and of the scaling function on either side of the strip (proof in SI Sec. S14.2). The corollary assumes the dynamically screened RPA line. A static line has infinite width, so its expansion has no term of order q / Λ q/\Lambda q /Λ . The low-density energy of the kite with a static line is at most of order r s − 3 / 4 r_{s}^{-3/4} r s − 3/4 (SI Sec. S17.3).
Corollary 2 applies the first statement of Corollary 1 to a kernel with the small-q q q endpoint term q log m ( 1 / q ) q\log^{m}(1/q) q log m ( 1/ q ) , that is, α 0 = 1 \alpha_{0}=1 α 0 = 1 . With the overall factor r s n − 2 r_{s}^{n-2} r s n − 2 it turns that statement into a counting rule for the powers of log r s \log r_{s} log r s .
Corollary 2 (Powers of log r s \log r_{s} log r s at high density in d d d dimensions). Let a diagram D D D with n n n bare interactions have a z z z -resolved radial kernel k D ( q ; z ) k_{D}(q;z) k D ( q ; z ) of Eq. (24) behaving as r s n − 2 κ D ( z ) q log m ( 1 / q ) r_{s}^{n-2}\kappa_{D}(z)q\log^{m}(1/q) r s n − 2 κ D ( z ) q log m ( 1/ q ) at small momentum transfer, with κ D \kappa_{D} κ D independent of the density, and let the scaling function have the Thomas–Fermi tail ∑ N ≥ 1 γ N ( z ) t − N ( d − 1 ) \sum_{N\geq1}\gamma_{N}(z)t^{-N(d-1)} ∑ N ≥ 1 γ N ( z ) t − N ( d − 1 ) [γ N = ( − h d ) N \gamma_{N}=(-h_{d})^{N} γ N = ( − h d ) N at full coupling]. Then the kernel pole at s = 2 s=2 s = 2 contributes a term r s n − 2 + 2 / ( d − 1 ) log m r s r_{s}^{n-2+2/(d-1)}\log^{m}r_{s} r s n − 2 + 2/ ( d − 1 ) log m r s to J D [ W ] − J D [ v ] \mathcal{J}_{D}[W]-\mathcal{J}_{D}[v] J D [ W ] − J D [ v ] if 2 / ( d − 1 ) 2/(d-1) 2/ ( d − 1 ) is not an integer. This difference is the single-chain resummation, the sum of all diagrams with a chain of free bubbles inserted into the selected line, each with the coefficient 1 / S D 1/S_{D} 1/ S D of D D D . At the resonant dimensions d = 1 + 2 / N ∗ d=1+2/N_{*} d = 1 + 2/ N ∗ , N ∗ = 1 , 2 , 3 , … N_{*}=1,2,3,\ldots N ∗ = 1 , 2 , 3 , … , the kernel pole coincides with the screening pole N = N ∗ N=N_{*} N = N ∗ and contributes r s n − 2 + N ∗ log m + 1 r s r_{s}^{n-2+N_{*}}\log^{m+1}r_{s} r s n − 2 + N ∗ log m + 1 r s with a coefficient proportional to ∫ 0 ∞ κ D ( z ) γ N ∗ ( z ) d z / π \int_{0}^{\infty}\kappa_{D}(z)\gamma_{N_{*}}(z)\,\mathrm{d}z/\pi ∫ 0 ∞ κ D ( z ) γ N ∗ ( z ) d z / π . The screening poles N ( d − 1 ) < 2 N(d-1)<2 N ( d − 1 ) < 2 contribute r s n − 2 + N r_{s}^{n-2+N} r s n − 2 + N without log r s \log r_{s} log r s . In three dimensions N ∗ = 1 N_{*}=1 N ∗ = 1 , the term is r s n − 1 log m + 1 r s r_{s}^{n-1}\log^{m+1}r_{s} r s n − 1 log m + 1 r s , and γ 1 = − 2 h / 3 \gamma_{1}=-2h/3 γ 1 = − 2 h /3 for the coupling average and − h -h − h at full coupling (Table 1 ). For a set of diagrams whose factors log ( 1 / q ) \log(1/q) log ( 1/ q ) cancel, such as vertex and self-energy corrections related by the Ward identity, the corollary applies to the summed kernel, and the power of log r s \log r_{s} log r s is set by the factors log ( 1 / q ) \log(1/q) log ( 1/ q ) that survive in the sum.
The proof is given in SI Sec. S14.2. Corollary 2 determines the leading power of log r s \log r_{s} log r s in the screening correction of a diagram, but it does not classify all contributions at the same order of the full high-density expansion.
Both kinds of universal pole of the scaling function are tied to conservation laws. The order of a high-density pole at a resonance depends on the Ward identity, which fixes the endpoint of the kernel that the pole meets. The low-density pole is constrained by the f f f -sum rule. § 3.4 shows how each law enters the correlation energy.
The high-density poles at s = N ( d − 1 ) s=N(d-1) s = N ( d − 1 ) come from the Thomas–Fermi tail c ≃ q T F d − 1 h d ( z ) / q d − 1 c \simeq q_{\mathrm{TF}}^{d-1}h_d(z)/q^{d-1} c ≃ q TF d − 1 h d ( z ) / q d − 1 of the long-range screening, with c c c and h d h_d h d of (8) and (9) (Sec. 2.6). The kernel of any polarization insertion has the small-q q q endpoint k ∝ q k \propto q k ∝ q . At fixed z z z the electron–hole factors stay finite as q → 0 q \to0 q → 0 , since both the shell volume and the excitation energies are of order q q q . The measure d ν = q d z \mathrm{d}\nu=q\,\mathrm{d}z d ν = q d z supplies one factor q q q , and the Coulomb line q 1 − d q^{1-d} q 1 − d compensates the factor q d − 1 q^{d-1} q d − 1 of d d q \mathrm{d}^d q d d q . The pole of this endpoint at s = 2 s=2 s = 2 coincides with a screening pole when N ( d − 1 ) = 2 N(d-1)=2 N ( d − 1 ) = 2 . This coincidence produces the terms of Corollary 2, such as the Carr–Maradudin part 0.01304 r s log r s R y 0.01304\,r_s\log r_s\ \mathrm{Ry} 0.01304 r s log r s Ry in three dimensions (Sec. 2.9) and the r s 2 log r s r_s^2\log r_s r s 2 log r s term in two (Sec. 3.1). The ring kernel carries no Coulomb line and has the endpoint q d q^d q d , whose pole at s = d + 1 s=d+1 s = d + 1 coincides with a screening pole when N ( d − 1 ) = d + 1 N(d-1)=d+1 N ( d − 1 ) = d + 1 . This coincidence gives the Gell-Mann–Brueckner term in three dimensions and the Rajagopal–Kimball term in two (Sec. 2.8). The Ward identity of charge conservation constrains the density and current vertices jointly through the inverse propagator. For the first-order exchange polarization considered here, the explicit vertex and self-energy expressions give equal and opposite terms with the Fermi-surface factor log ( 1 / q ) \log(1/q) log ( 1/ q ) , which cancel in the sum (Sec. 3.4). At higher orders the Ward identity still constrains vertex and self-energy jointly, but the surviving power of log r s \log r_s log r s must be determined order by order.
In three dimensions the low-density pole at s = − 2 s=-2 s = − 2 comes from the large-width term of the scaling function. In Corollary 1 it competes with the pole at s = 1 − α s=1-\alpha s = 1 − α from the large-q q q term q − α q^{-\alpha} q − α of the kernel, and the leading term comes from whichever lies closer to the strip. To trace the origin of (25), we use the low-density line of Sec. 2.6, which for q ≪ q ∗ q \ll q_* q ≪ q ∗ is W ≃ v ν 2 / ( ν 2 + ω p 2 ) W \simeq v\,\nu^2/(\nu^2+\omega_p^2) W ≃ v ν 2 / ( ν 2 + ω p 2 ) , with ω p = q ∗ 2 / 2 \omega_p=q_*^2/2 ω p = q ∗ 2 /2 the plasma frequency [(20)]. The line is small below ω p \omega_p ω p and close to v v v above it, so the diagram is probed only where Y D ≃ A D / ν 2 Y_D \simeq A_D/\nu^2 Y D ≃ A D / ν 2 . The energy per unit q q q is then 4 π ∫ 0 ∞ d ν π A D ( q ) ν 2 + ω p 2 = 4 π A D ( q ) / q ∗ 2 4\pi\int_0^\infty\frac{\mathrm{d}\nu}{\pi}\frac{A_D(q)}{\nu^2+\omega_p^2}=4\pi A_D(q)/q_*^2 4 π ∫ 0 ∞ π d ν ν 2 + ω p 2 A D ( q ) = 4 π A D ( q ) / q ∗ 2 . The residue at s = − 2 s=-2 s = − 2 is thus fixed by the high-frequency tail A D A_D A D , the contribution of D D D to the first frequency moment of the density response. At large ν \nu ν the exact response decays as − n e q 2 / ( m e ν 2 ) -n_{\mathrm{e}}q^2/(m_{\mathrm{e}}\nu^2) − n e q 2 / ( m e ν 2 ) in physical units, that is, as − q T F 2 q 2 / ( 3 ν 2 ) -q_{\mathrm{TF}}^2q^2/(3\nu^2) − q TF 2 q 2 / ( 3 ν 2 ) in the units of (4). The particle number fixes this coefficient for every interaction strength [the f f f -sum rule, (56) below]. The proper response Π \Pi Π has the same tail, since χ − Π = O ( ν − 4 ) \chi-\Pi=O(\nu^{-4}) χ − Π = O ( ν − 4 ) . In the sum of all corrections to Π \Pi Π of a given order the ν − 2 \nu^{-2} ν − 2 terms should therefore cancel.
Under an additional remainder condition, this cancellation of the tails also removes the residue at s = − 2 s=-2 s = − 2 (SI Lemma S1, proved in SI Sec. S14.4). For a set S \mathcal{S} S of diagrams closed with the same screened line and the same coupling weight, suppose that the summed integrand Y S = ∑ D ∈ S Y D Y_{\mathcal{S}}=\sum_{D\in\mathcal{S}}Y_D Y S = ∑ D ∈ S Y D has the high-frequency tail Y S ≃ A S ( q ) / ν 2 Y_{\mathcal{S}}\simeq A_{\mathcal{S}}(q)/\nu^2 Y S ≃ A S ( q ) / ν 2 . If A S ( q ) = 0 A_{\mathcal{S}}(q)=0 A S ( q ) = 0 for every q q q and the summed radial kernel satisfies the remainder condition of Corollary 1, the pole at s = − 2 s=-2 s = − 2 is absent from ∑ D ∈ S I D [ W ] \sum_{D\in\mathcal{S}}\mathcal{I}_D[W] ∑ D ∈ S I D [ W ] . The condition A S = 0 A_{\mathcal{S}}=0 A S = 0 holds for the first-order exchange corrections to the proper polarization at fixed density. These conditions concern the summed integrand and radial kernel, and the individual diagrams need not satisfy them. For the kite alone A D ( q ) A_D(q) A D ( q ) tends to the nonzero constant − 3 / ( 4 π 3 ) -3/(4\pi^3) − 3/ ( 4 π 3 ) at large q q q (SI Sec. S14.3), so ∫ 0 ∞ A D d q \int_0^\infty A_D\,\mathrm{d}q ∫ 0 ∞ A D d q diverges, consistent with the q − 2 q^{-2} q − 2 endpoint of its kernel (Sec. 3.2). The leading low-density term of the kite is therefore the kernel pole at s = − 1 s=-1 s = − 1 (α = 2 \alpha=2 α = 2 in Corollary 1), not the universal pole. The high-frequency tail vanishes only for the sum with the Fock self-energy corrections. The f f f -sum-rule argument for complete sets of corrections of higher order is given in SI Sec. S14.4, after the proof of SI Lemma S1. At higher order this argument assumes that the coupling expansion of the exact fixed-density response may be taken term by term at large frequency. Individual diagrams with bare internal lines need not have finite tails. Poles generated by the kernel at large momentum transfer obey no such general rule. They depend on the occupation restrictions and Coulomb factors of the loop momenta, as in the kite, where they come from transitions with both loop momenta in the same Fermi ball (§ 3.2 ).
The coupling averages (12) are Mellin convolutions in r s r_s r s . A term φ r s p \varphi r_s^p φ r s p of F D F_D F D , with constant φ \varphi φ , appears in e D e_D e D multiplied by the Mellin multiplier M ( p ) = ∫ 0 1 w D ( λ ) λ p d λ M(p) = \int_0^1 w_D(\lambda)\lambda^p\,\mathrm{d}\lambda M ( p ) = ∫ 0 1 w D ( λ ) λ p d λ , provided this moment converges (p > − 2 p > -2 p > − 2 for w D = 2 λ w_D = 2\lambda w D = 2 λ ). The same multiplier applies to the highest power of log r s \log r_s log r s in φ r s p log m r s \varphi r_s^p\log^m r_s φ r s p log m r s (SI Sec. S14.5 gives the lower powers). At low density the couplings λ ≲ 1 / r s \lambda\lesssim1/r_s λ ≲ 1/ r s , for which λ r s ≲ 1 \lambda r_s \lesssim1 λ r s ≲ 1 and F D F_D F D is not in its low-density form, add a term of order r s − 2 r_s^{-2} r s − 2 for w D = 2 λ w_D = 2\lambda w D = 2 λ . This term is smaller than all low-density terms obtained here.
Diagrams that enter one physical quantity together, such as the vertex and self-energy corrections of a response, are closed with the same line and weight. If their radial kernels have equal and opposite endpoint terms, the corresponding poles cancel in the sum for every common scaling function, and the sum starts at the next pole. At first order the endpoint terms cancel at both endpoints of the kernel (§ 3.4 ).
Ring series in d d d dimensions
The known log r s \log r_s log r s terms of (1) and (2) are residues of coincident poles in the sense of the endpoint–residue correspondence (§ 2.7 , SI Proposition S3). They were originally obtained by dedicated partial summations. Here each follows from two endpoint terms, one of the kernel and one of the scaling function, and the dimension enters only through the positions of the poles (Figure 2 ). The statements of § 2.8 and § 2.9 concern kernels without Fermi-surface factors log ( 1 / q ) \log(1/q) log ( 1/ q ) , either because the diagram has none, as for the rings, or because the factors log ( 1 / q ) \log(1/q) log ( 1/ q ) of the vertex and self-energy corrections cancel, as the Ward identity requires (Corollary 2). The screening poles lie at the integer multiples of d − 1 d - 1 d − 1 , and the kernel poles at s = d + 1 s = d + 1 s = d + 1 for the rings and at s = 2 s = 2 s = 2 for a polarization insertion. As d d d varies, the lines s = N ( d − 1 ) s = N(d - 1) s = N ( d − 1 ) cross the kernel poles at isolated dimensions, and only there does a log r s \log r_s log r s term appear.
The RPA correlation energy depends nonlinearly on the line, through the scalar function log ( 1 + c ) − c \log(1+c)-c log ( 1 + c ) − c of the screening ratio c c c of (8) . It is still an integral of a density-independent kernel against a function of the scaled momentum transfer. This integral follows from the d d d -dimensional form of the adiabatic-connection fluctuation–dissipation formula (53) of § 3.4 with the RPA response χ λ = Π 0 / ϵ λ \chi_\lambda= \Pi_0/\epsilon_\lambda χ λ = Π 0 / ϵ λ , whose coupling integral gives − [ log ( 1 + c ) − c ] -[\log(1+c)-c] − [ log ( 1 + c ) − c ] . In the units of § 2.2 , with the volume ( 2 π ) d / ( 2 V d ) (2\pi)^d/(2\mathcal{V}_d) ( 2 π ) d / ( 2 V d ) per electron, A d = d V d \mathcal{A}_d=d\mathcal{V}_d A d = d V d and ∫ 0 ∞ d ν = q 2 ∫ − ∞ ∞ d z \int_0^\infty\mathrm{d}\nu=\frac{q}{2}\int_{-\infty}^{\infty}\mathrm{d}z ∫ 0 ∞ d ν = 2 q ∫ − ∞ ∞ d z (SI Sec. S2), the RPA correlation energy per electron is
e R P A = d 8 π E 0 ∫ − ∞ ∞ d z ∫ 0 ∞ d q q d [ log ( 1 + c ) − c ] , (26) e_{\mathrm{RPA}}=\frac{d}{8\pi}E_0\int_{-\infty}^{\infty}\mathrm{d}z\int_0^\infty\mathrm{d}q\,q^d[\log(1+c)-c], \tag*{(26)} e RPA = 8 π d E 0 ∫ − ∞ ∞ d z ∫ 0 ∞ d q q d [ log ( 1 + c ) − c ] , (26)
with c c c of (8) at ν = q z \nu=qz ν = q z and E 0 E_0 E 0 in Ry. For d = 3 d=3 d = 3 , E 0 = 32 / ( π 2 q T F 4 ) E_0=32/(\pi^2q_{\mathrm{TF}}^4) E 0 = 32/ ( π 2 q TF 4 ) Ry and (26) is Eq. (20) of Ref. [68]. With ν = q z \nu=qz ν = q z , one factor q q q in (26) is the Jacobian of the frequency integral. Together with d d q = A d q d − 1 d q \mathrm{d}^dq=\mathcal{A}_dq^{d-1}\mathrm{d}q d d q = A d q d − 1 d q , this Jacobian makes the kernel at small momentum transfer the phase-space factor q d q^d q d . This small-q q q endpoint term gives the Mellin transform k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) a simple pole at s = d + 1 s=d+1 s = d + 1 . With c ≃ h d ( z ) / t d − 1 c\simeq h_d(z)/t^{d-1} c ≃ h d ( z ) / t d − 1 and t = q / q T F t=q/q_{\mathrm{TF}} t = q / q TF (Table 1 ), the scaling function log ( 1 + h d t 1 − d ) − h d t 1 − d \log(1+h_dt^{1-d})-h_dt^{1-d} log ( 1 + h d t 1 − d ) − h d t 1 − d has the Thomas–Fermi tail ∑ N ≥ 2 ( − 1 ) N + 1 h d N t N ( 1 − d ) / N \sum_{N\ge2}(-1)^{N+1}h_d^Nt^{N(1-d)}/N ∑ N ≥ 2 ( − 1 ) N + 1 h d N t N ( 1 − d ) / N , which gives simple poles at s = N ( d − 1 ) s=N(d-1) s = N ( d − 1 ) . The kernel pole coincides with one of them when
N r = d + 1 d − 1 ∈ { 2 , 3 , 4 , … } , d = N r + 1 N r − 1 = 3 , 2 , 5 3 , 3 2 , … (27) N_r=\frac{d+1}{d-1}\in\{2,3,4,\ldots\},\qquad d=\frac{N_r+1}{N_r-1}=3,2,\frac{5}{3},\frac{3}{2},\ldots \tag*{(27)} N r = d − 1 d + 1 ∈ { 2 , 3 , 4 , … } , d = N r − 1 N r + 1 = 3 , 2 , 3 5 , 2 3 , … (27)
At these resonant dimensions, the same set as in Corollary 2, the double pole gives, with log q T F = log r s / ( d − 1 ) + O ( 1 ) \log q_{\mathrm{TF}} = \log r_s/(d - 1) + O(1) log q TF = log r s / ( d − 1 ) + O ( 1 ) from (7) (SI Sec. S2),
e R P A ⊃ d 8 π E 0 q T F d + 1 ( − 1 ) N r N r ( d − 1 ) ∫ − ∞ ∞ h d ( z ) N r d z log r s , E 0 q T F d + 1 ∝ r s ( 3 − d ) / ( d − 1 ) . (28) e_{\mathrm{RPA}} \supset\frac{d}{8\pi} E_0 q_{\mathrm{TF}}^{d+1} \frac{(-1)^{N_r}}{N_r(d - 1)} \int_{-\infty}^{\infty} h_d(z)^{N_r}\,\mathrm{d}z \log r_s,\qquad E_0 q_{\mathrm{TF}}^{d+1} \propto r_s^{(3-d)/(d-1)}. \tag*{(28)} e RPA ⊃ 8 π d E 0 q TF d + 1 N r ( d − 1 ) ( − 1 ) N r ∫ − ∞ ∞ h d ( z ) N r d z log r s , E 0 q TF d + 1 ∝ r s ( 3 − d ) / ( d − 1 ) . (28)
Here and below, A ⊃ B A \supset B A ⊃ B means that the density expansion of A A A contains the term B B B . With h d h_d h d in place of the Lindhard function, the q q q integral diverges at large q q q , and its resonant term has an integrand ∝ 1 / q \propto1/q ∝ 1/ q . The Lindhard function cuts it off at q ∼ 1 q \sim1 q ∼ 1 and changes only the term without log r s \log r_s log r s . The coefficient of log r s \log r_s log r s therefore comes from the window q T F ≪ q ≪ 1 q_{\mathrm{TF}} \ll q \ll1 q TF ≪ q ≪ 1 (SI Sec. S14.6).
In three dimensions N r = 2 N_r = 2 N r = 2 , the prefactor is 3 / π 3 3/\pi^3 3/ π 3 , and as r s → 0 r_s \to0 r s → 0 [SI (141) ]
e R P A = 3 π 3 ∫ − ∞ ∞ h ( z ) 2 d z log r s + O ( 1 ) = 2 π 2 ( 1 − log 2 ) log r s + O ( 1 ) , (29) e_{\mathrm{RPA}} = \frac{3}{\pi^3}\int_{-\infty}^{\infty} h(z)^2\,\mathrm{d}z \log r_s + O(1) = \frac{2}{\pi^2}(1-\log2)\log r_s + O(1), \tag*{(29)} e RPA = π 3 3 ∫ − ∞ ∞ h ( z ) 2 d z log r s + O ( 1 ) = π 2 2 ( 1 − log 2 ) log r s + O ( 1 ) , (29)
which is the Gell-Mann–Brueckner coefficient A = 0.06218 A = 0.06218 A = 0.06218 of (1) [5 ] . In two dimensions N r = 3 N_r = 3 N r = 3 , E 0 q T F 3 = 8 2 r s E_0q_{\mathrm{TF}}^3 = 8\sqrt{2}r_s E 0 q TF 3 = 8 2 r s Ry, and
e R P A ⊃ − 2 2 3 π ∫ − ∞ ∞ h 2 ( z ) 3 d z r s log r s = − 2 2 ( 10 3 π − 1 ) r s log r s , (30) e_{\mathrm{RPA}} \supset-\frac{2\sqrt{2}}{3\pi}\int_{-\infty}^{\infty} h_2(z)^3\,\mathrm{d}z\ r_s\log r_s = -2\sqrt{2}\left(\frac{10}{3\pi}-1\right)r_s\log r_s, \tag*{(30)} e RPA ⊃ − 3 π 2 2 ∫ − ∞ ∞ h 2 ( z ) 3 d z r s log r s = − 2 2 ( 3 π 10 − 1 ) r s log r s , (30)
with ∫ − ∞ ∞ h 2 3 d z = 10 − 3 π \int_{-\infty}^{\infty}h_2^3\,\mathrm{d}z = 10 - 3\pi ∫ − ∞ ∞ h 2 3 d z = 10 − 3 π (SI Sec. S1). This is the Rajagopal–Kimball coefficient a ℓ a_\ell a ℓ of (2) [13 , 12 ] . In two dimensions the screening pole N = 2 N = 2 N = 2 lies at s = 2 s = 2 s = 2 , to the left of the kernel pole at s = 3 s = 3 s = 3 . The leading term of the ring series is therefore the constant of the second-order ring diagram, which is part of a 0 a_0 a 0 . The log r s \log r_s log r s term appears one order later. The Gell-Mann–Brueckner and Rajagopal–Kimball terms both contain a single log r s \log r_s log r s because the ring kernel has no Fermi-surface factor.
For d > 3 d>3 d > 3 , 1 < N r < 2 1<N_r<2 1 < N r < 2 . No screening pole then coincides with the kernel pole, and the integral over t = q / q T F t=q/q_{\mathrm{TF}} t = q / q TF of the scaling function against t d t^d t d converges at both ends. The leading term is then a pure power [19 ] ,
e R P A ≃ A d r s ( 3 − d ) / ( d − 1 ) , A d = d α d N r − 2 β d N r 4 ( d + 1 ) sin ( π N r ) ∫ − ∞ ∞ h d ( z ) N r d z R y , (31) e_{\mathrm{RPA}}\simeq A_d r_s^{(3-d)/(d-1)},\qquad A_d=\frac{d\,\alpha_d^{N_r-2}\beta_d^{N_r}}{4(d+1)\sin(\pi N_r)}\int_{-\infty}^{\infty}h_d(z)^{N_r}\,\mathrm{d}z\ \mathrm{Ry}, \tag*{(31)} e RPA ≃ A d r s ( 3 − d ) / ( d − 1 ) , A d = 4 ( d + 1 ) sin ( π N r ) d α d N r − 2 β d N r ∫ − ∞ ∞ h d ( z ) N r d z Ry , (31)
which follows from ∫ 0 ∞ t d [ log ( 1 + h t 1 − d ) − h t 1 − d ] d t = h N r π / [ ( d − 1 ) N r sin ( π N r ) ] \int_0^\infty t^d[\log(1+ht^{1-d})-ht^{1-d}]\,\mathrm{d}t=h^{N_r}\pi/[(d-1)N_r\sin(\pi N_r)] ∫ 0 ∞ t d [ log ( 1 + h t 1 − d ) − h t 1 − d ] d t = h N r π / [( d − 1 ) N r sin ( π N r )] (SI Sec. S2). We check the coefficient in three cases. (i) For d = 4 d=4 d = 4 , where N r = 5 / 3 N_r=5/3 N r = 5/3 , the long-wavelength Lindhard shape h 4 = ( 1 + z 2 − z ) 2 h_4=(\sqrt{1+z^2}-z)^2 h 4 = ( 1 + z 2 − z ) 2 gives ∫ h 4 5 / 3 d z = 60 / 91 \int h_4^{5/3}\,\mathrm{d}z=60/91 ∫ h 4 5/3 d z = 60/91 and A 4 = − 24 ⋅ 2 5 / 12 / ( 91 3 ) = − 0.20325 A_4=-24\cdot2^{5/12}/(91\sqrt{3})=-0.20325 A 4 = − 24 ⋅ 2 5/12 / ( 91 3 ) = − 0.20325 Ry. (ii) As d → 3 + d\to3^+ d → 3 + , ( d − 3 ) A d → − 2 A (d-3)A_d\to-2A ( d − 3 ) A d → − 2 A . Since r s ( 3 − d ) / ( d − 1 ) = 1 − 1 2 ( d − 3 ) log r s + O ( ( d − 3 ) 2 ) r_s^{(3-d)/(d-1)}=1-\frac{1}{2}(d-3)\log r_s+O((d-3)^2) r s ( 3 − d ) / ( d − 1 ) = 1 − 2 1 ( d − 3 ) log r s + O (( d − 3 ) 2 ) , this pole in the dimension turns into the term A log r s A\log r_s A log r s of (29). The constant − 2 A / ( d − 3 ) -2A/(d-3) − 2 A / ( d − 3 ) left over cancels against the opposite pole of the explicit r s 0 r_s^0 r s 0 term of Eq. (21) of Ref. [19 ] , so that the energy stays finite as d → 3 d\to3 d → 3 . The log r s \log r_s log r s term of three dimensions is thus the confluence of two powers of r s r_s r s . (iii) As d → ∞ d\to\infty d → ∞ , A d / d → − 1 / ( 2 e π ) = − 0.05855 A_d/d\to-1/(2e\pi)=-0.05855 A d / d → − 1/ ( 2 e π ) = − 0.05855 Ry. Equation (31) is a simplified form of Eq. (21) of Ref. [19 ] : the series in that equation sums to the factor π / [ N r sin ( π N r ) ] \pi/[N_r\sin(\pi N_r)] π / [ N r sin ( π N r )] . The two forms agree numerically to machine precision (SI Sec. S2). Table 2 lists A d A_d A d for d = 4 d=4 d = 4 –9. These values differ from those in Table I of the same work, in which a numerical error was later noted [69, Ref. 68].
Table 2. Coefficient A d A_d A d of the leading term A d r s ( 3 − d ) / ( d − 1 ) A_d r_s^{(3-d)/(d-1)} A d r s ( 3 − d ) / ( d − 1 ) of the RPA correlation energy for d > 3 d>3 d > 3 , (31), in Ry. The third column gives Table I of Ref. [19 ] , converted from Hartree to Ry, and the last column the ratio of the second column to the third.
First-order exchange polarization in d d d dimensions
After the rings, the bare diagrams with the next-strongest small-q q q divergence form a second series. It inserts the first-order exchange polarization Π ( 1 ) = Π v t x ( 1 ) + Π S E ( 1 ) \Pi^{(1)}=\Pi_{\mathrm{vtx}}^{(1)}+\Pi_{\mathrm{SE}}^{(1)} Π ( 1 ) = Π vtx ( 1 ) + Π SE ( 1 ) , the sum of the exchange vertex correction and the Fock self-energy insertions of Figure 4 (a)–(Figure 4 c), once into the ring chain [8 , 9 ] . Replacing Π 0 \Pi_0 Π 0 by Π 0 + Π ( 1 ) \Pi_0+\Pi^{(1)} Π 0 + Π ( 1 ) in (26) changes c c c by − v Π ( 1 ) -v\Pi^{(1)} − v Π ( 1 ) , and the first-order change of log ( 1 + c ) − c \log(1+c)-c log ( 1 + c ) − c is − v Π ( 1 ) [ 1 / ( 1 + c ) − 1 ] -v\Pi^{(1)}[1/(1+c)-1] − v Π ( 1 ) [ 1/ ( 1 + c ) − 1 ] . A ring of j j j lines with one insertion Π ( 1 ) \Pi^{(1)} Π ( 1 ) receives j j j times the coefficient 1 / ( 2 j ) 1/(2j) 1/ ( 2 j ) of the ring of j j j free bubbles, that is, 1 2 \frac{1}{2} 2 1 . This is the closed-diagram coefficient per insertion, so the expansion reproduces the closed-diagram series of Refs. [8 , 9 ] (SI Sec. S14.6). The energy of the series is
Figure 4. (a) Exchange vertex correction, (b,c) Fock self-energy insertions in the polarization, and (d–f) the energies obtained by closing them with the same screened line W λ W_\lambda W λ . Panel (d) is the kite. Arrows denote G 0 G_0 G 0 , and dashed stubs denote density probes.
e Π ( 1 ) = d 8 π E 0 ∫ − ∞ ∞ d z ∫ 0 ∞ d q q d [ − v Π ( 1 ) ( q , i q z ) ] [ 1 1 + c − 1 ] . (32) e_{\Pi^{(1)}}=\frac{d}{8\pi}E_0\int_{-\infty}^{\infty}\mathrm{d}z\int_0^\infty\mathrm{d}q\,q^d[-v\Pi^{(1)}(q,iqz)]\left[\frac{1}{1+c}-1\right]. \tag*{(32)} e Π ( 1 ) = 8 π d E 0 ∫ − ∞ ∞ d z ∫ 0 ∞ d q q d [ − v Π ( 1 ) ( q , i q z )] [ 1 + c 1 − 1 ] . (32)
The scaling function 1 / ( 1 + c ) − 1 1/(1+c)-1 1/ ( 1 + c ) − 1 has the tail ∑ N ≥ 1 ( − h d ) N t N ∗ − N ( d − 1 ) \sum_{N\ge1}(-h_d)^N t^{N_*-N(d-1)} ∑ N ≥ 1 ( − h d ) N t N ∗ − N ( d − 1 ) , the same as that of the full-coupling line. This tail gives poles at s = N ( d − 1 ) s=N(d-1) s = N ( d − 1 ) .
By the Ward identity, shown explicitly in (59) of § 3.4 , the Fermi-surface factors log ( 1 / q ) \log(1/q) log ( 1/ q ) of the vertex and self-energy kernels cancel, and Π ( 1 ) \Pi^{(1)} Π ( 1 ) tends to a finite limit at fixed z z z ,
Π ( 1 ) ( q , i q z ) → − q T F d − 1 α d r s R d ( z ) , R d ( z ) = − B d 4 ⟨ ⟨ μ μ ′ Re [ ( b − b ′ ) 2 ] ∣ p ^ − p ^ ′ ∣ d − 1 ⟩ ⟩ , (33) \Pi^{(1)}(q,iqz) \to-q_{\mathrm{TF}}^{d-1}\alpha_{\mathrm{d}r_s}R_d(z),\qquad R_d(z)=-\frac{\mathcal{B}_d}{4}\left\langle\!\left\langle\frac{\mu\mu'\operatorname{Re}\left[(b-b')^2\right]}{\lvert\hat{p}-\hat{p}'\rvert^{d-1}}\right\rangle\!\right\rangle, \tag*{(33)} Π ( 1 ) ( q , i q z ) → − q TF d − 1 α d r s R d ( z ) , R d ( z ) = − 4 B d ⟨ ⟨ ∣ p ^ − p ^ ′ ∣ d − 1 μ μ ′ Re [ ( b − b ′ ) 2 ] ⟩ ⟩ , (33)
with α d r s = e 2 / ( ℏ v F ) \alpha_{\mathrm{d}r_s}=e^2/(\hbar v_{\mathrm{F}}) α d r s = e 2 / ( ℏ v F ) the coupling constant of (3) , b = ( i z − μ ) − 1 b=(iz-\mu)^{-1} b = ( i z − μ ) − 1 , b ′ = ( i z − μ ′ ) − 1 b'=(iz-\mu')^{-1} b ′ = ( i z − μ ′ ) − 1 , μ = p ^ ⋅ q ^ \mu=\hat{p}\cdot\hat{q} μ = p ^ ⋅ q ^ , μ ′ = p ^ ′ ⋅ q ^ \mu'=\hat{p}'\cdot\hat{q} μ ′ = p ^ ′ ⋅ q ^ , and the double average running over pairs of directions on the unit sphere. Equation (33) is the long-wavelength limit of the first-order exchange polarization (SI Sec. S3). In this limit each occupation difference becomes a factor q μ q\mu q μ on the Fermi surface, and each electron–hole factor becomes b / q b/q b / q . The factor ( b − b ′ ) 2 = ( μ − μ ′ ) 2 b 2 b ′ 2 (b-b')^2=(\mu-\mu')^2b^2b'^2 ( b − b ′ ) 2 = ( μ − μ ′ ) 2 b 2 b ′2 vanishes quadratically for coincident directions and thereby removes the forward singularity of the Coulomb factor ∣ p ^ − p ^ ′ ∣ 1 − d \lvert\hat{p}-\hat{p}'\rvert^{1-d} ∣ p ^ − p ^ ′ ∣ 1 − d . The kernel of (32) therefore has the endpoint term ∝ q \propto q ∝ q without a factor log ( 1 / q ) \log(1/q) log ( 1/ q ) . Its pole at s = 2 s=2 s = 2 coincides with the screening pole N = N ∗ N=N_\ast N = N ∗ when N ∗ ( d − 1 ) = 2 N_\ast(d-1)=2 N ∗ ( d − 1 ) = 2 , that is, at the resonant dimensions d = 1 + 2 / N ∗ = 3 , 2 , 5 3 , … d=1+2/N_\ast=3,2,\frac{5}{3},\ldots d = 1 + 2/ N ∗ = 3 , 2 , 3 5 , … , and there, by SI Eq. (S78) with m = 0 m=0 m = 0 ,
e Π ( 1 ) ⊃ d 8 π E 0 α d r s q T F d + 1 ( − 1 ) N ∗ + 1 d − 1 ∫ − ∞ ∞ h d ( z ) N ∗ R d ( z ) d z log r s , E 0 r s q T F d + 1 ∝ r s N ∗ . (34) e_{\Pi^{(1)}}\supset\frac{d}{8\pi}E_0\alpha_{\mathrm{d}r_s}q_{\mathrm{TF}}^{d+1}\frac{(-1)^{N_\ast+1}}{d-1}\int_{-\infty}^{\infty}h_d(z)^{N_\ast}R_d(z)\,\mathrm{d}z\,\log r_s,\qquad E_0r_sq_{\mathrm{TF}}^{d+1}\propto r_s^{N_\ast}. \tag*{(34)} e Π ( 1 ) ⊃ 8 π d E 0 α d r s q TF d + 1 d − 1 ( − 1 ) N ∗ + 1 ∫ − ∞ ∞ h d ( z ) N ∗ R d ( z ) d z log r s , E 0 r s q TF d + 1 ∝ r s N ∗ . (34)
In three dimensions the average over the relative azimuth reduces (33) to a double integral over μ \mu μ and μ ′ \mu' μ ′ . The result is R 3 = R ( 1 ) / ( 4 π ) R_3=R^{(1)}/(4\pi) R 3 = R ( 1 ) / ( 4 π ) , with the function of DuBois [8 ] and [68, Eqs. (40) and (41)],
R ( 1 ) ( z ) = 4 1 + z 2 [ ( 1 + 3 z 2 ) − ( 2 + 3 z 2 ) z arctan 1 z ] , (35) R^{(1)}(z)=\frac{4}{1+z^2}\left[(1+3z^2)-(2+3z^2)z\arctan\frac{1}{z}\right], \tag*{(35)} R ( 1 ) ( z ) = 1 + z 2 4 [ ( 1 + 3 z 2 ) − ( 2 + 3 z 2 ) z arctan z 1 ] , (35)
so that Π ( 1 ) ( q , i q z ) → − ( q T F 4 / 16 ) R ( 1 ) ( z ) \Pi^{(1)}(q,iqz)\to-(q_{\mathrm{TF}}^4/16)R^{(1)}(z) Π ( 1 ) ( q , i q z ) → − ( q TF 4 /16 ) R ( 1 ) ( z ) . (Eq. (41) of Ref. [68] prints the argument of the arctangent as μ \mu μ , to be read as 1 / μ 1/\mu 1/ μ .) We checked the reduction by direct quadrature of the reduced double integral. The quadrature agrees with the closed form to within 2.5 × 10 − 17 2.5\times10^{-17} 2.5 × 1 0 − 17 (SI Sec. S3). With N ∗ = 1 N_\ast=1 N ∗ = 1 , q T F 2 = 4 α L r s / π q_{\mathrm{TF}}^2=4\alpha_{\mathrm{L}}r_s/\pi q TF 2 = 4 α L r s / π and E 0 q T F 4 = 32 / π 2 E_0q_{\mathrm{TF}}^4=32/\pi^2 E 0 q TF 4 = 32/ π 2 Ry, (34) gives the r s log r s r_s\log r_s r s log r s term of the series in Ry (SI Sec. S14.6),
e Π ( 1 ) = 3 α L 2 π 4 ∫ − ∞ ∞ h ( z ) R ( 1 ) ( z ) d z r s log r s + O ( r s ) = α L ( π 2 − 12 log 2 ) 2 π 3 r s log r s + O ( r s ) , (36) e_{\Pi^{(1)}}=\frac{3\alpha_{\mathrm{L}}}{2\pi^4}\int_{-\infty}^{\infty}h(z)R^{(1)}(z)\,\mathrm{d}z\,r_s\log r_s+O(r_s)=\frac{\alpha_{\mathrm{L}}(\pi^2-12\log2)}{2\pi^3}r_s\log r_s+O(r_s), \tag*{(36)} e Π ( 1 ) = 2 π 4 3 α L ∫ − ∞ ∞ h ( z ) R ( 1 ) ( z ) d z r s log r s + O ( r s ) = 2 π 3 α L ( π 2 − 12 log 2 ) r s log r s + O ( r s ) , (36)
that is, 0.01304 r s log r s 0.01304\,r_s\log r_s 0.01304 r s log r s , the exchange part of the coefficient of Carr and Maradudin [9 ] and [7 Eq. (5)] . The residue integral agrees with the closed form of [10 Table I] and [3 Table 1] to 40 digits in multiprecision arithmetic (SI Sec. S19). SI Sec. S14.6 bounds the remainder in (36) under the assumption that the correction to the limit (33) vanishes as a positive power of q q q .
The rest of the jellium r s log r s r_s\log r_s r s log r s term, α L ( π 2 − 6 ) / ( 12 π 3 ) r s log r s = 0.00542 r s log r s \alpha_{\mathrm{L}}(\pi^2-6)/(12\pi^3)\,r_s\log r_s=0.00542\,r_s\log r_s α L ( π 2 − 6 ) / ( 12 π 3 ) r s log r s = 0.00542 r s log r s , arises in the ring series. It comes from the cross term of h ( z ) h(z) h ( z ) and the next term of the small-q q q expansion L ( q , q z ) = h ( z ) + q 2 g 3 ( z ) + O ( q 4 ) \mathcal{L}(q,qz)=h(z)+q^2g_3(z)+O(q^4) L ( q , q z ) = h ( z ) + q 2 g 3 ( z ) + O ( q 4 ) , with g 3 ( z ) = − 1 / [ 12 ( 1 + z 2 ) 2 ] g_3(z)=-1/[12(1+z^2)^2] g 3 ( z ) = − 1/ [ 12 ( 1 + z 2 ) 2 ] from SI Eq. (S87). To first order the change q T F 4 g 3 q_{\mathrm{TF}}^4g_3 q TF 4 g 3 of c c c adds q 3 q T F 3 g 3 [ − c / ( 1 + c ) ] q^3q_{\mathrm{TF}}^3g_3[-c/(1+c)] q 3 q TF 3 g 3 [ − c / ( 1 + c )] to the integrand of (26). This addition is the ring series with one insertion of the polarization − q T F 2 q 2 g 3 -q_{\mathrm{TF}}^2q^2g_3 − q TF 2 q 2 g 3 . Its term q 3 q T F 3 g 3 h 2 / t 4 q^3q_{\mathrm{TF}}^3g_3h^2/t^4 q 3 q TF 3 g 3 h 2 / t 4 has the kernel q 3 q^3 q 3 , with its pole at s = 4 s=4 s = 4 . This is the pole s = 2 + m s=2+m s = 2 + m of a polarization insertion whose small-q q q form is q m q^m q m times a function of z z z , here with m = 2 m=2 m = 2 (the dashed line of figure 2 and the counting of Sec. 3.1). It coincides with the second screening pole s = 2 ( d − 1 ) = 4 s=2(d-1)=4 s = 2 ( d − 1 ) = 4 , the resonance of the ring series itself. The double pole gives q T F 6 log q T F q_{\mathrm{TF}}^6\log q_{\mathrm{TF}} q TF 6 log q TF [68, Eqs. (33) and (36)] and [7 Eq. (3)] (SI Sec. S14.6).
Together the two parts give the coefficient C C C of (1). Loos and Gill call the part from the ring series the RPA contribution and the part from the first-order exchange polarization the second-order exchange contribution [3 Eqs. (29) and (30)] . If the vertex correction were kept alone, its kernel would carry log ( 1 / q ) \log(1/q) log ( 1/ q ) , the pole would be triple, and the term would be r s log 2 r s r_s\log^2 r_s r s log 2 r s , as found for the kite in § 3.3 . The r s log r s r_s\log r_s r s log r s term of jellium thus carries a single log r s \log r_s log r s because the factors log ( 1 / q ) \log(1/q) log ( 1/ q ) of the vertex and self-energy corrections cancel, as the Ward identity requires. This cancellation removes the higher-order pole, in accordance with Corollary 2.
Results
Section 3.1 derives the new result, the coefficient a 2 ℓ a_{2\ell} a 2 ℓ of r s 2 log r s r_s^2\log r_s r s 2 log r s in two dimensions, by the rule of § 2 : the kernel pole of the polarization insertion at s = 2 s=2 s = 2 meets the second screening pole. It also states the hypotheses under which a 2 ℓ a_{2\ell} a 2 ℓ is the complete coefficient. Sections 3.2–3.4 apply the rule to the screened second-order exchange diagram, the kite, in three dimensions. They give its energy expressions and the endpoints of its radial kernel, its leading terms at high and low density, and the cancellation of these terms by the Fock self-energy corrections. Section 3.3 compares the high- and low-density terms with the fit of BRM.
The r s 2 log r s r_s^2\log r_s r s 2 log r s term of the two-dimensional gas
In two dimensions the kernel pole of the polarization insertion at s = 2 s=2 s = 2 coincides with the second screening pole, N ∗ = 2 N_* = 2 N ∗ = 2 (Figure 2 ). Equation (34) therefore predicts a term r s 2 log r s r_s^2\log r_s r s 2 log r s from the first-order exchange polarization. The coefficient of this term requires the long-wavelength form R 2 ( z ) R_2(z) R 2 ( z ) of Π ( 1 ) \Pi^{(1)} Π ( 1 ) at every z z z , which we now obtain in closed form.
For d = 2 d=2 d = 2 the average in (33) runs over pairs of angles on the Fermi circle, ∣ p ^ − p ^ ′ ∣ = 2 ∣ sin ( ϕ / 2 ) ∣ |\hat{\boldsymbol{p}}-\hat{\boldsymbol{p}}'|=2|\sin(\phi/2)| ∣ p ^ − p ^ ′ ∣ = 2∣ sin ( ϕ /2 ) ∣ with ϕ \phi ϕ the angle between the two directions, and B 2 = 2 \mathcal{B}_2=2 B 2 = 2 . We expand the three factors in Fourier series on the circle. Because the numerator vanishes at ϕ = 0 \phi=0 ϕ = 0 , the Coulomb factor 1 / ∣ 2 sin ( ϕ / 2 ) ∣ 1/|2\sin(\phi/2)| 1/∣2 sin ( ϕ /2 ) ∣ enters only through the differences of its Fourier coefficients, ( 2 / π ) ∑ j = 1 ∣ m ∣ 1 / ( 2 j − 1 ) (2/\pi)\sum_{j=1}^{|m|}1/(2j-1) ( 2/ π ) ∑ j = 1 ∣ m ∣ 1/ ( 2 j − 1 ) for the harmonic m m m . The Fourier coefficients of b = ( i z − cos θ ) − 1 b=(iz-\cos\theta)^{-1} b = ( i z − cos θ ) − 1 are geometric in r = 1 + z 2 − z r=\sqrt{1+z^2}-z r = 1 + z 2 − z , and the sum over m m m is elementary (SI § 15 ). With s z = 1 + z 2 s_z=\sqrt{1+z^2} s z = 1 + z 2 , r = s z − z r=s_z-z r = s z − z and a = z / s z a=z/s_z a = z / s z ,
R 2 ( z ) = 2 π [ h 2 + 2 a 2 r arctan r − r 2 1 + r 2 ] = 2 π [ h 2 + z h 2 ′ + 2 a 2 r arctan r 1 + r 2 ] , (37) R_2(z)=\frac{2}{\pi}\left[h_2+2a^2\frac{r\arctan r-r^2}{1+r^2}\right]=\frac{2}{\pi}\left[h_2+zh_2'+\frac{2a^2r\arctan r}{1+r^2}\right], \tag*{(37)} R 2 ( z ) = π 2 [ h 2 + 2 a 2 1 + r 2 r arctan r − r 2 ] = π 2 [ h 2 + z h 2 ′ + 1 + r 2 2 a 2 r arctan r ] , (37)
where h 2 = 1 − a h_2=1-a h 2 = 1 − a is the long-wavelength Lindhard shape of (10) and h 2 ′ = d h 2 / d z h_2'=dh_2/dz h 2 ′ = d h 2 / d z . Both forms hold for z ≥ 0 z\geq0 z ≥ 0 , and R 2 R_2 R 2 is even in z z z . A quadrature of the angular double integral (33) , independent of the Fourier sum, agrees with (37) to 10 − 17 10^{-17} 1 0 − 17 at z = 1 z=1 z = 1 and to 10 − 13 10^{-13} 1 0 − 13 at z = 0.3 z=0.3 z = 0.3 and 3 3 3 (SI § 15 ). The function is shown in Figure 3 . It has the following properties.
Figure 3. (a) Long-wavelength form R 2 ( z ) R_2(z) R 2 ( z ) of the first-order exchange polarization of the two-dimensional gas, (37) , as a function of the scaled frequency z z z , and the long-wavelength Lindhard shape h 2 ( z ) h_2(z) h 2 ( z ) , with the three-dimensional counterparts R 3 = R ( 1 ) / ( 4 π ) R_3 = R^{(1)}/(4\pi) R 3 = R ( 1 ) / ( 4 π ) of (35) and h 3 = h h_3 = h h 3 = h (dashed). (b) R 2 ( z ) R_2(z) R 2 ( z ) on log-log axes with its static value 2 / π 2/\pi 2/ π (Hartree–Fock compressibility) and its high-frequency tail 5 / ( 12 π z 4 ) 5/(12\pi z^4) 5/ ( 12 π z 4 ) .
(i) In the static limit, R 2 ( 0 ) = 2 / π R_2(0)=2/\pi R 2 ( 0 ) = 2/ π , so that Π 0 + Π ( 1 ) → − q T F ( 1 + 2 r s / π ) \Pi_0+\Pi^{(1)}\to-q_{\mathrm{TF}}(1+\sqrt{2}r_s/\pi) Π 0 + Π ( 1 ) → − q TF ( 1 + 2 r s / π ) for q → 0 q\to0 q → 0 at ν = 0 \nu=0 ν = 0 . This limit gives the Hartree–Fock compressibility of the two-dimensional gas. The ratio of this compressibility to the free value, 1 / ( 1 − 2 r s / π ) 1/(1-\sqrt{2}r_s/\pi) 1/ ( 1 − 2 r s / π ) to this order, was given by Rajagopal [32 ] , by Czachor, Holas, Sharma and Singwi (CHSS) [33 ] , Eq. (5.29), by Hameeuw, Brosens and Devreese [34 ] and recently, with the static wave-number dependence, by Das Sarma [35 ] .
(ii) In the high-frequency limit, R 2 ( z ) = 5 / ( 12 π z 4 ) − 11 / ( 20 π z 6 ) + O ( z − 8 ) R_2(z)=5/(12\pi z^4)-11/(20\pi z^6)+O(z^{-8}) R 2 ( z ) = 5/ ( 12 π z 4 ) − 11/ ( 20 π z 6 ) + O ( z − 8 ) . The absence of a z − 2 z^{-2} z − 2 term expresses the f f f -sum rule: the free response already exhausts the first frequency moment (SI Lemma S1). With Π ( 1 ) = − q T F ( r s / 2 ) R 2 \Pi^{(1)}=-q_{\mathrm{TF}}(r_s/\sqrt{2})R_2 Π ( 1 ) = − q TF ( r s / 2 ) R 2 and z → − i ω / q z\to-i\omega/q z → − iω / q , the leading term is − v Π ( 1 ) = 5 r s 2 q 2 / ( 12 π ω 4 ) -v\Pi^{(1)}=5r_s^2q^2/(12\pi\omega^4) − v Π ( 1 ) = 5 r s 2 q 2 / ( 12 π ω 4 ) at real frequency ω \omega ω in units of ω 0 \omega_0 ω 0 , which is Eq. (5.12) of CHSS [33 ] and Eq. (B.10) of Ref. [34 ] . This term fixes the exchange correction to the plasmon dispersion obtained by Rajagopal [32 ] . The second term predicts + 11 r s 2 q 2 / ( 20 π ω 6 ) +11r_s^2q^2/(20\pi\omega^6) + 11 r s 2 q 2 / ( 20 π ω 6 ) in the high-frequency expansion, which these works do not give.
(iii) In accordance with the Ward identity, the vertex and self-energy parts of Π ( 1 ) \Pi^{(1)} Π ( 1 ) carry ∓ q T F ( r s / 2 ) π − 1 [ h 2 + z h 2 ′ ] log ( 1 / q ) \mp q_{\mathrm{TF}}(r_s/\sqrt{2})\pi^{-1}[h_2 + zh_2']\log(1/q) ∓ q TF ( r s / 2 ) π − 1 [ h 2 + z h 2 ′ ] log ( 1/ q ) at small q q q and fixed z z z , the two-dimensional form of (59) (SI Sec. S4). In both two and three dimensions the coefficient is q T F d − 1 α d r s π − 1 [ h d + z h d ′ ] q_{\mathrm{TF}}^{d-1}\alpha_d r_s\pi^{-1}[h_d + zh_d'] q TF d − 1 α d r s π − 1 [ h d + z h d ′ ] , and its static value agrees with the separate terms in log ( 1 / q ) \log(1/q) log ( 1/ q ) found by Das Sarma [35 Eqs. (21) and (22)] . In the Fourier evaluation, the first two terms of the second form of (37) together give twice this coefficient. They come from the self-energy terms b 2 + b ′ 2 b^2+b'^2 b 2 + b ′2 of ( b − b ′ ) 2 (b-b')^2 ( b − b ′ ) 2 through the harmonics m = ± 1 m=\pm1 m = ± 1 . The last term comes from the vertex term − 2 b b ′ -2bb' − 2 b b ′ summed over all harmonics (SI Sec. S4).
(iv) At finite wave number, the corrections to the limit (37) at fixed z z z are of order q 2 log q q^2\log q q 2 log q , without an O ( q ) O(q) O ( q ) term, at the values of z z z examined. Das Sarma proved this for the static response and obtained the coefficient of q 2 log q q^2\log q q 2 log q [35 ] . Our evaluation of the finite-q q q response agrees with his static values within the accuracy of our quadrature. It shows the same behavior at z = 0.3 z=0.3 z = 0.3 and is consistent with this behavior at z = 1 z=1 z = 1 (SI Sec. S6).
The first-order exchange polarization of the two-dimensional gas has been evaluated numerically for arbitrary wave number and frequency [33 , 34 ] [70]. The limits (i) and (ii) have been obtained in closed form [32 , 33 , 34 ] . The residue at the coincident pole requires the closed form (37) at fixed z z z , which interpolates between these limits.
For d = 2 d=2 d = 2 we have N ∗ = 2 N_\ast=2 N ∗ = 2 , and the scaling function 1 / ( 1 + c ) − 1 1/(1+c)-1 1/ ( 1 + c ) − 1 of (32) has the tail − h 2 / t + h 2 2 / t 2 − ⋯ -h_2/t+h_2^2/t^2-\cdots − h 2 / t + h 2 2 / t 2 − ⋯ . The first term gives the simple pole at s = 1 s=1 s = 1 , an O ( r s ) O(r_s) O ( r s ) contribution to the term with coefficient a 1 a_1 a 1 of (2) . The pole of the second term coincides with the kernel pole at s = 2 s=2 s = 2 . With E 0 α 2 r s q T F 2 = 8 r s 2 R y E_0\alpha_2r_s q_{\mathrm{TF}}^2=8r_s^2\,\mathrm{Ry} E 0 α 2 r s q TF 2 = 8 r s 2 Ry , (34) gives (SI Sec. S5)
e Π ( 1 ) ⊃ a 2 ℓ r s 2 log r s , a 2 ℓ = − 2 π ∫ − ∞ ∞ h 2 ( z ) 2 R 2 ( z ) d z . (38) e_{\Pi^{(1)}} \supset a_{2\ell}r_s^2\log r_s,\qquad a_{2\ell}=-\frac{2}{\pi}\int_{-\infty}^{\infty}h_2(z)^2R_2(z)\,\mathrm{d}z. \tag*{(38)} e Π ( 1 ) ⊃ a 2 ℓ r s 2 log r s , a 2 ℓ = − π 2 ∫ − ∞ ∞ h 2 ( z ) 2 R 2 ( z ) d z . (38)
The substitution r = 1 + z 2 − z r=\sqrt{1+z^2}-z r = 1 + z 2 − z maps z > 0 z>0 z > 0 onto 0 < r < 1 0<r<1 0 < r < 1 , with h 2 = 2 r 2 / ( 1 + r 2 ) h_2=2r^2/(1+r^2) h 2 = 2 r 2 / ( 1 + r 2 ) and a = ( 1 − r 2 ) / ( 1 + r 2 ) a=(1-r^2)/(1+r^2) a = ( 1 − r 2 ) / ( 1 + r 2 ) . It reduces the integral to three elementary integrals. One of them contains arctan r \arctan r arctan r and produces Catalan’s constant (SI Sec. S5). The result is
a 2 ℓ = 7 π + 4 log 2 π − 160 9 π 2 − 16 G π 2 = − 0.1754613739208 … R y , (39) a_{2\ell}=\frac{7}{\pi}+\frac{4\log2}{\pi}-\frac{160}{9\pi^2}-\frac{16G}{\pi^2}=-0.1754613739208\ldots\,\mathrm{Ry}, \tag*{(39)} a 2 ℓ = π 7 + π 4 log 2 − 9 π 2 160 − π 2 16 G = − 0.1754613739208 … Ry , (39)
that is, − 0.0877306869604 -0.0877306869604 − 0.0877306869604 Hartree, in agreement with multiprecision quadrature of (38) to 100 digits. Catalan’s constant appears, as in the second-order exchange part of the constant a 0 a_0 a 0 of (2), through the Fermi-circle geometry. At r s = 1 r_s = 1 r s = 1 the term vanishes, and at r s = 0.1 r_s = 0.1 r s = 0.1 it amounts to 4.0 × 10 − 3 4.0 \times10^{-3} 4.0 × 1 0 − 3 Ry, about a tenth of the term a ℓ r s log r s a_\ell r_s \log r_s a ℓ r s log r s .
For a fully spin-polarized gas the same steps, with spin-resolved h 2 h_2 h 2 and R 2 R_2 R 2 , give a 2 ℓ / 8 a_{2\ell}/8 a 2 ℓ /8 , the analog of the factor 1 / ( 4 2 ) 1/(4\sqrt{2}) 1/ ( 4 2 ) of the Rajagopal–Kimball term [13 ] [71]. At partial polarization ζ \zeta ζ the coefficient depends weakly on ζ \zeta ζ , but it does not tend to a 2 ℓ / 8 a_{2\ell}/8 a 2 ℓ /8 as ζ → 1 \zeta\to1 ζ → 1 . The first-order exchange polarization of the minority spins grows as 1 / k F ↓ 1/k_{F\downarrow} 1/ k F ↓ , while their density of states stays constant in two dimensions. The expansion is therefore uniform only for r s ≪ ( 1 − ζ ) 1 / 2 r_s \ll(1-\zeta)^{1/2} r s ≪ ( 1 − ζ ) 1/2 (SI Sec. S8).
Equation (39) gives the coefficient of the series that inserts Π ( 1 ) \Pi^{(1)} Π ( 1 ) once into the ring chain. Whether this value is the complete coefficient of r s 2 log r s r_s^2 \log r_s r s 2 log r s in ϵ c 2 D \epsilon_c^{2\mathrm{D}} ϵ c 2 D depends on which other diagrams can produce a coincident pole at this order. Consider the class of diagrams with one soft line, that is, a screened line whose momentum transfer is of order k T F k_{\mathrm{TF}} k TF and whose frequency is of order v F v_F v F times this momentum transfer. Within this class, an insertion of order r s j r_s^j r s j whose small-q q q expansion at fixed z z z contains a term q m F ( z ) q^m F(z) q m F ( z ) gives, at a resonance, the term r s j + ( 3 − d + m ) / ( d − 1 ) log r s r_s^{j+(3-d+m)/(d-1)} \log r_s r s j + ( 3 − d + m ) / ( d − 1 ) log r s (SI Sec. S7). In three dimensions this is r s j + m / 2 log r s r_s^{j+m/2} \log r_s r s j + m /2 log r s . Within this class, the order r s log r s r_s \log r_s r s log r s receives exactly the two contributions of Sec. 2.9, Π ( 1 ) \Pi^{(1)} Π ( 1 ) (j = 1 j = 1 j = 1 , m = 0 m = 0 m = 0 ) and the q 2 q^2 q 2 term of the Lindhard function (j = 0 j = 0 j = 0 , m = 2 m = 2 m = 2 ). Their sum is the established coefficient C C C . In two dimensions the term is r s j + 1 + m log r s r_s^{j+1+m} \log r_s r s j + 1 + m log r s . The order r s 2 log r s r_s^2 \log r_s r s 2 log r s could therefore receive, besides Π ( 1 ) \Pi^{(1)} Π ( 1 ) , only a term linear in q q q of the Lindhard function. This term is absent because L ( q , q z ) \mathcal{L}(q,qz) L ( q , q z ) is even in q q q at fixed z z z (SI Sec. S1). The q 2 q^2 q 2 term of the Lindhard function, g 2 ( z ) = − z / [ 8 ( 1 + z 2 ) 5 / 2 ] g_2(z) = -z/[8(1+z^2)^{5/2}] g 2 ( z ) = − z / [ 8 ( 1 + z 2 ) 5/2 ] , enters at r s 3 log r s r_s^3 \log r_s r s 3 log r s , as do the second-order polarization and two insertions of Π ( 1 ) \Pi^{(1)} Π ( 1 ) . The corrections of order q 2 log q q^2 \log q q 2 log q to Π ( 1 ) \Pi^{(1)} Π ( 1 ) enter at r s 4 log 2 r s r_s^4 \log^2 r_s r s 4 log 2 r s .
Diagrams with several soft lines are counted by the number L L L of independent soft loops, the loops formed by the soft lines when each fermion loop is shrunk to a vertex. Lines whose momentum transfer is of order k F k_F k F are hard. The nonlinear vertices that couple soft density fluctuations are the closed fermion loops with three or more density vertices. In the sum over diagrams, each such loop appears symmetrized over the orderings of its vertices. The symmetrized loops are bounded at soft momenta Q Q Q and imaginary frequencies of order v F Q v_F Q v F Q . Neumayr and Metzner showed this in two dimensions [36 Eq. (73)] , and Kopper and Magnen proved it with Ward identities for d ≤ 3 d \le3 d ≤ 3 [37 ] . The bound rests on a cancellation among the terms of the symmetrized loop, the loop cancellation. Without it, the unsymmetrized three-point loops would each be larger by a factor k F / k T F k_F/k_{\mathrm{TF}} k F / k TF . The bound holds when no partial sum of the external frequencies of a loop is small compared with v F Q v_F Q v F Q , and its constant grows as such a sum tends to zero. A diagram with L L L soft loops then scales as E Q T F L ( d + 1 ) E_{\mathrm{QTF}}^{L(d+1)} E QTF L ( d + 1 ) , which is r s 3 L − 2 r_s^{3L-2} r s 3 L − 2 in two dimensions and r s 2 L − 2 r_s^{2L-2} r s 2 L − 2 in three (SI Sec. S7). The ring series is L = 1 L = 1 L = 1 , and L = 2 L = 2 L = 2 starts at r s 4 r_s^4 r s 4 in two dimensions. Without the loop cancellation, two-loop diagrams would enter at r s 2 r_s^2 r s 2 , the order considered here.
The completeness of a 2 ℓ a_{2\ell} a 2 ℓ rests on two hypotheses. (H1) The bound on the symmetrized loops holds uniformly on the soft region away from its degenerate parts (small partial sums of the soft frequencies, collinear or canceling soft momenta). The degenerate parts and the mixed regions with hard subdiagrams admit integrable bounds compatible with this bound, and the hard regions contribute powers of r s r_s r s without log r s \log r_s log r s . (H2) The proper polarization at fixed z z z is − q T F [ L ( q , q z ) + ( r s / 2 ) R 2 + r s 2 P 2 ( z ) ] + O ( r s q 2 log q ) + O ( r s 3 ) -q_{\mathrm{TF}}[\mathcal{L}(q,qz)+(r_s/\sqrt{2})R_2+r_s^2P_2(z)]+O(r_s q^2\log q)+O(r_s^3) − q TF [ L ( q , q z ) + ( r s / 2 ) R 2 + r s 2 P 2 ( z )] + O ( r s q 2 log q ) + O ( r s 3 ) , with the free response L \mathcal{L} L kept exactly, remainders integrable in z z z and P 2 P_2 P 2 bounded and decaying as z − 4 z^{-4} z − 4 , where the internal lines of the second-order terms are screened. Under (H1) and (H2), the coefficient of r s 2 log r s r_s^2 \log r_s r s 2 log r s in the correlation energy of the paramagnetic two-dimensional electron gas is a 2 ℓ a_{2\ell} a 2 ℓ of (39) (SI Proposition S1).
SI Sec. S7 gives the counting behind this statement and two checks. In three dimensions the same counting places the two-loop diagrams at r s 2 r_s^2 r s 2 , consistent with the established coefficient C C C , which contains only one-line terms. Replacing the bare exchange line inside Π ( 1 ) \Pi^{(1)} Π ( 1 ) by a Thomas–Fermi screened line changes R 2 R_2 R 2 at fixed z z z by an amount linear in the screening wave number κ s \kappa_s κ s , without log κ s \log\kappa_s log κ s . Screening of the internal line therefore does not shift the coefficient. Without (H1) and (H2), (39) still gives the coefficient of the Π ( 1 ) \Pi^{(1)} Π ( 1 ) series, the two-dimensional counterpart of the exchange part of the Carr–Maradudin coefficient.
The screened kite: energy expressions and endpoints of the radial kernel
Sections 3.2–3.4 treat the three-dimensional gas, d = 3 d = 3 d = 3 , except for the last paragraph of Sec. 3.4. The screened second-order exchange diagram, the kite, is evaluated here in the energy expression of BRM [Eqs. (21)–(24)] of Ref. [25 ] . BRM call it the kite-diagram series. They distinguish it from the adiabatic-connection SOSEX (AC-SOSEX) and from the coupled-cluster SOSEX of Freeman [72], whose values lie closer to a quasistatic approximation of the kite [Eq. (61)] of Ref. [25 ] . The radial kernel of the kite has an endpoint term with a Fermi-surface factor log ( 1 / q ) \log(1/q) log ( 1/ q ) at small momentum transfer and a q − 2 q^{-2} q − 2 endpoint term at large momentum transfer. Together with the scaling functions of table 1, these endpoint terms give the leading terms of the kite energy (Sec. 3.3). The Fock self-energy insertions that complete the first-order exchange polarization Π ( 1 ) \Pi^{(1)} Π ( 1 ) have the same endpoint terms with opposite coefficients and cancel these leading terms (Sec. 3.4).
In the kite (Figure 4 d), one interaction line is screened and the other remains bare. We denote the functional (11) of the kite by K [ f ] \mathcal{K}[f] K [ f ] , normalized so that K [ v ] \mathcal{K}[v] K [ v ] is the bare kite energy. bare kite is a single four-vertex fermion loop with two Coulomb lines. Two Wick contractions produce this topology. They give the closed-diagram coefficient 2 / ( 2 ! 2 2 ) = 1 / 4 2/(2! \, 2^2)=1/4 2/ ( 2 ! 2 2 ) = 1/4 , where 1 / 2 ! 1/2! 1/2 ! is the second-order factor and 1 / 2 2 1/2^2 1/ 2 2 comes from the factor 1 / 2 1/2 1/2 of each Coulomb interaction. The spin sum over the single loop gives a factor 2. With the Fourier measures, Coulomb units and conversion to Ry per electron (SI Sec. S9), the integrand Y D Y_D Y D of (11) is A 0 X A_0X A 0 X with S D = 4 S_D=4 S D = 4 and n = 2 n=2 n = 2 , so that
K [ f ] = A 0 ∫ d 3 q ∫ 0 ∞ d ν π f ( q , i ν ) X ( q , ν ) , A 0 = 3 32 π 5 , (40) \mathcal{K}[f]=A_0\int\mathrm{d}^3q\int_0^\infty\frac{\mathrm{d}\nu}{\pi}\,f(q,i\nu)X(q,\nu),\qquad A_0=\frac{3}{32\pi^5}, \tag*{(40)} K [ f ] = A 0 ∫ d 3 q ∫ 0 ∞ π d ν f ( q , i ν ) X ( q , ν ) , A 0 = 32 π 5 3 , (40)
X ( q , ν ) = ∫ d 3 p d 3 k ∣ p − k ∣ 2 Re Δ p Δ k ( i ν − δ p ) ( i ν − δ k ) , (41) X(q,\nu)=\int\frac{\mathrm{d}^3p\,\mathrm{d}^3k}{\lvert p-k\rvert^2}\operatorname{Re}\frac{\Delta_p\Delta_k}{(i\nu-\delta_p)(i\nu-\delta_k)}, \tag*{(41)} X ( q , ν ) = ∫ ∣ p − k ∣ 2 d 3 p d 3 k Re ( i ν − δ p ) ( i ν − δ k ) Δ p Δ k , (41)
Δ p = Θ ( 1 − ∣ p ∣ ) − Θ ( 1 − ∣ p + q ∣ ) , δ p = p ⋅ q + q 2 2 . \Delta_p=\Theta(1-\lvert p\rvert)-\Theta(1-\lvert p+q\rvert),\qquad\delta_p=p\cdot q+\frac{q^2}{2}. Δ p = Θ ( 1 − ∣ p ∣) − Θ ( 1 − ∣ p + q ∣) , δ p = p ⋅ q + 2 q 2 .
The occupation difference Δ p \Delta_p Δ p is nonzero only when one of p p p and p + q p+q p + q is occupied and the other is empty, and δ p \delta_p δ p is the signed excitation energy of the free electron–hole pair. Each electron–hole factor Δ p / ( i ν − δ p ) \Delta_p/(i\nu-\delta_p) Δ p / ( i ν − δ p ) arises from the internal fermion frequency integral of two propagators, and the internal Coulomb factor ∣ p − k ∣ − 2 \lvert p-k\rvert^{-2} ∣ p − k ∣ − 2 couples the two factors. The real part is taken because the frequency integral runs over ν > 0 \nu>0 ν > 0 only (Sec. 2.5).
As a check on the normalization, the bare limit gives, in Ry,
K [ v ] = e 2 x = 1 3 log 2 − 3 ζ ( 3 ) 2 π 2 , (42) \mathcal{K}[v]=e_{2x}=\frac{1}{3}\log2-\frac{3\zeta(3)}{2\pi^2}, \tag*{(42)} K [ v ] = e 2 x = 3 1 log 2 − 2 π 2 3 ζ ( 3 ) , (42)
the Onsager–Mittag–Stephen second-order exchange energy [6 ] [73]. A direct Monte Carlo evaluation of (40) with f = v f=v f = v gives 0.04836(2) Ry, in agreement with this value (SI Sec. S19).
The full-coupling kite energy and the BRM coupling-averaged energy, which is (12) with w D = 2 λ w_D=2\lambda w D = 2 λ , are [25 ]
F ( r s ) = K [ W λ = 1 ( r s ) ] , e k i t e ( r s ) = 2 ∫ 0 1 d λ K [ W λ ( r s ) ] = 2 K [ W ‾ ] , (43) F(r_s)=\mathcal{K}[W_{\lambda=1}(r_s)],\qquad e_{\mathrm{kite}}(r_s)=2\int_0^1\mathrm{d}\lambda\,\mathcal{K}[W_\lambda(r_s)]=2\mathcal{K}[\overline{W}], \tag*{(43)} F ( r s ) = K [ W λ = 1 ( r s )] , e kite ( r s ) = 2 ∫ 0 1 d λ K [ W λ ( r s )] = 2 K [ W ] , (43)
where W ‾ = ∫ 0 1 W λ d λ \overline{W}=\int_0^1W_\lambda\,\mathrm{d}\lambda W = ∫ 0 1 W λ d λ . SI Eq. (S16) derives e k i t e e_{\mathrm{kite}} e kite from the self-energy trace, and SI Sec. S11 gives the correspondence with the BRM representation. Without screening, W λ = λ v W_\lambda=\lambda v W λ = λ v and W ‾ = v / 2 \overline{W}=v/2 W = v /2 , so both energies reduce to e 2 x e_{2x} e 2 x . With c c c of (8) , 2 W ‾ = v ρ ( c ) 2\overline{W}=v\rho(c) 2 W = v ρ ( c ) by (19) . Both F F F and e k i t e e_{\mathrm{kite}} e kite describe the isolated kite, and e k i t e e_{\mathrm{kite}} e kite at zero spin polarization is the energy studied in the companion paper [50].
A third energy expression sums closed diagrams with their closed-diagram coefficients. The chain family consists of the bare kite and every diagram with a nonempty chain of free polarization bubbles on one kite line. The closed-diagram coefficient is 1 / 4 1/4 1/4 for the bare kite and 1 / 2 1/2 1/2 for every nonempty chain (SI Proposition S2). The coefficient doubles because the two kite lines are equivalent and a chain on one of them breaks the symmetry that exchanges them. The family therefore sums to
e c l o s e d = K [ v ] + 2 K [ W λ = 1 − v ] = 2 F ( r s ) − e 2 x . (44) e_{\mathrm{closed}}=\mathcal{K}[v]+2\mathcal{K}[W_{\lambda=1}-v]=2F(r_s)-e_{2x}. \tag*{(44)} e closed = K [ v ] + 2 K [ W λ = 1 − v ] = 2 F ( r s ) − e 2 x . (44)
For a chain of j ≥ 1 j\geq1 j ≥ 1 bubbles the BRM average assigns 1 / ( j + 2 ) 1/(j+2) 1/ ( j + 2 ) of the chain-family coefficient, because it marks only the two original kite interactions and retains the bare coefficient 1 / 4 1/4 1/4 (SI Sec. S12.2). Equivalently, the j j j -bubble term of W λ W_\lambda W λ carries λ j + 1 \lambda^{j+1} λ j + 1 , and 2 ∫ 0 1 λ j + 1 d λ = 2 / ( j + 2 ) 2\int_0^1\lambda^{j+1}\,\mathrm{d}\lambda=2/(j+2) 2 ∫ 0 1 λ j + 1 d λ = 2/ ( j + 2 ) .
The three energy expressions are related by
e k i t e ( r s ) = 2 ∫ 0 1 λ F ( λ r s ) d λ = 2 r s 2 ∫ 0 r s ρ F ( ρ ) d ρ , (45) e_{\mathrm{kite}}(r_s)=2\int_0^1\lambda F(\lambda r_s)\,\mathrm{d}\lambda=\frac{2}{r_s^2}\int_0^{r_s}\rho F(\rho)\,\mathrm{d}\rho, \tag*{(45)} e kite ( r s ) = 2 ∫ 0 1 λ F ( λ r s ) d λ = r s 2 2 ∫ 0 r s ρF ( ρ ) d ρ , (45)
F ( r s ) = e k i t e ( r s ) + r s 2 e k i t e ′ ( r s ) , e c l o s e d ( r s ) = 2 e k i t e ( r s ) + r s e k i t e ′ ( r s ) − e 2 x . F(r_s)=e_{\mathrm{kite}}(r_s)+\frac{r_s}{2}e_{\mathrm{kite}}'(r_s),\qquad e_{\mathrm{closed}}(r_s)=2e_{\mathrm{kite}}(r_s)+r_se_{\mathrm{kite}}'(r_s)-e_{2x}. F ( r s ) = e kite ( r s ) + 2 r s e kite ′ ( r s ) , e closed ( r s ) = 2 e kite ( r s ) + r s e kite ′ ( r s ) − e 2 x .
where the prime denotes d / d r s d/dr_s d / d r s . By the first relation, a term φ r s p \varphi r_s^p φ r s p of F F F appears in e k i t e e_{\mathrm{kite}} e kite multiplied by the Mellin multiplier M ( p ) = 2 / ( p + 2 ) M(p)=2/(p+2) M ( p ) = 2/ ( p + 2 ) of § 2.7 . The first relation is (12) with w D = 2 λ w_D=2\lambda w D = 2 λ and ϱ = λ r s \varrho=\lambda r_s ϱ = λ r s . Multiplying it by r s 2 r_s^2 r s 2 and differentiating with respect to r s r_s r s gives 2 r s e k i t e + r s 2 e k i t e ′ = 2 r s F 2r_se_{\mathrm{kite}}+r_s^2e_{\mathrm{kite}}'=2r_sF 2 r s e kite + r s 2 e kite ′ = 2 r s F , which is the second relation. Inserting the second relation into (44) gives the third. SI Sec. S12.4 explains why asymptotic forms are carried from F F F to e k i t e e_{\mathrm{kite}} e kite and not in the reverse direction.
For the kite, Y D = A 0 X Y_D=A_0X Y D = A 0 X by (40) , and we drop the subscript D D D : the radial kernel e D , Λ e_{D,\Lambda} e D , Λ of (16) becomes
K [ b Λ ] = ∫ 0 ∞ e Λ ( q ) d q , \mathcal{K}[b_\Lambda]=\int_0^\infty e_\Lambda(q)\,dq, K [ b Λ ] = ∫ 0 ∞ e Λ ( q ) d q ,
e Λ ( q ) = 4 π A 0 ∫ 0 ∞ d ν π Λ 2 Λ 2 + ν 2 / q 2 X ( q , ν ) , (46) e_\Lambda(q)=4\pi A_0\int_0^\infty\frac{d\nu}{\pi}\frac{\Lambda^2}{\Lambda^2+\nu^2/q^2}X(q,\nu), \tag*{(46)} e Λ ( q ) = 4 π A 0 ∫ 0 ∞ π d ν Λ 2 + ν 2 / q 2 Λ 2 X ( q , ν ) , (46)
with bare-line limit e ∞ ( q ) e_\infty(q) e ∞ ( q ) , whose integral is e 2 x e_{2x} e 2 x . The two endpoints of e Λ ( q ) e_\Lambda(q) e Λ ( q ) are as follows.
At small momentum transfer, q → 0 q\to0 q → 0 , the occupation difference confines the loop momenta to thin shells of width O ( q ) O(q) O ( q ) at the Fermi surface, and the Coulomb interaction between nearly coincident momenta produces a factor log ( 1 / q ) \log(1/q) log ( 1/ q ) . Inside the two shells the separation p − k \mathbf{p}-\mathbf{k} p − k has a component of order q q q normal to the Fermi surface and a tangential component r r r , so the Coulomb factor integrates schematically as ∫ d 2 r / ( r 2 + q 2 ) ≃ 2 π log ( 1 / q ) \int d^2r/(r^2+q^2)\simeq2\pi\log(1/q) ∫ d 2 r / ( r 2 + q 2 ) ≃ 2 π log ( 1/ q ) over q ≲ r ≲ 1 q\lesssim r\lesssim1 q ≲ r ≲ 1 . This factor log ( 1 / q ) \log(1/q) log ( 1/ q ) is the Fermi-surface factor of the radial kernel of the kite. At fixed z = ν / q z=\nu/q z = ν / q , with h ( z ) h(z) h ( z ) of (9) , the shell expansion gives
X ( q , q z ) = a x ( z ) log ( 1 / q ) + b x ( z ) + o ( 1 ) , X(q,qz)=a_x(z)\log(1/q)+b_x(z)+o(1), X ( q , q z ) = a x ( z ) log ( 1/ q ) + b x ( z ) + o ( 1 ) ,
a x ( z ) = 8 π 2 [ h ( z ) + z h ′ ( z ) ] , ∫ 0 ∞ d z π h ( z ) a x ( z ) = 4 π 2 3 ( 1 − log 2 ) , (47) a_x(z)=8\pi^2[h(z)+zh'(z)],\qquad\int_0^\infty\frac{dz}{\pi}h(z)a_x(z)=\frac{4\pi^2}{3}(1-\log2), \tag*{(47)} a x ( z ) = 8 π 2 [ h ( z ) + z h ′ ( z )] , ∫ 0 ∞ π d z h ( z ) a x ( z ) = 3 4 π 2 ( 1 − log 2 ) , (47)
where h ′ = d h / d z h'=dh/dz h ′ = d h / d z and o ( 1 ) o(1) o ( 1 ) refers to q → 0 q\to0 q → 0 at fixed z z z . The finite part b x ( z ) b_x(z) b x ( z ) of the shell expansion enters the energy only through the r s log r s r_s\log r_s r s log r s term (§ 3.3 ). The angular integrals that give a x a_x a x and the z z z integral are in SI Sec. S10.1, Eqs. (92) and (93) . Integrating (47) against the Lorentzian factor, with d ν = q d z d\nu=q\,dz d ν = q d z supplying the factor q q q , gives e Λ ( q ) = κ ( Λ ) q log ( 1 / q ) + O ( q ) e_\Lambda(q)=\kappa(\Lambda)q\log(1/q)+O(q) e Λ ( q ) = κ ( Λ ) q log ( 1/ q ) + O ( q ) with κ ( Λ ) = 4 π A 0 ∫ 0 ∞ ( d z / π ) Λ 2 a x ( z ) / ( Λ 2 + z 2 ) \kappa(\Lambda)=4\pi A_0\int_0^\infty(dz/\pi)\,\Lambda^2a_x(z)/(\Lambda^2+z^2) κ ( Λ ) = 4 π A 0 ∫ 0 ∞ ( d z / π ) Λ 2 a x ( z ) / ( Λ 2 + z 2 ) [closed form in SI Eq. (S26), where the coefficient of the O ( q ) O(q) O ( q ) term also follows from the pointwise expansion]. The frequency integration is performed first to control the integrated shell remainder (SI Sec. S16).
At large momentum transfer, q > 2 q>2 q > 2 , the unit Fermi ball B \mathbb{B} B and its translate B − q \mathbb{B}_{-\mathbf{q}} B − q by − q -\mathbf{q} − q do not overlap, and Δ p \Delta_p Δ p is + 1 +1 + 1 on the former and − 1 -1 − 1 on the latter (Figure 5 ). The product Δ p Δ k \Delta_p\Delta_k Δ p Δ k gives two same-ball regions , in which both loop momenta lie inside the same ball, and two different-ball regions , in which they lie in different balls. Each same-ball region has the internal Coulomb integral 4 π 2 4\pi^2 4 π 2 and excitation energies ± q 2 / 2 + O ( q ) \pm q^2/2+O(q) ± q 2 /2 + O ( q ) of common sign. The different-ball regions are suppressed by q − 2 q^{-2} q − 2 . In a same-ball region δ p ≃ δ k ≃ ± q 2 / 2 \delta_p\simeq\delta_k\simeq\pm q^2/2 δ p ≃ δ k ≃ ± q 2 /2 , so (17) with Ω = q Λ \Omega=q\Lambda Ω = q Λ gives the frequency integral of the two electron–hole factors, 2 w Λ / [ q 2 ( 1 + 2 w Λ ) 2 ] 2w_\Lambda/[q^2(1+2w_\Lambda)^2] 2 w Λ / [ q 2 ( 1 + 2 w Λ ) 2 ] with w Λ = Λ / q w_\Lambda=\Lambda/q w Λ = Λ/ q . Because both loop momenta stay in one ball, the Coulomb integral does not decrease with q q q . The kernel therefore decays only as q − 2 q^{-2} q − 2 . By (18) this same-ball term is absent for a static line, since both excitation energies have the same sign. For w Λ w_\Lambda w Λ bounded away from zero, the radial kernel at large momentum transfer is
Figure 5. Momentum regions of the kite loop for q > 2 q > 2 q > 2 (in units of k F k_{\mathrm{F}} k F ), where the Fermi ball B \mathbb{B} B and its translate B − q \mathbb{B}_{-q} B − q do not overlap. (a) Cross section through the axis of q q q , drawn for q = 3.2 q = 3.2 q = 3.2 . The occupation difference Δ p \Delta_p Δ p is + 1 +1 + 1 on B \mathbb{B} B and − 1 -1 − 1 on B − q \mathbb{B}_{-q} B − q , and the color gives the excitation energy δ p = p ⋅ q + q 2 / 2 \delta_p = \mathbf{p} \cdot\mathbf{q} + q^2/2 δ p = p ⋅ q + q 2 /2 in units of q 2 / 2 q^2/2 q 2 /2 . (b) The four regions of the two loop momenta. In the same-ball regions (shaded) the two excitation energies have the same sign and ∣ p − k ∣ ≤ 2 \lvert\mathbf{p} - \mathbf{k} \rvert\le2 ∣ p − k ∣ ≤ 2 . These regions contribute only when the line depends on frequency [18 ] . In the different-ball regions the signs are opposite and ∣ p − k ∣ ≃ q \lvert\mathbf{p} - \mathbf{k} \rvert\simeq q ∣ p − k ∣ ≃ q , so the Coulomb factor suppresses them by q − 2 q^{-2} q − 2 . The signs are exact for q > 2 q > 2 q > 2 , whereas δ p ≃ ± q 2 / 2 \delta_p \simeq\pm q^2/2 δ p ≃ ± q 2 /2 and ∣ p − k ∣ ≃ q \lvert\mathbf{p} - \mathbf{k} \rvert\simeq q ∣ p − k ∣ ≃ q denote the leading large-q q q behavior.
e Λ ( q ) = 3 π 3 q 2 ρ l o c ( w Λ ) + O ( 1 q 3 ( 1 + w Λ ) + 1 q 4 ) , ρ l o c ( w ) = 2 π w ( 1 + 2 w ) 2 . (48) e_\Lambda(q)=\frac{3}{\pi^3q^2}\rho_{\mathrm{loc}}(w_\Lambda)+O\left(\frac{1}{q^3(1+w_\Lambda)}+\frac{1}{q^4}\right),\qquad\rho_{\mathrm{loc}}(w)=\frac{2\pi w}{(1+2w)^2}. \tag*{(48)} e Λ ( q ) = π 3 q 2 3 ρ loc ( w Λ ) + O ( q 3 ( 1 + w Λ ) 1 + q 4 1 ) , ρ loc ( w ) = ( 1 + 2 w ) 2 2 π w . (48)
The frequency integral and the bounds on the different-ball regions are in SI Sec. S17.1. The factor ρ l o c \rho_{\mathrm{loc}} ρ loc is maximal at w Λ = 1 / 2 w_\Lambda=1/2 w Λ = 1/2 , where the spectral frequency q Λ q\Lambda q Λ equals the excitation frequency q 2 / 2 q^2/2 q 2 /2 .
The kite at high and low density
At high density, r s → 0 r_s \to0 r s → 0 , the scale is ε = q T F \varepsilon= q_{\mathrm{TF}} ε = q TF . We consider the coupling-averaged line, with c c c of (8). At small momentum transfer we set ν = q z \nu= qz ν = q z , and L ( q , q z ) → h ( z ) \mathcal{L}(q,qz) \to h(z) L ( q , q z ) → h ( z ) . Let f = 2 W ‾ − v = v [ ρ ( c ) − 1 ] f = 2\overline{W} - v = v[\rho(c) - 1] f = 2 W − v = v [ ρ ( c ) − 1 ] , so that ε k i t e − ε 2 x = K [ f ] \varepsilon_{\mathrm{kite}} - \varepsilon_{2x} = \mathcal{K}[f] ε kite − ε 2 x = K [ f ] by (42) and (43). With d 3 q = 4 π q 2 d q \mathrm{d}^3q = 4\pi q^2\mathrm{d}q d 3 q = 4 π q 2 d q and d ν = q d z \mathrm{d}\nu= q\,\mathrm{d}z d ν = q d z , (40) gives [cf. SI Eq. (S29)]
ε k i t e − ε 2 x ≃ 4 π A 0 ∫ 0 ∞ d z π ∫ 0 ∞ d q k z ( q ) g z ( q / q T F ) , (49) \varepsilon_{\mathrm{kite}} - \varepsilon_{2x} \simeq4\pi A_0 \int_0^\infty\frac{\mathrm{d}z}{\pi}\int_0^\infty\mathrm{d}q\, k_z(q)\,g_z(q/q_{\mathrm{TF}}), \tag*{(49)} ε kite − ε 2 x ≃ 4 π A 0 ∫ 0 ∞ π d z ∫ 0 ∞ d q k z ( q ) g z ( q / q TF ) , (49)
k z ( q ) = q X ( q , q z ) , g z ( t ) = ρ ( h ( z ) / t 2 ) − 1. k_z(q) = qX(q,qz), \qquad g_z(t) = \rho(h(z)/t^2) - 1. k z ( q ) = q X ( q , q z ) , g z ( t ) = ρ ( h ( z ) / t 2 ) − 1.
Here k z = k D / ( 4 π A 0 ) k_z = k_D/(4\pi A_0) k z = k D / ( 4 π A 0 ) with k D k_D k D of (24), and ≃ \simeq ≃ denotes equality up to the O ( r s ∣ log r s ∣ ) O(r_s\lvert\log r_s\rvert) O ( r s ∣ log r s ∣) corrections discussed after (50). The full denominator of the screened line is retained for q ≲ q T F q \lesssim q_{\mathrm{TF}} q ≲ q TF , as in the ring resummation of Gell-Mann and Brueckner [5 ] . Expanding ρ ( c ) \rho(c) ρ ( c ) in powers of c ∝ q T F 2 / q 2 c \propto q_{\mathrm{TF}}^2/q^2 c ∝ q TF 2 / q 2 would generate bare chain diagrams whose q q q integrals diverge at small q q q .
The scaling function (table 1) has the tail g z ≃ − 2 h ( z ) / ( 3 t 2 ) g_z \simeq-2h(z)/(3t^2) g z ≃ − 2 h ( z ) / ( 3 t 2 ) , so that g ~ z ( s ) ≃ 2 h ( z ) / [ 3 ( s − 2 ) ] \widetilde{g}_z(s) \simeq2h(z)/[3(s-2)] g z ( s ) ≃ 2 h ( z ) / [ 3 ( s − 2 )] near s = 2 s = 2 s = 2 , up to terms regular there. By (47) the radial kernel has the endpoint term k z ≃ a x ( z ) q log ( 1 / q ) k_z \simeq a_x(z)q\log(1/q) k z ≃ a x ( z ) q log ( 1/ q ) , so that k ~ z ( 1 − s ) ≃ a x ( z ) / ( s − 2 ) 2 \widetilde{k}_z(1-s) \simeq a_x(z)/(s-2)^2 k z ( 1 − s ) ≃ a x ( z ) / ( s − 2 ) 2 . The two poles coincide at s = 2 s = 2 s = 2 . The resulting triple pole contributes − 1 3 a x ( z ) h ( z ) q T F log 2 q T F -\frac{1}{3}a_x(z)h(z)q_{\mathrm{TF}}\log^2 q_{\mathrm{TF}} − 3 1 a x ( z ) h ( z ) q TF log 2 q TF to the q q q integral of (49) at fixed z z z (SI Sec. S14.6), with log q T F = 1 2 log r s + O ( 1 ) \log q_{\mathrm{TF}} = \frac{1}{2}\log r_s + O(1) log q TF = 2 1 log r s + O ( 1 ) . Integrating over z z z with ∫ 0 ∞ h a x d z / π = 4 π 2 ( 1 − log 2 ) / 3 \int_0^\infty h\,a_x\,\mathrm{d}z/\pi= 4\pi^2(1-\log2)/3 ∫ 0 ∞ h a x d z / π = 4 π 2 ( 1 − log 2 ) /3 and using q T F 2 = 4 α L r s / π q_{\mathrm{TF}}^{2}=4\alpha Lr_s/\pi q TF 2 = 4 α L r s / π from (7) , we obtain the energies in Ry,
e k i t e ( r s ) = e 2 x + c h d r s log 2 r s + O ( r s ∣ log r s ∣ ) , F ( r s ) = e 2 x + 3 2 c h d r s log 2 r s + O ( r s ∣ log r s ∣ ) , e c l o s e d ( r s ) = e 2 x + 3 c h d r s log 2 r s + O ( r s ∣ log r s ∣ ) , c h d = − α L ( 1 − log 2 ) 6 π 3 = − 0.0008594457 … . (50) \begin{aligned}
e_{\mathrm{kite}}(r_s) &= e_{2x}+c_{\mathrm{hd}}r_s\log^{2}r_s+O(r_s\lvert\log r_s\rvert),\\
F(r_s) &= e_{2x}+\frac{3}{2}c_{\mathrm{hd}}r_s\log^{2}r_s+O(r_s\lvert\log r_s\rvert),\\
e_{\mathrm{closed}}(r_s) &= e_{2x}+3c_{\mathrm{hd}}r_s\log^{2}r_s+O(r_s\lvert\log r_s\rvert),\\
c_{\mathrm{hd}} &= -\frac{\alpha L(1-\log2)}{6\pi^{3}}=-0.0008594457\ldots. \tag*{(50)}
\end{aligned} e kite ( r s ) F ( r s ) e closed ( r s ) c hd = e 2 x + c hd r s log 2 r s + O ( r s ∣ log r s ∣) , = e 2 x + 2 3 c hd r s log 2 r s + O ( r s ∣ log r s ∣) , = e 2 x + 3 c hd r s log 2 r s + O ( r s ∣ log r s ∣) , = − 6 π 3 α L ( 1 − log 2 ) = − 0.0008594457 … . (50)
SI Eqs. (S29)–(S32) give the window form of this calculation. Replacing h h h by L ( q , q z ) \mathcal{L}(q,q_z) L ( q , q z ) and controlling the regions q ≲ q T F q\lesssim q_{\mathrm{TF}} q ≲ q TF and q ≳ 1 q\gtrsim1 q ≳ 1 changes the energy only by O ( r s ∣ log r s ∣ ) O(r_s\lvert\log r_s\rvert) O ( r s ∣ log r s ∣) (SI Secs. S10.1 and S16). The coefficients of F F F and e c l o s e d e_{\mathrm{closed}} e closed are derived in SI Secs. S14.6 and S12.4. At full coupling the tail is − h / t 2 -h/t^2 − h / t 2 instead of − 2 h / ( 3 t 2 ) -2h/(3t^2) − 2 h / ( 3 t 2 ) , so F F F follows directly from the same triple pole. The ratio of the e k i t e e_{\mathrm{kite}} e kite coefficient to the F F F coefficient is the Mellin multiplier 2 / 3 2/3 2/3 at p = 1 p=1 p = 1 , and e c l o s e d = 2 F − e 2 x e_{\mathrm{closed}}=2F-e_{2x} e closed = 2 F − e 2 x doubles the r s log 2 r s r_s\log^2r_s r s log 2 r s correction of F F F . The finite part b x ( z ) b_x(z) b x ( z ) of the kernel gives k ~ z ( 1 − s ) \tilde{k}_z(1-s) k ~ z ( 1 − s ) a simple pole at the same point. This simple pole and the pole of g ~ z \tilde{g}_z g ~ z form a double pole, which gives an r s log r s r_s\log r_s r s log r s term. The coefficient of this term, c 1 = 0.01150 c_1=0.01150 c 1 = 0.01150 Ry for e k i t e e_{\mathrm{kite}} e kite , requires the finite shell parts and has been evaluated in the companion paper [50]. The Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) series of Sec. 2.9, 0.0130 r s log r s − 0.021 r s 0.0130r_s\log r_s-0.021r_s 0.0130 r s log r s − 0.021 r s Ry [9 ] and [7 Eq. (5)] , is a different quantity: it contains Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) as well and closes Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) with the line v / ( 1 + c ) v/(1+c) v / ( 1 + c ) of the ring series instead of 2 W ‾ 2\overline{W} 2 W .
Direct numerical integration of e k i t e e_{\mathrm{kite}} e kite at r s = 10 − 3 r_s=10^{-3} r s = 1 0 − 3 –10 − 1 10^{-1} 1 0 − 1 [50, 74] shows the r s log 2 r s r_s\log^2r_s r s log 2 r s term in this range. The values are listed in SI table S8. A weighted least-squares fit of the five values of e k i t e − e 2 x e_{\mathrm{kite}}-e_{2x} e kite − e 2 x (weights 1 / σ 2 1/\sigma^2 1/ σ 2 , with σ \sigma σ the quoted error of each value) to c f i t r s log 2 r s + c 1 , f i t r s log r s + c 2 r s c_{\mathrm{fit}}r_s\log^2r_s+c_{1,\mathrm{fit}}r_s\log r_s+c_{2}r_s c fit r s log 2 r s + c 1 , fit r s log r s + c 2 r s , with all three coefficients free, gives c f i t = − 0.00091 ( 5 ) c_{\mathrm{fit}}=-0.00091(5) c fit = − 0.00091 ( 5 ) Ry. The same fit to the four values with r s ≤ 0.03 r_s\leq0.03 r s ≤ 0.03 gives c f i t = − 0.00092 ( 12 ) c_{\mathrm{fit}}=-0.00092(12) c fit = − 0.00092 ( 12 ) Ry. Both agree with c h d c_{\mathrm{hd}} c hd . The weighted sum of squared residuals is 0.02 for two degrees of freedom. Without the r s log 2 r s r_s\log^2r_s r s log 2 r s term it rises to 380 for three degrees of freedom. With c f i t c_{\mathrm{fit}} c fit fixed at c h d c_{\mathrm{hd}} c hd , the fitted coefficient of r s log r s r_s\log r_s r s log r s is 0.01125 ( 6 ) 0.01125(6) 0.01125 ( 6 ) for r s ≤ 0.1 r_s\leq0.1 r s ≤ 0.1 and approaches c 1 c_1 c 1 as the fit range shrinks (within 1.4 σ 1.4\sigma 1.4 σ for r s ≤ 0.03 r_s\leq0.03 r s ≤ 0.03 ). This drift is expected from the terms of order r s 2 log k r s r_s^2\log^k r_s r s 2 log k r s omitted in the fit.
BRM fitted their Monte Carlo values for r s ≤ 1 r_s\leq1 r s ≤ 1 with e 2 x + C 1 r s log r s + C 2 r s + C 3 r s 2 log r s e_{2x}+C_1r_s\log r_s+C_2r_s+C_3r_s^2\log r_s e 2 x + C 1 r s log r s + C 2 r s + C 3 r s 2 log r s and obtained C 1 = 0.01212 C_1=0.01212 C 1 = 0.01212 Ry [Eq. (39) and Table I of Ref. [25]]. They regarded the closeness of this coefficient to 0.01304 Ry, the exchange part of the Carr–Maradudin coefficient of (36) , as a strong test of their calculation. For r s ≤ 1 r_s\leq1 r s ≤ 1 their values agree with the direct integration, for example 43.53(18) against 43.433(6) mRy at r s = 0.1 r_s=0.1 r s = 0.1 . With these error bars, however, a fit without the r s log 2 r s r_s\log^2r_s r s log 2 r s term absorbs that term into C 1 C_1 C 1 . The closeness of C 1 C_1 C 1 to 0.01304 is therefore not a test of the kite. This exchange part belongs to the complete first-order exchange polarization Π ( 1 ) \Pi^{(1)} Π ( 1 ) (Sec. 2.9). The isolated kite has c h d r s log 2 r s + c 1 r s log r s c_{\mathrm{hd}}r_s\log^2r_s+c_1r_s\log r_s c hd r s log 2 r s + c 1 r s log r s instead.
At low density, with the form (21) of the line, the scale is ϵ = q ∗ \epsilon=q_* ϵ = q ∗ . Inserting (21) into the kite functional gives K [ W λ ] = λ e 2 x − λ ∫ 0 ∞ θ ϵ Λ d q + … \mathcal{K}[W_\lambda]=\lambda e_{2x}-\lambda\int_{0}^{\infty}\theta\epsilon\Lambda\,dq+\ldots K [ W λ ] = λ e 2 x − λ ∫ 0 ∞ θ ϵ Λ d q + … . Here e Λ e_{\Lambda} e Λ is the radial kernel e Λ e_{\Lambda} e Λ of (46) at the q q q -dependent width Λ ( q ) \Lambda(q) Λ ( q ) of (21) and the dots denote the contribution of the remainder δ λ \delta\lambda δ λ . With e 2 x = ∫ 0 ∞ e ∞ d q e_{2x}=\int_{0}^{\infty}e_{\infty}\,dq e 2 x = ∫ 0 ∞ e ∞ d q , the right-hand side becomes λ ∫ 0 ∞ [ ( 1 − θ ) e ∞ + θ ( e ∞ − e Λ ) ] d q + … \lambda\int_{0}^{\infty}[(1-\theta)e_{\infty}+\theta(e_{\infty}-e_{\Lambda})]\,dq+\ldots λ ∫ 0 ∞ [( 1 − θ ) e ∞ + θ ( e ∞ − e Λ )] d q + … . The bare-line part ( 1 − θ ) e ∞ (1-\theta)e_{\infty} ( 1 − θ ) e ∞ is concentrated at q ≳ q ∗ q\gtrsim q_* q ≳ q ∗ , where e ∞ = O ( q − 4 ) e_{\infty}=O(q^{-4}) e ∞ = O ( q − 4 ) , and contributes only a remainder O ( q ∗ − 3 ) O(q_*^{-3}) O ( q ∗ − 3 ) (SI Sec. S18.2). In the second part the relevant momentum transfers q ∼ q ∗ ≫ 1 q\sim q_*\gg1 q ∼ q ∗ ≫ 1 lie where (48) holds. Since ρ l o c → 0 \rho_{\mathrm{loc}}\to0 ρ loc → 0 for w → ∞ w\to\infty w → ∞ , the integrand of the second part is − 3 Ψ Λ ( t ) / ( π 2 q 2 ) -3\Psi_\Lambda(t)/(\pi^2q^2) − 3 Ψ Λ ( t ) / ( π 2 q 2 ) with Ψ Λ ( t ) = θ Λ ρ l o c ( w Λ , λ ) \Psi_\Lambda(t)=\theta_\Lambda\rho_{\mathrm{loc}}(w_\Lambda,\lambda) Ψ Λ ( t ) = θ Λ ρ loc ( w Λ , λ ) , w Λ = Λ / q w_\Lambda=\Lambda/q w Λ = Λ/ q and t = q / q ∗ t=q/q_* t = q / q ∗ . The function Ψ Λ \Psi_\Lambda Ψ Λ depends on q q q only through t t t and vanishes as t 2 t^2 t 2 for t → 0 t\to0 t → 0 and as t − 4 t^{-4} t − 4 for t → ∞ t\to\infty t → ∞ . The substitution q = q ∗ x q=q_*x q = q ∗ x therefore gives the factor 1 / q ∗ 1/q_* 1/ q ∗ directly. In the language of the endpoint–residue correspondence (SI Proposition S3), this leading term is the residue of the simple pole at s = − 1 s=-1 s = − 1 that the q − 2 q^{-2} q − 2 endpoint term of the kernel gives k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) . With e k i t e = 2 ∫ 0 1 K [ W λ ] d λ e_{\mathrm{kite}}=2\int_0^1\mathcal{K}[W_\lambda]\,\mathrm{d}\lambda e kite = 2 ∫ 0 1 K [ W λ ] d λ of (43), θ λ = λ / ( x 4 + λ ) \theta_\lambda=\lambda/(x^4+\lambda) θ λ = λ / ( x 4 + λ ) and p l o c p_{\mathrm{loc}} p loc of (48) written out, the leading term is
e k i t e ( r s ) = A 1 q ∗ + O ( q ∗ − 3 / 2 ) , e_{\mathrm{kite}}(r_s)=\frac{A_1}{q_*}+O\left(q_*^{-3/2}\right), e kite ( r s ) = q ∗ A 1 + O ( q ∗ − 3/2 ) ,
A 1 = − 12 π 2 ∫ 0 1 λ d λ ∫ 0 ∞ d x x 2 λ x 4 + λ w Λ , λ ( x ) [ 1 + 2 w Λ , λ ( x ) ] 2 , (51) A_1=-\frac{12}{\pi^2}\int_0^1\lambda\,\mathrm{d}\lambda\int_0^\infty\frac{\mathrm{d}x}{x^2}\frac{\lambda}{x^4+\lambda}\frac{w_{\Lambda,\lambda}(x)}{\left[1+2w_{\Lambda,\lambda}(x)\right]^2}, \tag*{(51)} A 1 = − π 2 12 ∫ 0 1 λ d λ ∫ 0 ∞ x 2 d x x 4 + λ λ [ 1 + 2 w Λ , λ ( x ) ] 2 w Λ , λ ( x ) , (51)
w Λ , λ ( x ) = 1 2 1 + λ x − 4 , A 1 = − 4 Γ ( 1 / 4 ) 7 2 π 3 / 2 Γ ( 3 / 4 ) = − 0.2146937609 … . w_{\Lambda,\lambda}(x)=\frac{1}{2}\sqrt{1+\lambda x^{-4}},\qquad A_1=-\frac{4\Gamma(1/4)}{7\sqrt{2}\pi^{3/2}\Gamma(3/4)}=-0.2146937609\ldots. w Λ , λ ( x ) = 2 1 1 + λ x − 4 , A 1 = − 7 2 π 3/2 Γ ( 3/4 ) 4Γ ( 1/4 ) = − 0.2146937609 … .
With q ∗ − 1 = [ 3 π / ( 16 α L ) ] 1 / 4 r s − 1 / 4 = 1.0311348 … r s − 1 / 4 q_*^{-1}=[3\pi/(16\alpha_L)]^{1/4}r_s^{-1/4}=1.0311348\ldots\,r_s^{-1/4} q ∗ − 1 = [ 3 π / ( 16 α L ) ] 1/4 r s − 1/4 = 1.0311348 … r s − 1/4 , from (20) and q T F 2 = 4 α L r s / π q_{\mathrm{TF}}^2=4\alpha_Lr_s/\pi q TF 2 = 4 α L r s / π , the leading term is − 0.2213782 … r s − 1 / 4 -0.2213782\ldots\,r_s^{-1/4} − 0.2213782 … r s − 1/4 (Table 3 ). The substitution x = λ 1 / 4 y x=\lambda^{1/4}y x = λ 1/4 y separates the coupling integral ∫ 0 1 λ 3 / 4 d λ = 4 / 7 \int_0^1\lambda^{3/4}\,\mathrm{d}\lambda=4/7 ∫ 0 1 λ 3/4 d λ = 4/7 and leaves the beta function $B(5/4,1/2) (SI Sec. S17.3, Eqs. (161) and (162) , with the split (S98) written out). The coefficient agrees with the matched expansion of the companion paper [50]. That paper also gives the further terms and a comparison with numerical integration up to r s = 10 8 r_s=10^8 r s = 1 0 8 .
Table 3.
BRM inferred a decay as r s − 1 / 2 r_s^{-1/2} r s − 1/2 from their Monte Carlo values and built it into their fit form [Eq. (58) of Ref. [25]]. The leading term derived here, however, is of order r s − 1 / 4 r_s^{-1/4} r s − 1/4 . Their limit of a vanishing kite as r s → ∞ r_s\to\infty r s → ∞ [their Eq. (38) ] agrees with (51) . At r s = 10 8 r_s=10^8 r s = 1 0 8 the direct value, − 2.22 ( 21 ) -2.22(21) − 2.22 ( 21 ) mRy, is consistent with A 1 / q ∗ = − 2.214 A_1/q_*=-2.214 A 1 / q ∗ = − 2.214 mRy [50, 74]. The Monte Carlo values of BRM at r s = 2 × 10 5 r_s=2\times10^5 r s = 2 × 1 0 5 and 10 6 10^6 1 0 6 , − 7.13 -7.13 − 7.13 and − 3.89 -3.89 − 3.89 mRy [25 ] , lie above the direct values − 9.13 -9.13 − 9.13 and − 6.35 -6.35 − 6.35 mRy. BRM sampled their Eqs. (28) and (29) with 10 9 10^9 1 0 9 points. The integration over the momentum transfer in these equations extends over all space. Their paper does not describe how this integration was mapped or truncated. The paper alone therefore does not determine the origin of the difference. The integrand extends to momentum transfers of order q ∗ k F ≈ 20 k F q_*k_F\approx20k_F q ∗ k F ≈ 20 k F and 31 k F 31k_F 31 k F at these densities. A momentum cutoff near 12 k F 12k_F 12 k F in the direct integration reproduces the difference [50]. Such a cutoff is one possible origin of the difference, not an identified error.
The remainder bound O ( q ∗ − 3 / 2 ) O(q_*^{-3/2}) O ( q ∗ − 3/2 ) is proved in SI Sec. S18.2. The bounds of that section hold at each fixed coupling. The x x x integral of (51) at λ = 1 \lambda=1 λ = 1 therefore gives F ( r s ) = 7 8 A 1 / q ∗ + O ( q ∗ − 3 / 2 ) F(r_s)=\frac{7}{8}A_1/q_*+O(q_*^{-3/2}) F ( r s ) = 8 7 A 1 / q ∗ + O ( q ∗ − 3/2 ) , consistent with the Mellin multiplier 2 / ( p + 2 ) = 8 / 7 2/(p+2)=8/7 2/ ( p + 2 ) = 8/7 at p = − 1 / 4 p=-1/4 p = − 1/4 . In the chain family, e c l o s e d = 2 F − e 2 x e_{\mathrm{closed}}=2F-e_{2x} e closed = 2 F − e 2 x tends to − e 2 x -e_{2x} − e 2 x as F → 0 F\to0 F → 0 (Table 3 ). This constant belongs to the chain family and is expected to be removed once both kite lines are screened (SI Sec. S12.3).
Table 3 : Leading energy terms in Ry per electron for the kite alone, in the three energy expressions of § 3.2 (full-coupling line, BRM coupling average, chain family). High density: correction to the bare value e 2 x e_{2x} e 2 x of (42), with c h d c_{\mathrm{hd}} c hd of (50). Low density: the energy, a formal limit (§ 1 ). The r s − 1 / 4 r_s^{-1/4} r s − 1/4 terms are A 1 / q ∗ = − 0.2213782 … r s − 1 / 4 A_1/q_*=-0.2213782\ldots\,r_s^{-1/4} A 1 / q ∗ = − 0.2213782 … r s − 1/4 of (51) for e k i t e e_{\mathrm{kite}} e kite , 7 / 8 7/8 7/8 of it for F F F and 7 / 4 7/4 7/4 of it for e c l o s e d e_{\mathrm{closed}} e closed , by (45). Derivations are in SI Sec. S12.4.
Cancellation by the Fock self-energy corrections
The kite is obtained by closing the vertex part Π v t x ( 1 ) \Pi_{\mathrm{vtx}}^{(1)} Π vtx ( 1 ) of the first-order exchange polarization Π ( 1 ) \Pi^{(1)} Π ( 1 ) with a screened line. This polarization also contains a Fock self-energy insertion on each propagator of the bubble [41 ] [42] (Figure 4 ). At fixed density the insertion is G 0 ( Σ x − δ μ x ) G 0 G_{0}(\Sigma_{\mathrm{x}}-\delta\mu_{\mathrm{x}})G_{0} G 0 ( Σ x − δ μ x ) G 0 , where Σ x ( k ) \Sigma_{\mathrm{x}}(k) Σ x ( k ) is the Fock self-energy at momentum k k k [SI Eq. (S19)]. The chemical-potential shift δ μ x = Σ x ( 1 ) \delta\mu_{\mathrm{x}}=\Sigma_{\mathrm{x}}(1) δ μ x = Σ x ( 1 ) is the self-energy at the Fermi surface k = 1 k=1 k = 1 and keeps the Fermi surface at its free position. The sum Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) of the two Fock self-energy insertions [explicit form in SI Eq. (S21)], the self-energy pair , is closed with the line f f f in the kite normalization,
K Π S E ( 1 ) [ f ] = − 3 π 3 q T F 4 ∫ d 3 q ∫ 0 ∞ d ν π f ( q , i ν ) Π S E ( 1 ) ( q , i ν ) , (52) \mathcal{K}_{\Pi_{\mathrm{SE}}^{(1)}}[f] = -\frac{3}{\pi^{3}q_{\mathrm{TF}}^{4}}\int\mathrm{d}^{3}q\int_{0}^{\infty}\frac{\mathrm{d}\nu}{\pi}f(q,i\nu)\Pi_{\mathrm{SE}}^{(1)}(q,i\nu), \tag*{(52)} K Π SE ( 1 ) [ f ] = − π 3 q TF 4 3 ∫ d 3 q ∫ 0 ∞ π d ν f ( q , i ν ) Π SE ( 1 ) ( q , i ν ) , (52)
which defines K Π [ f ] \mathcal{K}_{\Pi}[f] K Π [ f ] for any insertion Π \Pi Π closed with f f f in the kite normalization. Its integrand is Y D Y_{\mathrm{D}} Y D of § 2.3 with S D = 4 S_{\mathrm{D}}=4 S D = 4 . The closed diagram with one Fock self-energy insertion has the coefficient 1 / 2 1/2 1/2 . Its two symmetries, in the sense of SI Sec. S12.1, are the identity and the rotation of the loop that exchanges its two interaction lines. Since Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) contains two insertions that both close to this diagram, the coefficient for the pair is 1 / 4 1/4 1/4 . In this normalization K Π v t x ( 1 ) = K \mathcal{K}_{\Pi_{\mathrm{vtx}}^{(1)}}=\mathcal{K} K Π vtx ( 1 ) = K (SI Sec. S9.2).
In energy expressions of the RPA-with-exchange (RPAx) type [75] the first-order exchange polarization Π ( 1 ) \Pi^{(1)} Π ( 1 ) enters closed with λ v / ϵ λ 2 \lambda v/\epsilon_{\lambda}^{2} λ v / ϵ λ 2 instead of W λ W_{\lambda} W λ , that is, through the factor λ Π ( 1 ) / ϵ λ 2 \lambda\Pi^{(1)}/\epsilon_{\lambda}^{2} λ Π ( 1 ) / ϵ λ 2 . Since ϵ λ = 1 + λ c \epsilon_{\lambda}=1+\lambda c ϵ λ = 1 + λ c by (5) and (8), this line has the scaling function ( 1 + λ c ) − 2 − 1 (1+\lambda c)^{-2}-1 ( 1 + λ c ) − 2 − 1 , which also has a t − 2 t^{-2} t − 2 tail. The vertex and self-energy kernels have opposite endpoint terms, so the corresponding poles cancel for any common line (§ 2.7 ). The leading cancellations derived below therefore also hold for this line. The self-consistency of a conserving approximation [46] and the terms from the Legendre transform between fixed chemical potential and fixed particle number are not considered.
To see why the self-energy corrections are needed, note that the correlation energy is determined by the density response. By the adiabatic-connection fluctuation–dissipation theorem [76, 77], the correlation energy E c E_{\mathrm{c}} E c per unit volume, in the units of § 2.2 , is
E c = − 1 2 ∫ 0 1 d λ ∫ d 3 q ( 2 π ) 3 v ( q ) ∫ 0 ∞ d ν π [ χ λ ( q , i ν ) − χ 0 ( q , i ν ) ] . (53) E_{\mathrm{c}}=-\frac{1}{2}\int_{0}^{1}\mathrm{d}\lambda\int\frac{\mathrm{d}^{3}q}{(2\pi)^{3}}v(q)\int_{0}^{\infty}\frac{\mathrm{d}\nu}{\pi}\left[\chi_{\lambda}(q,i\nu)-\chi_{0}(q,i\nu)\right]. \tag*{(53)} E c = − 2 1 ∫ 0 1 d λ ∫ ( 2 π ) 3 d 3 q v ( q ) ∫ 0 ∞ π d ν [ χ λ ( q , i ν ) − χ 0 ( q , i ν ) ] . (53)
Multiplying this energy density by the volume 3 π 2 3\pi^{2} 3 π 2 per electron and by E 0 = 32 / ( π 2 q T F 4 ) E_{0}=32/(\pi^{2}q_{\mathrm{TF}}^{4}) E 0 = 32/ ( π 2 q TF 4 ) Ry converts it to Ry per electron. An approximation to E c E_{\mathrm{c}} E c is therefore an approximation to χ λ \chi_{\lambda} χ λ . By (5), a change δ Π λ \delta\Pi_{\lambda} δ Π λ of the proper response changes the full response, to first order, by δ χ λ = δ Π λ / ϵ λ 2 \delta\chi_{\lambda}=\delta\Pi_{\lambda}/\epsilon_{\lambda}^{2} δ χ λ = δ Π λ / ϵ λ 2 . The resulting change of the correlation energy is
δ E c = − 1 2 ∫ 0 1 d λ ∫ d 3 q ( 2 π ) 3 ∫ 0 ∞ d ν π v ( q ) ϵ λ ( q , i ν ) 2 δ Π λ ( q , i ν ) . (54) \delta E_{\mathrm{c}}=-\frac{1}{2}\int_{0}^{1}\mathrm{d}\lambda\int\frac{\mathrm{d}^{3}q}{(2\pi)^{3}}\int_{0}^{\infty}\frac{\mathrm{d}\nu}{\pi}\frac{v(q)}{\epsilon_{\lambda}(q,i\nu)^{2}}\delta\Pi_{\lambda}(q,i\nu). \tag*{(54)} δ E c = − 2 1 ∫ 0 1 d λ ∫ ( 2 π ) 3 d 3 q ∫ 0 ∞ π d ν ϵ λ ( q , i ν ) 2 v ( q ) δ Π λ ( q , i ν ) . (54)
For first-order exchange, δ Π λ = λ Π ( 1 ) \delta\Pi_{\lambda}=\lambda\Pi^{(1)} δ Π λ = λ Π ( 1 ) with Π ( 1 ) = Π v t x ( 1 ) + Π S E ( 1 ) \Pi^{(1)}=\Pi_{\mathrm{vtx}}^{(1)}+\Pi_{\mathrm{SE}}^{(1)} Π ( 1 ) = Π vtx ( 1 ) + Π SE ( 1 ) , and (54) is the RPAx form above. The energy expressions of § 2.3 close Π ( 1 ) \Pi^{(1)} Π ( 1 ) with other screened lines and coupling weights. These lines share a t − 2 t^{-2} t − 2 Thomas–Fermi tail at high density and the suppression below the plasma frequency at low density. Evaluating the kite alone amounts to δ Π λ = λ Π v t x ( 1 ) \delta\Pi_{\lambda}=\lambda\Pi_{\mathrm{vtx}}^{(1)} δ Π λ = λ Π vtx ( 1 ) . The exact response obeys two constraints that this choice violates.
The Ward identity, which expresses charge conservation, ties the vertex correction to the self-energy corrections. At small momentum transfer their terms in log ( 1 / q ) \log(1/q) log ( 1/ q ) are equal and
opposite, as shown in (59) below. In (54) the measure is d 3 q d ν = 4 π q 3 d q d z d^{3}q\,d\nu= 4\pi q^{3}\,dq\,dz d 3 q d ν = 4 π q 3 d q d z , the bare line is q − 2 q^{-2} q − 2 , and for q ≫ q T F q \gg q_{\mathrm{TF}} q ≫ q TF the screening changes it by a relative amount of order q T F 2 h ( z ) / q 2 q_{\mathrm{TF}}^{2}h(z)/q^{2} q TF 2 h ( z ) / q 2 . The vertex part behaves as Π v t x ( 1 ) ∝ q T F 4 log ( 1 / q ) \Pi_{\mathrm{vtx}}^{(1)} \propto q_{\mathrm{TF}}^{4}\log(1/q) Π vtx ( 1 ) ∝ q TF 4 log ( 1/ q ) , and the conversion to Ry per electron (Sec. 2.8) supplies the factor q T F − 4 q_{\mathrm{TF}}^{-4} q TF − 4 of E 0 = 32 / ( π 2 q T F ) E_{0}=32/(\pi^{2}q_{\mathrm{TF}}) E 0 = 32/ ( π 2 q TF ) Ry. Up to the z z z integral and in the window q T F ≪ q ≪ 1 q_{\mathrm{TF}}\ll q\ll1 q TF ≪ q ≪ 1 of Sec. 2.8, the screening correction to the energy per electron therefore behaves as
q T F 2 ∫ q T F 1 log ( 1 / q ) q d q = 1 2 q T F 2 log 2 q T F ∝ r s log 2 r s . (55) q_{\mathrm{TF}}^{2}\int_{q_{\mathrm{TF}}}^{1}\frac{\log(1/q)}{q}\,dq=\tfrac{1}{2}q_{\mathrm{TF}}^{2}\log^{2}q_{\mathrm{TF}}\propto r_{s}\log^{2}r_{s}. \tag*{(55)} q TF 2 ∫ q TF 1 q log ( 1/ q ) d q = 2 1 q TF 2 log 2 q TF ∝ r s log 2 r s . (55)
For the conserving sum, Π ( 1 ) \Pi^{(1)} Π ( 1 ) tends to the finite limit − ( q T F 4 / 16 ) R ( 1 ) ( z ) -(q_{\mathrm{TF}}^{4}/16)R^{(1)}(z) − ( q TF 4 /16 ) R ( 1 ) ( z ) . The same integral without the factor log ( 1 / q ) \log(1/q) log ( 1/ q ) then gives q T F 2 log ( 1 / q T F ) ∝ r s log r s q_{\mathrm{TF}}^{2}\log(1/q_{\mathrm{TF}})\propto r_{s}\log r_{s} q TF 2 log ( 1/ q TF ) ∝ r s log r s , the form of Sec. 2.9. The r s log 2 r s r_{s}\log^{2}r_{s} r s log 2 r s term of the kite is thus the signature in E c E_{c} E c of a response that violates charge conservation.
(ii) In the f f f -sum rule [78, Sec. 2.2], the double commutator [ [ ρ ^ q , H ^ λ ] , ρ ^ − q ] [[\hat{\rho}_{q},\hat{H}_{\lambda}],\hat{\rho}_{-q}] [[ ρ ^ q , H ^ λ ] , ρ ^ − q ] , with H ^ λ = T ^ + λ V ^ \hat{H}_{\lambda}=\hat{T}+\lambda\hat{V} H ^ λ = T ^ + λ V ^ of Sec. 2.2 and the density operator ρ ^ q = ∑ j e − i q ⋅ r j \hat{\rho}_{q}=\sum_{j}e^{-i\mathbf{q}\cdot\mathbf{r}_{j}} ρ ^ q = ∑ j e − i q ⋅ r j , contains only the kinetic energy, because the interaction commutes with the density. Through the spectral representation of χ λ \chi_{\lambda} χ λ , the double commutator fixes the coefficient of − ν − 2 -\nu^{-2} − ν − 2 in the large-frequency expansion of the response at n e q 2 / m e n_{e}q^{2}/m_{e} n e q 2 / m e (SI Sec. S14.4). With n e = k F 3 / ( 3 π 2 ) n_{e}=k_{F}^{3}/(3\pi^{2}) n e = k F 3 / ( 3 π 2 ) and q T F 2 = 4 / ( π a B k F ) q_{\mathrm{TF}}^{2}=4/(\pi a_{B}k_{F}) q TF 2 = 4/ ( π a B k F ) , in the units of Sec. 2.2 this gives, for every λ \lambda λ ,
χ λ ( q , i ν ) ≃ − q T F 2 F q 2 3 ν 2 , ν → ∞ , (56) \chi_{\lambda}(q,i\nu)\simeq-\frac{q_{\mathrm{TF}}^{2}Fq^{2}}{3\nu^{2}},\qquad\nu\to\infty, \tag*{(56)} χ λ ( q , i ν ) ≃ − 3 ν 2 q TF 2 F q 2 , ν → ∞ , (56)
which is the free value. Since χ λ − Π λ = O ( ν − 4 ) \chi_{\lambda}-\Pi_{\lambda}=O(\nu^{-4}) χ λ − Π λ = O ( ν − 4 ) , the ν − 2 \nu^{-2} ν − 2 tail of δ Π λ \delta\Pi_{\lambda} δ Π λ must vanish in a complete set of corrections. At low density and q ≪ q ∗ q\ll q_{*} q ≪ q ∗ the factor 1 / ϵ λ 1/\epsilon_{\lambda} 1/ ϵ λ can be replaced by ν 2 / ( ν 2 + λ ω p 2 ) \nu^{2}/(\nu^{2}+\lambda\omega_{p}^{2}) ν 2 / ( ν 2 + λ ω p 2 ) , with ω p \omega_{p} ω p of (20) . This replacement holds because ϵ λ = 1 + λ c \epsilon_{\lambda}=1+\lambda c ϵ λ = 1 + λ c is the RPA dielectric function and c = q T F 2 L / q 2 ≃ q T F 2 / ( 3 ν 2 ) = ω p 2 / ν 2 c=q_{\mathrm{TF}}^{2}\mathcal{L}/q^{2}\simeq q_{\mathrm{TF}}^{2}/(3\nu^{2})=\omega_{p}^{2}/\nu^{2} c = q TF 2 L / q 2 ≃ q TF 2 / ( 3 ν 2 ) = ω p 2 / ν 2 for ν ≫ q + q 2 / 2 \nu\gg q+q^{2}/2 ν ≫ q + q 2 /2 , the largest particle–hole energy at momentum transfer q q q . The relative error of this high-frequency form is at most ( q + q 2 / 2 ) 2 / ν 2 (q+q^{2}/2)^{2}/\nu^{2} ( q + q 2 /2 ) 2 / ν 2 . The high-frequency form is the free value of (56) , and it follows from the representation of L \mathcal{L} L and the f f f -sum identity in SI Sec. S17.4. For q ≪ q ∗ q\ll q_{*} q ≪ q ∗ its range of validity contains the frequencies ν ∼ ω p = q ∗ 2 / 2 \nu\sim\omega_{p}=q_{*}^{2}/2 ν ∼ ω p = q ∗ 2 /2 at which the factor changes. For ν ≲ q \nu\lesssim q ν ≲ q , both 1 / ϵ λ 1/\epsilon_{\lambda} 1/ ϵ λ and its replacement are O ( q 2 / ( λ q T F 2 ) ) O(q^{2}/(\lambda q_{\mathrm{TF}}^{2})) O ( q 2 / ( λ q TF 2 )) [SI Eq. (S105)]. The line v / ϵ λ 2 v/\epsilon_{\lambda}^{2} v / ϵ λ 2 in (54) thus suppresses the frequencies ν ≲ ω p \nu\lesssim\omega_{p} ν ≲ ω p . A tail A / ν 2 A/\nu^{2} A / ν 2 of δ Π λ \delta\Pi_{\lambda} δ Π λ , with A A A independent of ν \nu ν , then contributes to the frequency integral of (54)
∫ 0 ∞ d ν π ν 4 ( ν 2 + λ ω p 2 ) 2 A ν 2 = A 4 λ ω p ∝ A q ∗ 2 , (57) \int_{0}^{\infty}\frac{d\nu}{\pi}\frac{\nu^{4}}{(\nu^{2}+\lambda\omega_{p}^{2})^{2}}\frac{A}{\nu^{2}}=\frac{A}{4\sqrt{\lambda}\omega_{p}}\propto\frac{A}{q_{*}^{2}}, \tag*{(57)} ∫ 0 ∞ π d ν ( ν 2 + λ ω p 2 ) 2 ν 4 ν 2 A = 4 λ ω p A ∝ q ∗ 2 A , (57)
at fixed q q q . When the tail is integrable in q q q , this contribution gives the universal pole at s = − 2 s=-2 s = − 2 [cf. (25) ]. For the kite alone the high-frequency tail does not vanish. It tends to a constant at large q q q and is therefore not integrable, so the leading term is instead the kernel pole at s = − 1 s=-1 s = − 1 , the r s − 1 / 4 r_{s}^{-1/4} r s − 1/4 term of Sec. 3.3. In the sum with Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) the high-frequency tails cancel for every q q q [SI Eq. (98) ]. The cancellation of the r s − 1 / 4 r_{s}^{-1/4} r s − 1/4 term is established below from the combined kernel.
An energy computed from a response that violates one of these two constraints thus contains a term that the constraint removes once the accompanying diagrams of the same order are included. Closed with the bare line, the self-energy insertions give zero, as shown below. The second-order exchange energy is therefore complete by itself, and the insertions are needed only when the line is screened.
In terms of reference states, the Fock self-energy insertions are the first-order change of the orbital energies. A Kohn–Sham reference with a local potential shifts all orbital energies by a constant and lacks the insertions. The dynamically screened one-line kite evaluated on such a reference therefore retains the r s log 2 r s r_s \log^2 r_s r s log 2 r s and r s − 1 / 4 r_s^{-1/4} r s − 1/4 terms, whereas Hartree–Fock orbital energies contain the insertions to all orders. SI Sec. S10.4 discusses these references, other SOSEX variants [79, 80, 81, 82, 75, 83, 84, 85], and a related cancellation in the structure factor observed by Yasuhara et al. [86].
The cancellation follows from the frequency denominators. With g p = ( i ν − δ p ) − 1 g_p=(i\nu-\delta_p)^{-1} g p = ( i ν − δ p ) − 1 and g k = ( i ν − δ k ) − 1 g_k=(i\nu-\delta_k)^{-1} g k = ( i ν − δ k ) − 1 , the self-energy shift Δ Σ x \Delta\Sigma_x Δ Σ x of the excitation energy, an integral over the second loop momentum, turns g p g_p g p into g p + Δ Σ x g p 2 g_p+\Delta\Sigma_xg_p^2 g p + Δ Σ x g p 2 to first order (SI Sec. S9.2). Since Δ Σ x ( p ) = Σ x ( p + q ) − Σ x ( p ) \Delta\Sigma_x(\mathbf{p})=\Sigma_x(\mathbf{p}+\mathbf{q})-\Sigma_x(\mathbf{p}) Δ Σ x ( p ) = Σ x ( p + q ) − Σ x ( p ) is itself an integral over k \mathbf{k} k of Δ k ∣ p − k ∣ − 2 \Delta_k|\mathbf{p}-\mathbf{k}|^{-2} Δ k ∣ p − k ∣ − 2 , the self-energy and vertex terms share the factor Δ p Δ k ∣ p − k ∣ − 2 \Delta_p\Delta_k|\mathbf{p}-\mathbf{k}|^{-2} Δ p Δ k ∣ p − k ∣ − 2 and differ only in their frequency factors. This common factor is the form the Ward identity takes here. After symmetrization in p ↔ k \mathbf{p}\leftrightarrow\mathbf{k} p ↔ k , the self-energy terms give g p 2 + g k 2 g_p^2+g_k^2 g p 2 + g k 2 and the vertex term gives − 2 g p g k -2g_pg_k − 2 g p g k :
g p 2 + g k 2 − 2 g p g k = ( g p − g k ) 2 = [ q ⋅ ( p − k ) ] 2 ( i ν − δ p ) 2 ( i ν − δ k ) 2 . (58) g_p^2+g_k^2-2g_pg_k=(g_p-g_k)^2=\frac{[\mathbf{q}\cdot(\mathbf{p}-\mathbf{k})]^2}{(i\nu-\delta_p)^2(i\nu-\delta_k)^2}. \tag*{(58)} g p 2 + g k 2 − 2 g p g k = ( g p − g k ) 2 = ( i ν − δ p ) 2 ( i ν − δ k ) 2 [ q ⋅ ( p − k ) ] 2 . (58)
The numerator vanishes quadratically at coincident loop momenta. This quadratic zero removes the Coulomb singularity ∣ p − k ∣ − 2 |\mathbf{p}-\mathbf{k}|^{-2} ∣ p − k ∣ − 2 from the combined integrand. The numerator also suppresses the same-ball regions by q − 2 q^{-2} q − 2 . The self-energy radial kernel therefore has the two endpoint terms of the kite with opposite coefficients. At small momentum transfer the cusp ∝ u log ∣ u ∣ \propto u\log|u| ∝ u log ∣ u ∣ of the Fock self-energy at the Fermi surface, where ∣ p ∣ = 1 + u |\mathbf{p}|=1+u ∣ p ∣ = 1 + u , integrated across a shell of width O ( q ) O(q) O ( q ) , gives
Π v t x ( 1 ) ( q , i q z ) = − q T F 4 4 [ h + z h ′ ] log 1 q + O ( q T F 4 ) , Π S E ( 1 ) ( q , i q z ) = + q T F 4 4 [ h + z h ′ ] log 1 q + O ( q T F 4 ) . (59) \Pi_{\mathrm{vtx}}^{(1)}(q,iq_z)=-\frac{q_{\mathrm{TF}}^4}{4}[h+zh']\log\frac{1}{q}+O(q_{\mathrm{TF}}^4),\qquad\Pi_{\mathrm{SE}}^{(1)}(q,iq_z)=+\frac{q_{\mathrm{TF}}^4}{4}[h+zh']\log\frac{1}{q}+O(q_{\mathrm{TF}}^4). \tag*{(59)} Π vtx ( 1 ) ( q , i q z ) = − 4 q TF 4 [ h + z h ′ ] log q 1 + O ( q TF 4 ) , Π SE ( 1 ) ( q , i q z ) = + 4 q TF 4 [ h + z h ′ ] log q 1 + O ( q TF 4 ) . (59)
Here h = h ( z ) h=h(z) h = h ( z ) and h ′ = d h / d z h'=dh/dz h ′ = d h / d z as in (47) , so that the vertex term is − q T F 4 a x ( z ) log ( 1 / q ) / ( 32 π 2 ) -q_{\mathrm{TF}}^4a_x(z)\log(1/q)/(32\pi^2) − q TF 4 a x ( z ) log ( 1/ q ) / ( 32 π 2 ) . The remainders refer to q → 0 q\to0 q → 0 at fixed z z z . The two terms in log ( 1 / q ) \log(1/q) log ( 1/ q ) are derived in SI Secs. S10.1 and S10.2 [Eqs. (82) , (83) , (92) , (95) and (96) ]. Their cancellation persists after the frequency integration, direction by direction [SI Eq. (S35)]. At large momentum transfer the self-energy radial kernel, defined as in (16) , is ϵ Σ , Λ ( q ) = − 3 p l o c ( w Λ ) / ( π 3 q 2 ) + O ( q − 4 ) \epsilon_{\Sigma,\Lambda}(q)=-3p_{\mathrm{loc}}(w_\Lambda)/(\pi^3q^2)+O(q^{-4}) ϵ Σ , Λ ( q ) = − 3 p loc ( w Λ ) / ( π 3 q 2 ) + O ( q − 4 ) , the term of (48) with opposite sign (SI Secs. S17.4 and S19).
The same balance between vertex and self-energy terms underlies the G W Γ GW\Gamma G W Γ scheme of Takada [87, 48], which adds a vertex function Γ \Gamma Γ to the G W GW G W approximation. This scheme constructs the density vertex as a ratio of inverse propagators consistent with the Ward identity and describes the remaining part by an electron–hole interaction. At first order, the Ward part of this construction, the part fixed by the Ward identity, is the summand − ( g p 2 + g k 2 ) -(g_p^2+g_k^2) − ( g p 2 + g k 2 ) of the vertex term − 2 g p g k -2g_pg_k − 2 g p g k in (58) . This part cancels the self-energy terms g p 2 + g k 2 g_p^2+g_k^2 g p 2 + g k 2 , and the remainder ( g p − g k ) 2 (g_p-g_k)^2 ( g p − g k ) 2 gives the finite limit R ( 1 ) R^{(1)} R ( 1 ) of Sec. 2.9 (SI Sec. S10.3).
The bare-line value K Π S E ( 1 ) [ ν ] K_{\Pi_{\mathrm{SE}}^{(1)}[\nu]} K Π SE ( 1 ) [ ν ] vanishes, consistent with second-order perturbation theory, whose energy consists of ring and exchange diagrams only. With one line screened, the same topology is the second diagram in Fig. 6(b) of Ref. [25 ] , which BRM set to zero because it contains two free propagators with the same four-momentum. For a frequency-independent insertion the frequency integral of G 0 ( k ) 2 G_0(k)^2 G 0 ( k ) 2 leaves only a term localized at the Fermi surface. This term vanishes for the fixed-density insertion Σ x − δ μ x \Sigma_x-\delta\mu_x Σ x − δ μ x , so the closure with the bare line vanishes [SI Eq. (S84)]. With W λ W_\lambda W λ the insertion depends on frequency, and its closure does not vanish in general. The diagram then has the topology of the self-energy pair closed with the screened line.
By the criterion of § 2.7 , the opposite endpoint terms remove the triple pole at s = 2 s=2 s = 2 and the pole at s = − 1 s=-1 s = − 1 that produce the leading terms of the kite. With e S E = 2 ∫ 0 1 K Π S E ( 1 ) [ W λ ] d λ e_{\mathrm{SE}}=2\int_0^1 K_{\Pi_{\mathrm{SE}}^{(1)}}[W_\lambda]\,\mathrm{d}\lambda e SE = 2 ∫ 0 1 K Π SE ( 1 ) [ W λ ] d λ ,
e S E ( r s ) = − A 1 q ∗ + O ( q ∗ − 3 / 2 ) , e k i t e ( r s ) + e S E ( r s ) = O ( q ∗ − 3 / 2 ) = O ( r s − 3 / 8 ) (60) e_{\mathrm{SE}}(r_s)=-\frac{A_1}{q_*}+O(q_*^{-3/2}), \qquad e_{\mathrm{kite}}(r_s)+e_{\mathrm{SE}}(r_s)=O(q_*^{-3/2})=O(r_s^{-3/8}) \tag*{(60)} e SE ( r s ) = − q ∗ A 1 + O ( q ∗ − 3/2 ) , e kite ( r s ) + e SE ( r s ) = O ( q ∗ − 3/2 ) = O ( r s − 3/8 ) (60)
(SI Sec. S17.4 bounds the remainders). At high density the two r s log 2 r s r_s\log^2r_s r s log 2 r s terms cancel in the same way. Since K Π S E ( 1 ) [ v ] = 0 K_{\Pi_{\mathrm{SE}}^{(1)}}[v]=0 K Π SE ( 1 ) [ v ] = 0 , the high-density bounds of § 3.3 and SI Sec. S10.2 leave e k i t e + e S E = e 2 x + O ( r s ∣ log r s ∣ ) e_{\mathrm{kite}}+e_{\mathrm{SE}}=e_{2x}+O(r_s|\log r_s|) e kite + e SE = e 2 x + O ( r s ∣ log r s ∣ ) . The high-frequency tail of the sum vanishes for every q q q [SI Eq. (S36)], whereas the tails of the two corrections are not separately integrable. The vanishing of the summed tail is the tail condition of SI Lemma S1 (§ 2.7 ). However, the remainder condition of the lemma is not verified here for the summed kernel. The absence of the universal pole at s = − 2 s=-2 s = − 2 therefore follows from the lemma only under that condition. The proved bound O ( q ∗ − 3 / 2 ) O(q_*^{-3/2}) O ( q ∗ − 3/2 ) cannot resolve a term of order q ∗ − 2 q_*^{-2} q ∗ − 2 . By (58) the combined kernel decays as q − 4 q^{-4} q − 4 , so its first possible kernel pole is at s = − 3 s=-3 s = − 3 . A nonzero residue there would give a sum of order q ∗ − 3 ∝ r s − 3 / 4 q_*^{-3}\propto r_s^{-3/4} q ∗ − 3 ∝ r s − 3/4 , whereas the proved bound is O ( r s − 3 / 8 ) O(r_s^{-3/8}) O ( r s − 3/8 ) . The self-energy pair thus cancels the r s log 2 r s r_s\log^2r_s r s log 2 r s term and the r s − 1 / 4 r_s^{-1/4} r s − 1/4 term of every row of Table 3 , but not the constant − e 2 x -e_{2x} − e 2 x of e c l o s e d e_{\mathrm{closed}} e closed .
In two dimensions, the same cancellation operates at the resonance of Sec. 3.1. By property (iii) of Sec. 3.1, the vertex part alone has the kernel endpoint q log ( 1 / q ) q\log(1/q) q log ( 1/ q ) , and the coincident pole at s = 2 s=2 s = 2 becomes triple. The vertex correction inserted alone into the ring chain, the two-dimensional counterpart of the chain family of the kite, would contribute
e v t x 2 D ⊃ 2 3 π 2 ( 10 − 3 π ) r s 2 log 2 r s = 0.03885 r s 2 log 2 r s R y . (61) e_{\mathrm{vtx}}^{2\mathrm{D}}\supset\frac{2}{3\pi^2}(10-3\pi)r_s^2\log^2r_s=0.03885\,r_s^2\log^2r_s\ \mathrm{Ry}. \tag*{(61)} e vtx 2 D ⊃ 3 π 2 2 ( 10 − 3 π ) r s 2 log 2 r s = 0.03885 r s 2 log 2 r s Ry . (61)
The coefficient follows from (32) with R d R_d R d replaced by π − 1 [ h d + z h d ′ ] log ( 1 / q ) \pi^{-1}[h_d+zh_d']\log(1/q) π − 1 [ h d + z h d ′ ] log ( 1/ q ) , which turns the double pole of (34) into a triple pole. The derivation also uses the identity ∫ h d N ∗ ( h d + z h d ′ ) d z = [ N ∗ / ( N ∗ + 1 ) ] ∫ h d N ∗ + 1 d z \int h_d^{N_*}(h_d+zh_d')\,\mathrm{d}z=[N_*/(N_*+1)]\int h_d^{N_*+1}\,\mathrm{d}z ∫ h d N ∗ ( h d + z h d ′ ) d z = [ N ∗ / ( N ∗ + 1 )] ∫ h d N ∗ + 1 d z , obtained by integrating z h d N ∗ h d ′ zh_d^{N_*}h_d' z h d N ∗ h d ′ by parts (SI Sec. S4). Since N ∗ + 1 = ( d + 1 ) / ( d − 1 ) = N r N_*+1=(d+1)/(d-1)=N_r N ∗ + 1 = ( d + 1 ) / ( d − 1 ) = N r , the integral on the right is that of the log r s \log r_s log r s term (28) of the rings at the same dimension. In three dimensions the same steps give 3 c h d r s log 2 r s 3c_{\mathrm{hd}}r_s\log^2r_s 3 c hd r s log 2 r s , the chain-family coefficient of (50) . In both dimensions, therefore, the log 2 r s \log^2r_s log 2 r s term of the isolated vertex correction is fixed by the integral that gives the log r s \log r_s log r s term of the rings, ∫ h 2 d z \int h^2\,\mathrm{d}z ∫ h 2 d z in three dimensions and ∫ h 2 2 d z \int h_2^2\,\mathrm{d}z ∫ h 2 2 d z in two. The self-energy pair removes this log 2 r s \log^2r_s log 2 r s term, and the conserving sum leaves the single log r s \log r_s log r s of (39) .
Conclusions
For the ring series and for diagrams linear in one screened interaction line in d d d dimensions, the powers of r s r_s r s and of log r s \log r_s log r s arise from residues of a Mellin convolution (table 4). Screening places its poles at the multiples of d − 1 d-1 d − 1 . The phase space places the kernel pole at s = d + 1 s=d+1 s = d + 1 for the rings and at s = 2 s=2 s = 2 for a polarization insertion. Once the Fermi-surface factors log ( 1 / q ) \log(1/q) log ( 1/ q ) of the vertex and self-energy corrections cancel, as the Ward identity requires, a log r s \log r_s log r s term appears where two poles coincide. This rule reproduces the known log r s \log r_s log r s terms in two and three dimensions and the ring power law for d > 3 d>3 d > 3 .
In two dimensions the insertion pole at s = 2 s=2 s = 2 meets the second screening pole, and the residue gives
e c 2 D ( r s ) = a 0 + a ℓ r s log r s + a 1 r s + a 2 ℓ r s 2 log r s + O ( r s 2 ) , a 2 ℓ = 7 π + 4 log 2 π − 160 9 π 2 − 16 G π 2 , (62) e_{\mathrm{c}}^{2\mathrm{D}}(r_s)=a_0+a_\ell r_s\log r_s+a_1r_s+a_{2\ell}r_s^2\log r_s+O(r_s^2),\qquad a_{2\ell}=\frac{7}{\pi}+\frac{4\log2}{\pi}-\frac{160}{9\pi^2}-\frac{16G}{\pi^2}, \tag*{(62)} e c 2 D ( r s ) = a 0 + a ℓ r s log r s + a 1 r s + a 2 ℓ r s 2 log r s + O ( r s 2 ) , a 2 ℓ = π 7 + π 4 log 2 − 9 π 2 160 − π 2 16 G , (62)
Table 4 : Asymptotic terms from the endpoint–residue correspondence (§ 2.7 , SI Proposition S3), in Ry per electron. The columns “Kernel” and “g g g ” give the endpoint terms whose poles produce the term, with t = q / ϵ t=q/\epsilon t = q / ϵ . SE denotes the Fock self-energy corrections. In the Kernel column, “(Ward)” marks the kernel of the sum of vertex and self-energy parts, in which their Fermi-surface factors log ( 1 / q ) \log(1/q) log ( 1/ q ) cancel (§ 2.7 ). The entry “opposite” marks the equal and opposite endpoint terms of the kite and SE (§ 3.4 ). The entry “in rings” denotes one insertion into the ring chain. The entries of the g g g column are shown without the factors h d ( z ) N h_d(z)^N h d ( z ) N of § 2.6 (for the kite, − h -h − h at full coupling and − 2 h / 3 -2h/3 − 2 h /3 for the coupling average, table 1). The last column gives the reference for a known coefficient, notes a proved remainder bound, or refers to the completeness statement under hypotheses (H1) and (H2) of Sec. 3.1 (SI Proposition S1). The two kite coefficients agree with those of the companion paper [50].
Table 4.
a 2 ℓ = − 0.17546 a_{2\ell}=-0.17546 a 2 ℓ = − 0.17546 Ry, with G G G Catalan’s constant. This value is the coefficient of the series with one insertion of the first-order exchange polarization Π ( 1 ) \Pi^{(1)} Π ( 1 ) into the ring chain. By power counting with the loop cancellation in symmetrized fermion loops, diagrams with two or more soft loops start at r s 4 r_s^4 r s 4 . Under hypotheses (H1) and (H2) of Sec. 3.1, a 2 ℓ a_{2\ell} a 2 ℓ is the complete coefficient. It constrains parametrizations of the two-dimensional correlation energy fitted to quantum Monte Carlo data, such as that of Ref. [88].
In three dimensions the screened kite alone acquires r s log 2 r s r_s\log^2r_s r s log 2 r s and r s − 1 / 4 r_s^{-1/4} r s − 1/4 terms. The Fock self-energy corrections remove both terms, and the remainders of the sum are bounded. Screened-exchange corrections to RPA-type correlation energies should therefore be evaluated together with the self-energy corrections. This conclusion holds at least for the leading terms of energy expressions with one screened line, without the Legendre-transform terms of § 3.4 or self-consistency.
Open problems are a proof of hypotheses (H1) and (H2) of Sec. 3.1, the constant a 1 a_1 a 1 of the two-dimensional gas, the next log r s \log r_s log r s terms in three dimensions, which belong to the complete second-order polarization in the ring chain (Sec. 3.1), and a proof of the f f f -sum-rule cancellation of SI Lemma S1 beyond first order.
Acknowledgments
This work received no external funding. The author used Claude (Anthropic) and ChatGPT (OpenAI) for assistance with analytic manipulations and numerical code development, reviewed the resulting material, and takes responsibility for the content.
Data availability statement
The Supporting Information provides the derivations and numerical data for the reported checks, including the d d d -dimensional coefficients and the closed forms of R 2 R_2 R 2 and a 2 ℓ a_{2\ell} a 2 ℓ . The Python scripts and raw output are provided as a supplementary code archive. The code for the direct numerical integration of the kite used in § 3.2 and § 3.3 and the scripts for most checks of SI Sec. S19 are available in a public repository [74]. Other data are available from the author upon reasonable request.
Conflict of interest
The author has no conflicts to disclose.
The Supporting Information (SI) contains the d d d -dimensional derivations behind § 2.8 , § 2.9 and 3.1: the long-wavelength Lindhard shape h d h_d h d , the ring coefficients, the long-wavelength limit R d R_d R d of the first-order exchange polarization, the closed form of R 2 R_2 R 2 , the coefficient a 2 ℓ a_{2\ell} a 2 ℓ , the finite-q q q checks, the order counting behind the completeness statement of Sec. 3.1 (SI Proposition S1) and the spin dependence. It also contains the normalization of the kite functional and the derivations behind § 2.1 –§ 2.7 and § 3.2 –§ 3.4 , the statement and proof of the endpoint–residue correspondence (SI Proposition S3), the proof of the cancellation of the universal low-density residue (SI Lemma S1) at first order and the argument for higher orders, the proof of SI Proposition S2, and the numerical checks. Its sections are cited as SI Sec. S S S n, and its equations, figures and tables carry the prefix S.
Supplementary Material
This document contains the normalization, the proofs of the remainder bounds and the numerical details behind the main article. Table S1 lists the sections that support each result of the main text. Equation, section, figure and table numbers without the prefix S refer to the main article, and numbers with the prefix S refer to this document.
Table S1. Where each result of the main article is supported.
Notation
The dimension of space is d d d . Sections S1–S8 treat general d d d , with the notation of main-text Secs. 2.2 and 2.8, 2.9 and 3.1: h d h_d h d is the long-wavelength Lindhard shape [h ≡ h 3 h \equiv h_3 h ≡ h 3 ], g d g_d g d its q 2 q^2 q 2 correction, R d R_d R d the long-wavelength first-order exchange polarization with R 3 = R ( 1 ) / ( 4 π ) R_3 = R^{(1)}/(4\pi) R 3 = R ( 1 ) / ( 4 π ) , A d A_d A d , B d B_d B d , C d C_d C d and α d \alpha_d α d the constants of table S3, N ∗ = 2 / ( d − 1 ) N_* = 2/(d - 1) N ∗ = 2/ ( d − 1 ) and N r = ( d + 1 ) / ( d − 1 ) N_r = (d + 1)/(d - 1) N r = ( d + 1 ) / ( d − 1 ) the indices of the resonant screening poles for a polarization insertion and for the rings, respectively, A d A_d A d the ring coefficient for d > 3 d > 3 d > 3 and a 2 ℓ a_{2\ell} a 2 ℓ the coefficient of r s 2 log r s r_s^2\log r_s r s 2 log r s in two dimensions. The proofs of Sec. S14.2 also hold in d d d dimensions. All other sections treat d = 3 d = 3 d = 3 . As in the main text, A ⊃ B A \supset B A ⊃ B means that the density expansion of A A A contains the term B B B . For d = 3 d = 3 d = 3 the conventions are those of main-text Sec. 2 (rederived from dimensional quantities in Sec. S9): momenta in units of k F k_F k F , frequencies in units of ℏ k F 2 / m e \hbar k_F^2/m_e ℏ k F 2 / m e , energies per electron in Ry, the imaginary frequency i ν i\nu i ν with ν > 0 \nu> 0 ν > 0 , and
α L = [ 4 / ( 9 π ) ] 1 / 3 , q T F 2 = 4 α L r s / π , A 0 = 3 / ( 32 π 5 ) , v ( q ) = q − 2 . \alpha_{\mathrm{L}} = [4/(9\pi)]^{1/3},\qquad q_{\mathrm{TF}}^2 = 4\alpha_{\mathrm{L}r_s}/\pi,\qquad A_0 = 3/(32\pi^5),\qquad v(q) = q^{-2}. α L = [ 4/ ( 9 π ) ] 1/3 , q TF 2 = 4 α L r s / π , A 0 = 3/ ( 32 π 5 ) , v ( q ) = q − 2 .
The screened line at coupling λ \lambda λ and its coupling average are
W λ = λ v / ϵ λ , ϵ λ = 1 − λ v Π 0 , Π 0 = − q T F 2 L , W ‾ = ∫ 0 1 W λ d λ . (63) W_\lambda= \lambda v/\epsilon_\lambda,\qquad\epsilon_\lambda= 1-\lambda v\Pi_0,\qquad\Pi_0 = -q_{\mathrm{TF}}^2\mathcal{L},\qquad\overline{W} = \int_0^1 W_\lambda\,\mathrm{d}\lambda. \tag*{(63)} W λ = λ v / ϵ λ , ϵ λ = 1 − λ v Π 0 , Π 0 = − q TF 2 L , W = ∫ 0 1 W λ d λ . (63)
Main-text Sec. 3.2 defines the kite functional K [ f ] \mathcal{K}[f] K [ f ] and
e 2 x = K [ v ] , e k i t e = 2 K [ W ‾ ] , e c l o s e d = 2 K [ W ‾ λ = 1 ] − e 2 x . e_{2x}=\mathcal{K}[v],\qquad e_{\mathrm{kite}}=2\mathcal{K}[\overline{W}],\qquad e_{\mathrm{closed}}=2\mathcal{K}[\overline{W}_{\lambda=1}]-e_{2x}. e 2 x = K [ v ] , e kite = 2 K [ W ] , e closed = 2 K [ W λ = 1 ] − e 2 x .
In remainder estimates, C b C_b C b is a positive constant that may change from one estimate to the next and does not depend on the small or large parameter of that estimate (q q q , r s r_s r s , Λ \Lambda Λ or ε \varepsilon ε , as stated there). The elementary functions used throughout are
h ( z ) = 1 − z arctan ( 1 / z ) , a ‾ x ( z ) = h ( z ) + z h ′ ( z ) , h(z)=1-z\arctan(1/z),\qquad\overline{a}_x(z)=h(z)+zh'(z), h ( z ) = 1 − z arctan ( 1/ z ) , a x ( z ) = h ( z ) + z h ′ ( z ) ,
f L ( k ) = 1 2 + 1 − k 2 4 k log ∣ 1 + k 1 − k ∣ . (64) f_{\mathrm{L}}(k)=\frac{1}{2}+\frac{1-k^2}{4k}\log\left|\frac{1+k}{1-k}\right|. \tag*{(64)} f L ( k ) = 2 1 + 4 k 1 − k 2 log 1 − k 1 + k . (64)
Here z = ν / q > 0 z=\nu/q>0 z = ν / q > 0 , h ′ = d h / d z h'=dh/dz h ′ = d h / d z , a ‾ x = a x / ( 8 π 2 ) \overline{a}_x=a_x/(8\pi^2) a x = a x / ( 8 π 2 ) with a x a_x a x of main-text Eq. (47) , and the endpoint values of f L f_{\mathrm{L}} f L are defined by real-axis limits. The two uses of f L f_{\mathrm{L}} f L are
Σ x ( k ) = − q T F 2 f L ( k ) / 2 (Fock self-energy) , L ( q , 0 ) = f L ( q / 2 ) (static Lindhard response) . \Sigma_x(k)=-q_{\mathrm{TF}}^2f_{\mathrm{L}}(k)/2\quad\text{(Fock self-energy)},\qquad\mathcal{L}(q,0)=f_{\mathrm{L}}(q/2)\quad\text{(static Lindhard response)}. Σ x ( k ) = − q TF 2 f L ( k ) /2 (Fock self-energy) , L ( q , 0 ) = f L ( q /2 ) (static Lindhard response) .
The first is derived after Eq. (81) . The second is the static Lindhard function of Ref. [1 ] , and for 0 < q < 1 0<q<1 0 < q < 1 it also follows from Eq. (149) at z = 0 z=0 z = 0 . Abbreviations: RPA (random-phase approximation), RPAx (RPA with exchange), BRM (Benites, Rosado and Manousakis [2 ] ), SOSEX (second-order screened exchange), CHSS (Czachor, Holas, Sharma and Singwi [41 ] ), TF (Thomas–Fermi) and G W Γ GW\Gamma G W Γ (the G W GW G W approximation with a vertex function Γ \Gamma Γ ). Table S2 lists the main symbols and terms with the place where each is defined.
Table S2. Main symbols and terms, with the place where each is defined in the main article (M) or in this document.
Conventions in d d d dimensions and the long-wavelength Lindhard shape
This section and the next seven support main-text Secs. 2.2 and 2.8, 2.9 and 3.1. They use the Fermi units of main-text Eq. (4) in d d d dimensions, with v ( q ) = q 1 − d v(q)=q^{1-d} v ( q ) = q 1 − d , and convert energies per electron to Ry with E 0 = 2 / ( α d r s ) 2 E_0=2/(\alpha_dr_s)^2 E 0 = 2/ ( α d r s ) 2 Ry.
Constants. Table S3 collects the constants of main-text Eqs. (3) and (7). With g p = ( i ν − δ p ) − 1 g_p=(i\nu-\delta_p)^{-1} g p = ( i ν − δ p ) − 1 and the spin factor 2, the factor C C C of a closed fermion loop in response units and the free response are
Table S3. Constants of the d d d -dimensional gas for d = 2 d=2 d = 2 and 3.
C = v 0 k F d / E 0 = C d / ( k F a B ) = C d α d r s , Π 0 = 2 C ∫ d d p ( 2 π ) − d Δ p Re g p , C=v_0k_{\mathrm{F}}^d/E_0=C_d/(k_{\mathrm{F}}a_B)=C_d\alpha_dr_s,\qquad\Pi_0=2C\int d^dp\,(2\pi)^{-d}\Delta_p\operatorname{Re}g_p, C = v 0 k F d / E 0 = C d / ( k F a B ) = C d α d r s , Π 0 = 2 C ∫ d d p ( 2 π ) − d Δ p Re g p ,
with C = π 2 q T F 2 C=\pi^2q_{\mathrm{TF}}^2 C = π 2 q TF 2 for d = 3 d=3 d = 3 [Eq. (81) ]. The value of Π 0 \Pi_0 Π 0 at q → 0 q\to0 q → 0 , ν = 0 \nu=0 ν = 0 defines q T F q_{\mathrm{TF}} q TF ,
Π 0 → − 2 C A d / ( 2 π ) d , q T F d − 1 = 2 C A d / ( 2 π ) d = B d α d r s . \Pi_0\to-2CA_d/(2\pi)^d,\qquad q_{\mathrm{TF}}^{d-1}=2CA_d/(2\pi)^d=B_d\alpha_dr_s. Π 0 → − 2 C A d / ( 2 π ) d , q TF d − 1 = 2 C A d / ( 2 π ) d = B d α d r s .
Long-wavelength limit. For q → 0 q\to0 q → 0 at fixed z = ν / q z=\nu/q z = ν / q , with μ = p ^ ⋅ q ^ \mu=\hat{p}\cdot\hat{q} μ = p ^ ⋅ q ^ ,
Δ p → δ ( 1 − ∣ p ∣ ) q μ , δ p → q μ , g p → b / q , b = ( i z − μ ) − 1 , Re b = − μ μ 2 + z 2 , \Delta_p\to\delta(1-|p|)\,q\mu,\qquad\delta_p\to q\mu,\qquad g_p\to b/q,\qquad b=(iz-\mu)^{-1},\qquad\operatorname{Re}b=-\frac{\mu}{\mu^2+z^2}, Δ p → δ ( 1 − ∣ p ∣ ) q μ , δ p → q μ , g p → b / q , b = ( i z − μ ) − 1 , Re b = − μ 2 + z 2 μ ,
and therefore
Π 0 → 2 C A d ( 2 π ) − d ⟨ μ Re b ⟩ = − q T F d − 1 ⟨ μ 2 / ( μ 2 + z 2 ) ⟩ , \Pi_0\to2CA_d(2\pi)^{-d}\langle\mu\operatorname{Re}b\rangle=-q_{\mathrm{TF}}^{d-1}\langle\mu^2/(\mu^2+z^2)\rangle, Π 0 → 2 C A d ( 2 π ) − d ⟨ μ Re b ⟩ = − q TF d − 1 ⟨ μ 2 / ( μ 2 + z 2 )⟩ ,
which is main-text Eq. (9) . The average over S d − 1 S^{d-1} S d − 1 has the weight ( 1 − μ 2 ) ( d − 3 ) / 2 / B ( 1 / 2 , ( d − 1 ) / 2 ) (1-\mu^2)^{(d-3)/2}/B(1/2,(d-1)/2) ( 1 − μ 2 ) ( d − 3 ) /2 / B ( 1/2 , ( d − 1 ) /2 ) on − 1 ≤ μ ≤ 1 -1\le\mu\le1 − 1 ≤ μ ≤ 1 , and Euler’s integral gives
⟨ ( μ 2 + z 2 ) − 1 ⟩ = z − 2 2 F 1 ( 1 , 1 2 ; d 2 ; − z − 2 ) , \left\langle(\mu^2+z^2)^{-1}\right\rangle=z^{-2}\,{}_2F_1\left(1,\frac{1}{2};\frac{d}{2};-z^{-2}\right), ⟨ ( μ 2 + z 2 ) − 1 ⟩ = z − 2 2 F 1 ( 1 , 2 1 ; 2 d ; − z − 2 ) ,
hence
h d = 1 − z 2 ⟨ ( μ 2 + z 2 ) − 1 ⟩ = 1 − 2 F 1 ( 1 , 1 2 ; d 2 ; − z − 2 ) . h_d=1-z^2\left\langle(\mu^2+z^2)^{-1}\right\rangle=1-{}_2F_1\left(1,\frac{1}{2};\frac{d}{2};-z^{-2}\right). h d = 1 − z 2 ⟨ ( μ 2 + z 2 ) − 1 ⟩ = 1 − 2 F 1 ( 1 , 2 1 ; 2 d ; − z − 2 ) .
For d = 2 d=2 d = 2 , ⟨ ( μ 2 + z 2 ) − 1 ⟩ = 1 / ( z 1 + z 2 ) \langle(\mu^2+z^2)^{-1}\rangle=1/(z\sqrt{1+z^2}) ⟨( μ 2 + z 2 ) − 1 ⟩ = 1/ ( z 1 + z 2 ) , and for d = 4 d=4 d = 4 the hypergeometric function is elementary. These cases give main-text Eq. (10) , and ⟨ μ 2 ⟩ = 1 / d \langle\mu^2\rangle=1/d ⟨ μ 2 ⟩ = 1/ d gives h d ≃ ⟨ μ 2 ⟩ z − 2 = 1 / ( d z 2 ) h_d\simeq\langle\mu^2\rangle z^{-2}=1/(dz^2) h d ≃ ⟨ μ 2 ⟩ z − 2 = 1/ ( d z 2 ) at large z z z . Direct quadrature of the average for d = 2 , 3 , 3.5 , 4 , 5 d=2,3,3.5,4,5 d = 2 , 3 , 3.5 , 4 , 5 agrees with the hypergeometric form to 18 digits.
The variable r r r . For d = 2 d=2 d = 2 and d = 4 d=4 d = 4 the substitution r = 1 + z 2 − z r=\sqrt{1+z^2}-z r = 1 + z 2 − z , which maps 0 < z < ∞ 0<z<\infty 0 < z < ∞ onto 1 > r > 0 1>r>0 1 > r > 0 , gives h 4 = r 2 h_4=r^2 h 4 = r 2 and
z = 1 − r 2 2 r , 1 + z 2 = 1 + r 2 2 r , a ≡ z 1 + z 2 = 1 − r 2 1 + r 2 , z=\frac{1-r^2}{2r},\qquad\sqrt{1+z^2}=\frac{1+r^2}{2r},\qquad a\equiv\frac{z}{\sqrt{1+z^2}}=\frac{1-r^2}{1+r^2}, z = 2 r 1 − r 2 , 1 + z 2 = 2 r 1 + r 2 , a ≡ 1 + z 2 z = 1 + r 2 1 − r 2 ,
{#eq:S2}
h_2=\frac{2r^2}{1+r^2},\qquad |dz|=\frac{1+r^2}{2r^2},dr.
h_2^3|dz|=\frac{4r^4,dr}{(1+r^2)^2},\qquad h_4^{5/3}|dz|=\frac{1}{2}(r^{4/3}+r^{10/3}),dr,
and since h d h_d h d is even, the ring coefficients contain the integrals
∫ − ∞ ∞ h 2 3 d z = 8 ∫ 0 1 r 4 d r ( 1 + r 2 ) 2 = 10 − 3 π , ∫ − ∞ ∞ h 4 5 / 3 d z = ∫ 0 1 ( r 4 / 3 + r 10 / 3 ) d r = 60 91 , (65) \int_{-\infty}^{\infty}h_2^3\,dz=8\int_0^1\frac{r^4\,dr}{(1+r^2)^2}=10-3\pi,\qquad\int_{-\infty}^{\infty}h_4^{5/3}\,dz=\int_0^1(r^{4/3}+r^{10/3})\,dr=\frac{60}{91}, \tag*{(65)} ∫ − ∞ ∞ h 2 3 d z = 8 ∫ 0 1 ( 1 + r 2 ) 2 r 4 d r = 10 − 3 π , ∫ − ∞ ∞ h 4 5/3 d z = ∫ 0 1 ( r 4/3 + r 10/3 ) d r = 91 60 , (65)
and in three dimensions ∫ − ∞ ∞ h 2 d z = 2 π 3 ( 1 − log 2 ) \int_{-\infty}^{\infty}h^2\,dz=\frac{2\pi}{3}(1-\log2) ∫ − ∞ ∞ h 2 d z = 3 2 π ( 1 − log 2 ) [Eq. (141) ].
Small-q q q expansion. The Lindhard function L ( q , ν ) \mathcal{L}(q,\nu) L ( q , ν ) is invariant under ( q , ν ) → ( − q , − ν ) (q,\nu)\to(-q,-\nu) ( q , ν ) → ( − q , − ν ) and even in ν \nu ν . Hence L ( q , q z ) \mathcal{L}(q,qz) L ( q , q z ) is even in q q q at fixed z z z , in every dimension, and, with g 3 g_3 g 3 from Sec. S14.6,
L ( q , q z ) = h d ( z ) + q 2 g d ( z ) + O ( q 4 ) , g 3 ( z ) = − 1 12 ( 1 + z 2 ) 2 . \mathcal{L}(q,qz)=h_d(z)+q^2g_d(z)+O(q^4),\qquad g_3(z)=-\frac{1}{12(1+z^2)^2}. L ( q , q z ) = h d ( z ) + q 2 g d ( z ) + O ( q 4 ) , g 3 ( z ) = − 12 ( 1 + z 2 ) 2 1 .
In two dimensions the angular integral of the free response is elementary. At imaginary frequency
L ( q , q z ) = 1 − 2 q Re ( q / 2 + i z ) 2 − 1 , \mathcal{L}(q,qz)=1-\frac{2}{q}\operatorname{Re}\sqrt{(q/2+iz)^2-1}, L ( q , q z ) = 1 − q 2 Re ( q /2 + i z ) 2 − 1 ,
with the branch that tends to i 1 + z 2 i\sqrt{1+z^2} i 1 + z 2 as q → 0 q\to0 q → 0 . For z ≥ 0 z\ge0 z ≥ 0 the square root expands as
Re ( q / 2 + i z ) 2 − 1 = q z 2 1 + z 2 + q 3 z 16 ( 1 + z 2 ) 5 / 2 + O ( q 5 ) . \operatorname{Re}\sqrt{(q/2+iz)^2-1}=\frac{qz}{2\sqrt{1+z^2}}+\frac{q^3z}{16(1+z^2)^{5/2}}+O(q^5). Re ( q /2 + i z ) 2 − 1 = 2 1 + z 2 q z + 16 ( 1 + z 2 ) 5/2 q 3 z + O ( q 5 ) .
This gives h 2 ( z ) h_2(z) h 2 ( z ) , a vanishing coefficient of q q q , and
g 2 ( z ) = − z 8 ( 1 + z 2 ) 5 / 2 (66) g_2(z)=-\frac{z}{8(1+z^2)^{5/2}} \tag*{(66)} g 2 ( z ) = − 8 ( 1 + z 2 ) 5/2 z (66)
again with a vanishing coefficient of q 3 q^3 q 3 . A 40-digit evaluation of the exact free response confirms (S4) and the vanishing coefficient of q q q at z = 0.3 z=0.3 z = 0.3 , 1 1 1 and 3 3 3 , where
g 2 ( 0.3 ) = − 0.0302318708 , g 2 ( 1 ) = − 1 / ( 32 2 ) , g 2 ( 3 ) = − 0.0011858541. g_2(0.3)=-0.0302318708,\qquad g_2(1)=-1/(32\sqrt{2}),\qquad g_2(3)=-0.0011858541. g 2 ( 0.3 ) = − 0.0302318708 , g 2 ( 1 ) = − 1/ ( 32 2 ) , g 2 ( 3 ) = − 0.0011858541.
For z = 0 z=0 z = 0 , g 2 = 0 g_2=0 g 2 = 0 , as required by the flat static response of the two-dimensional gas below 2 k F 2k_F 2 k F [40 ] .
Ring series in d d d dimensions
Energy. In units of E 0 k F d E_0 k_F^d E 0 k F d the RPA correlation energy per unit volume is
1 2 ∫ d d q ( 2 π ) − d ∫ − ∞ ∞ d ν ( 2 π ) − 1 [ log ( 1 + c ) − c ] . \frac{1}{2}\int d^d q\,(2\pi)^{-d}\int_{-\infty}^{\infty}d\nu\,(2\pi)^{-1}[\log(1+c)-c]. 2 1 ∫ d d q ( 2 π ) − d ∫ − ∞ ∞ d ν ( 2 π ) − 1 [ log ( 1 + c ) − c ] .
Multiplying by the volume ( 2 π ) d / ( 2 V d ) (2\pi)^d/(2V_d) ( 2 π ) d / ( 2 V d ) per electron, with d d q = A d q d − 1 d q d^d q=A_dq^{d-1}dq d d q = A d q d − 1 d q , d ν = q d z d\nu=q\,dz d ν = q d z and A d = d V d A_d=dV_d A d = d V d , gives
e R P A = E 0 A d 8 π V d ∫ − ∞ ∞ d z ∫ 0 ∞ d q q d [ log ( 1 + c ) − c ] , E 0 A d 8 π V d = d E 0 8 π , e_{\mathrm{RPA}}=\frac{E_0A_d}{8\pi V_d}\int_{-\infty}^{\infty}dz\int_0^\infty dq\,q^d[\log(1+c)-c],\qquad\frac{E_0A_d}{8\pi V_d}=\frac{dE_0}{8\pi}, e RPA = 8 π V d E 0 A d ∫ − ∞ ∞ d z ∫ 0 ∞ d q q d [ log ( 1 + c ) − c ] , 8 π V d E 0 A d = 8 π d E 0 ,
d E 0 8 π ∣ d = 3 = 3 8 π ⋅ 32 π 2 q T F 4 = 12 π 3 q T F 4 R y . \left.\frac{dE_0}{8\pi}\right|_{d=3}=\frac{3}{8\pi}\cdot\frac{32}{\pi^2q_{\mathrm{TF}}^4}=\frac{12}{\pi^3q_{\mathrm{TF}}^4}\ \mathrm{Ry}. 8 π d E 0 d = 3 = 8 π 3 ⋅ π 2 q TF 4 32 = π 3 q TF 4 12 Ry .
The prefactor d E 0 / ( 8 π ) dE_0/(8\pi) d E 0 / ( 8 π ) is that of main-text Eq. (26) , and its d = 3 d=3 d = 3 value is Eq. (20) of Ref. [29 ] .
Resonant dimensions. With c = h d t 1 − d c=h_dt^{1-d} c = h d t 1 − d and t = q / q T F t=q/q_{\mathrm{TF}} t = q / q TF , the q q q integral at fixed z z z is
q T F d + 1 ∫ 0 ∞ t d [ log ( 1 + h d t 1 − d ) − h d t 1 − d ] d t , q_{\mathrm{TF}}^{d+1}\int_0^\infty t^d[\log(1+h_dt^{1-d})-h_dt^{1-d}]\,dt, q TF d + 1 ∫ 0 ∞ t d [ log ( 1 + h d t 1 − d ) − h d t 1 − d ] d t ,
cut off at t ∼ 1 / q T F t\sim1/q_{\mathrm{TF}} t ∼ 1/ q TF by the Lindhard function. In the expansion of log ( 1 + h d t 1 − d ) \log(1+h_dt^{1-d}) log ( 1 + h d t 1 − d ) , the term of order N r N_r N r , with N r ( d − 1 ) = d + 1 N_r(d-1)=d+1 N r ( d − 1 ) = d + 1 , makes the integrand proportional to t − 1 t^{-1} t − 1 , and the window 1 ≲ t ≲ 1 / q T F 1\lesssim t\lesssim1/q_{\mathrm{TF}} 1 ≲ t ≲ 1/ q TF gives
t d ( − 1 ) N r + 1 h d N r N r t − N r ( d − 1 ) = ( − 1 ) N r + 1 h d N r N r t , ∫ 1 1 / q T F d t t = − log q T F , t^d\frac{(-1)^{N_r+1}h_d^{N_r}}{N_r}t^{-N_r(d-1)}=\frac{(-1)^{N_r+1}h_d^{N_r}}{N_rt},\qquad\int_1^{1/q_{\mathrm{TF}}}\frac{dt}{t}=-\log q_{\mathrm{TF}}, t d N r ( − 1 ) N r + 1 h d N r t − N r ( d − 1 ) = N r t ( − 1 ) N r + 1 h d N r , ∫ 1 1/ q TF t d t = − log q TF ,
q T F d + 1 ∫ 0 ∞ t d [ ⋯ ] d t ⊃ ( − 1 ) N r ( h d N r N r ) q T F d + 1 log q T F , q_{\mathrm{TF}}^{d+1}\int_0^\infty t^d[\cdots]\,dt\supset(-1)^{N_r}\left(\frac{h_d^{N_r}}{N_r}\right)q_{\mathrm{TF}}^{d+1}\log q_{\mathrm{TF}}, q TF d + 1 ∫ 0 ∞ t d [ ⋯ ] d t ⊃ ( − 1 ) N r ( N r h d N r ) q TF d + 1 log q TF ,
in agreement with Eq. (140) for m = 0 m=0 m = 0 . With log q T F = log r s / ( d − 1 ) + O ( 1 ) \log q_{\mathrm{TF}}=\log r_s/(d-1)+O(1) log q TF = log r s / ( d − 1 ) + O ( 1 ) this is main-text Eq. (28) . The regions q ≲ q T F q\lesssim q_{\mathrm{TF}} q ≲ q TF and q ≳ 1 q\gtrsim1 q ≳ 1 are controlled as for d = 3 d=3 d = 3 in § 25.6 , with the two-dimensional Lindhard function of § 12 for d = 2 d=2 d = 2 . With Table S3 and Eq. (65) the coefficients of log r s \log r_s log r s are
d = 3 : 12 π 3 ⋅ 1 4 ∫ h 2 2 d z = 2 π 2 ( 1 − log 2 ) , d = 2 : − 2 2 r s π ⋅ 1 3 ∫ h 2 3 d z = − 2 2 ( 10 3 π − 1 ) r s , d=3:\quad\frac{12}{\pi^3}\cdot\frac{1}{4}\int h_2^2\,dz=\frac{2}{\pi^2}(1-\log2),\qquad d=2:\quad-\frac{2\sqrt{2}r_s}{\pi}\cdot\frac{1}{3}\int h_2^3\,dz=-2\sqrt{2}\left(\frac{10}{3\pi}-1\right)r_s, d = 3 : π 3 12 ⋅ 4 1 ∫ h 2 2 d z = π 2 2 ( 1 − log 2 ) , d = 2 : − π 2 2 r s ⋅ 3 1 ∫ h 2 3 d z = − 2 2 ( 3 π 10 − 1 ) r s ,
which are main-text Eqs. (29) and (30) .
Dimensions d > 3 d>3 d > 3 . The scale-free integral then converges, and the substitution u = h d t 1 − d u=h_dt^{1-d} u = h d t 1 − d gives
∫ 0 ∞ t d [ log ( 1 + h d t 1 − d ) − h d t 1 − d ] d t = h d N r d − 1 ∫ 0 ∞ u − N r − 1 [ log ( 1 + u ) − u ] d u = h d N r d − 1 π N r sin ( π N r ) . (67) \int_0^\infty t^d[\log(1+h_dt^{1-d})-h_dt^{1-d}]\,dt
=\frac{h_d^{N_r}}{d-1}\int_0^\infty u^{-N_r-1}[\log(1+u)-u]\,du
=\frac{h_d^{N_r}}{d-1}\frac{\pi}{N_r\sin(\pi N_r)}. \tag*{(67)} ∫ 0 ∞ t d [ log ( 1 + h d t 1 − d ) − h d t 1 − d ] d t = d − 1 h d N r ∫ 0 ∞ u − N r − 1 [ log ( 1 + u ) − u ] d u = d − 1 h d N r N r sin ( π N r ) π . (67)
for 1 < N r < 2 1<N_{\mathrm{r}}<2 1 < N r < 2 . The last integral follows by one integration by parts, whose boundary terms vanish for 1 < N r < 2 1<N_{\mathrm{r}}<2 1 < N r < 2 , from
∫ 0 ∞ u a − 1 ( 1 + u ) − 1 d u = π sin ( π a ) \int_{0}^{\infty}u^{a-1}(1+u)^{-1}\mathrm{d}u=\frac{\pi}{\sin(\pi a)} ∫ 0 ∞ u a − 1 ( 1 + u ) − 1 d u = sin ( π a ) π
with a = 2 − N r a=2-N_{\mathrm{r}} a = 2 − N r ,
∫ 0 ∞ u − N r − 1 [ log ( 1 + u ) − u ] d u = − 1 N r ∫ 0 ∞ u 1 − N r 1 + u d u = − π N r sin ( π ( 2 − N r ) ) = π N r sin ( π N r ) . \int_{0}^{\infty}u^{-N_{\mathrm{r}}-1}\left[\log(1+u)-u\right]\mathrm{d}u=-\frac{1}{N_{\mathrm{r}}}\int_{0}^{\infty}\frac{u^{1-N_{\mathrm{r}}}}{1+u}\mathrm{d}u=-\frac{\pi}{N_{\mathrm{r}}\sin\left(\pi(2-N_{\mathrm{r}})\right)}=\frac{\pi}{N_{\mathrm{r}}\sin(\pi N_{\mathrm{r}})}. ∫ 0 ∞ u − N r − 1 [ log ( 1 + u ) − u ] d u = − N r 1 ∫ 0 ∞ 1 + u u 1 − N r d u = − N r sin ( π ( 2 − N r ) ) π = N r sin ( π N r ) π .
With ( d − 1 ) N r = d + 1 (d-1)N_{\mathrm{r}}=d+1 ( d − 1 ) N r = d + 1 and E 0 = 2 / ( α d r s 2 ) E_{0}=2/(\alpha_{d}r_{\mathrm{s}}^{2}) E 0 = 2/ ( α d r s 2 ) Ry,
E 0 q T F d + 1 = 2 ( B d α d r s ) N r ( α d r s ) 2 = 2 α d N r − 2 B d N r r s N r − 2 R y , E_{0}q_{\mathrm{TF}}^{d+1}=\frac{2(B_{d}\alpha_{d}r_{\mathrm{s}})^{N_{\mathrm{r}}}}{(\alpha_{d}r_{\mathrm{s}})^{2}}=2\alpha_{d}^{N_{\mathrm{r}}-2}B_{d}^{N_{\mathrm{r}}}r_{\mathrm{s}}^{N_{\mathrm{r}}-2}\,\mathrm{Ry}, E 0 q TF d + 1 = ( α d r s ) 2 2 ( B d α d r s ) N r = 2 α d N r − 2 B d N r r s N r − 2 Ry ,
e R P A ≃ d 8 π E 0 q T F d + 1 d − 1 π N r sin ( π N r ) ∫ − ∞ ∞ h d N r d z = d α d N r − 2 B d N r 4 ( d + 1 ) sin ( π N r ) ∫ − ∞ ∞ h d N r d z r s N r − 2 R y , e_{\mathrm{RPA}}\simeq\frac{d}{8\pi}\frac{E_{0}q_{\mathrm{TF}}^{d+1}}{d-1}\frac{\pi}{N_{\mathrm{r}}\sin(\pi N_{\mathrm{r}})}\int_{-\infty}^{\infty}h_{d}^{N_{\mathrm{r}}}\,\mathrm{d}z=\frac{d\alpha_{d}^{N_{\mathrm{r}}-2}B_{d}^{N_{\mathrm{r}}}{}}{4(d+1)\sin(\pi N_{\mathrm{r}})}\int_{-\infty}^{\infty}h_{d}^{N_{\mathrm{r}}}\,\mathrm{d}z\,r_{\mathrm{s}}^{N_{\mathrm{r}}-2}\,\mathrm{Ry}, e RPA ≃ 8 π d d − 1 E 0 q TF d + 1 N r sin ( π N r ) π ∫ − ∞ ∞ h d N r d z = 4 ( d + 1 ) sin ( π N r ) d α d N r − 2 B d N r ∫ − ∞ ∞ h d N r d z r s N r − 2 Ry ,
which gives main-text Eq. (31) . For d = 4 d=4 d = 4 ,
N r = 5 / 3 , A 4 = 2 π 2 , C 4 = 4 π 2 , B 4 = 1 , α 4 = 2 − 5 / 4 , sin ( 5 π / 3 ) = − 3 / 2 , N_{\mathrm{r}}=5/3,\qquad A_{4}=2\pi^{2},\qquad C_{4}=4\pi^{2},\qquad B_{4}=1,\qquad\alpha_{4}=2^{-5/4},\qquad\sin(5\pi/3)=-\sqrt{3}/2, N r = 5/3 , A 4 = 2 π 2 , C 4 = 4 π 2 , B 4 = 1 , α 4 = 2 − 5/4 , sin ( 5 π /3 ) = − 3 /2 ,
and with (S3)
A 4 = 4 ⋅ 2 5 / 12 20 ⋅ ( − 3 / 2 ) ⋅ 60 91 R y = − 24 ⋅ 2 5 / 12 91 3 R y . A_{4}=\frac{4\cdot2^{5/12}}{20\cdot(-\sqrt{3}/2)}\cdot\frac{60}{91}\,\mathrm{Ry}=-\frac{24\cdot2^{5/12}}{91\sqrt{3}}\,\mathrm{Ry}. A 4 = 20 ⋅ ( − 3 /2 ) 4 ⋅ 2 5/12 ⋅ 91 60 Ry = − 91 3 24 ⋅ 2 5/12 Ry .
The limit d → 3 d\to3 d → 3 . Near d = 3 d=3 d = 3 ,
N r = 2 − d − 3 d − 1 , sin ( π N r ) = − π 2 ( d − 3 ) + O ( ( d − 3 ) 2 ) , N_{\mathrm{r}}=2-\frac{d-3}{d-1},\qquad\sin(\pi N_{\mathrm{r}})=-\frac{\pi}{2}(d-3)+O\left((d-3)^{2}\right), N r = 2 − d − 1 d − 3 , sin ( π N r ) = − 2 π ( d − 3 ) + O ( ( d − 3 ) 2 ) ,
and at d = 3 d=3 d = 3 the remaining factors of A d A_{d} A d give 3 B 3 2 h 2 d z / 16 = ( 2 / π ) ( 1 − log 2 ) 3B_{3}^{2}h^{2}\mathrm{d}z/16=(2/\pi)(1-\log2) 3 B 3 2 h 2 d z /16 = ( 2/ π ) ( 1 − log 2 ) . Hence
( d − 3 ) A d → ( 2 / π ) ( 1 − log 2 ) − π / 2 = − 4 ( 1 − log 2 ) π 2 = − 2 A . (d-3)A_{d}\to\frac{(2/\pi)(1-\log2)}{-\pi/2}=-\frac{4(1-\log2)}{\pi^{2}}=-2A. ( d − 3 ) A d → − π /2 ( 2/ π ) ( 1 − log 2 ) = − π 2 4 ( 1 − log 2 ) = − 2 A .
In Eq. (21) of Ref. [32 ] the explicit r s 0 r_{\mathrm{s}}^{0} r s 0 term D ∫ R ( D − 1 ) / 2 2 d u / [ π 3 ( D − 3 ) ] D\int R_{(D-1)/2}^{2}\mathrm{d}u/[\pi^{3}(D-3)] D ∫ R ( D − 1 ) /2 2 d u / [ π 3 ( D − 3 )] has the opposite pole, since R ( D − 1 ) / 2 R_{(D-1)/2} R ( D − 1 ) /2 reduces to h h h at D = 3 D=3 D = 3 . This term cancels the constant − 2 A / ( d − 3 ) -2A/(d-3) − 2 A / ( d − 3 ) (main-text § 2.8 ). Hence the remaining part δ D \delta_{D} δ D of their r s 0 r_{\mathrm{s}}^{0} r s 0 term must be regular at d = 3 d=3 d = 3 . Numerically
( d − 3 ) A d / 2 = − 0.0639390 , − 0.0623257 , − 0.0621955 ( d = 3.1 , 3.01 , 3.001 ) , (d-3)A_{d}/2=-0.0639390,\quad-0.0623257,\quad-0.0621955\qquad(d=3.1,\ 3.01,\ 3.001), ( d − 3 ) A d /2 = − 0.0639390 , − 0.0623257 , − 0.0621955 ( d = 3.1 , 3.01 , 3.001 ) ,
approaching − A = − 0.0621814 -A=-0.0621814 − A = − 0.0621814 linearly in d − 3 d-3 d − 3 .
The limit d → ∞ d\to\infty d → ∞ . For large d d d the component μ \mu μ is Gaussian with variance 1 / d 1/d 1/ d . Hence h d ( y / d ) h_{d}(y/\sqrt{d}) h d ( y / d ) tends to a fixed function of y y y , and with x x x a standard normal variable
∫ − ∞ ∞ h d ( y / d ) d y → ⟨ ∫ − ∞ ∞ x 2 ( x 2 + y 2 ) − 1 d y ⟩ x = π ⟨ ∣ x ∣ ⟩ = 2 π , \int_{-\infty}^{\infty}h_{d}(y/\sqrt{d})\,\mathrm{d}y\to\left\langle\int_{-\infty}^{\infty}x^{2}(x^{2}+y^{2})^{-1}\,\mathrm{d}y\right\rangle_{x}=\pi\langle|x|\rangle=\sqrt{2\pi}, ∫ − ∞ ∞ h d ( y / d ) d y → ⟨ ∫ − ∞ ∞ x 2 ( x 2 + y 2 ) − 1 d y ⟩ x = π ⟨ ∣ x ∣ ⟩ = 2 π ,
hence
∫ h d N r d z ≃ 2 π / d . \int h_{d}^{N_{\mathrm{r}}}\,\mathrm{d}z\simeq\sqrt{2\pi/d}. ∫ h d N r d z ≃ 2 π / d .
Stirling’s formula gives B d ≃ 2 2 / π d B_{d}\simeq2\sqrt{2}/\sqrt{\pi d} B d ≃ 2 2 / π d and α d − 1 ≃ d / e \alpha_{d}^{-1}\simeq d/e α d − 1 ≃ d / e , and sin ( π N r ) ≃ − 2 π / d \sin(\pi N_{\mathrm{r}})\simeq-2\pi/d sin ( π N r ) ≃ − 2 π / d . With these, (31) gives
A d ≃ d ( d / e ) B d 2 π / d 4 d ( − 2 π / d ) = − d 2 e π R y . A_{d}\simeq\frac{d(d/e)B_{d}\sqrt{2\pi/d}}{4d(-2\pi/d)}=-\frac{d}{2e\pi}\,\mathrm{Ry}. A d ≃ 4 d ( − 2 π / d ) d ( d / e ) B d 2 π / d = − 2 e π d Ry .
Numerically A d / d A_{d}/d A d / d is − 0.052756 -0.052756 − 0.052756 , − 0.057680 -0.057680 − 0.057680 and − 0.058216 -0.058216 − 0.058216 at d = 100 d=100 d = 100 , 1000 1000 1000 and 3000 3000 3000 , approaching − 1 / ( 2 e π ) = − 0.058550 -1/(2e\pi)=-0.058550 − 1/ ( 2 e π ) = − 0.058550 with corrections of order ( log d ) / d (\log d)/d ( log d ) / d .
Comparison with Ref. [32 ] . Equation (21) of Ref. [32 ] , with its series Σ D \Sigma_{D} Σ D , the functions R ( D − 1 ) / 2 R_{(D-1)/2} R ( D − 1 ) /2 of its Eq. (18) and α D = 1 / α d \alpha_{D}=1/\alpha_{d} α D = 1/ α d , gives the same coefficient as (31) to machine precision for every d d d tested (3.001 ≤ d ≤ 3000 3.001 \le d \le3000 3.001 ≤ d ≤ 3000 ). A closed form of the series Σ D \Sigma_D Σ D agrees with its printed definition to 40 digits for d = 4 , 5 , 6 , 9 d = 4,5,6,9 d = 4 , 5 , 6 , 9 . The values in Table I of Ref. [32 ] , in Hartree, are smaller in magnitude than A d A_d A d by the factors in main-text Table 2 . The large-d d d limit stated there, − D / ( 2 e π 2 ) -D/(2e\pi^2) − D / ( 2 e π 2 ) Hartree, is smaller in magnitude than − d / ( 4 e π ) -d/(4e\pi) − d / ( 4 e π ) Hartree by a factor of π / 2 \pi/2 π /2 . Both differences appear in the published and arXiv (v2) versions. Reference [33 ] states in its Ref. 68 that a small error in the numerical evaluation of that table was corrected, but gives no corrected values. Since main-text Eq. (31) reproduces Eq. (21) of Ref. [32 ] , the difference lies in the numerical evaluation, not in the formula.
First-order exchange polarization at long wavelength
The steps of Sec. S9.2 hold in d d d dimensions with d 3 p → d d p d^3p \to d^dp d 3 p → d d p , ( 2 π ) 3 → ( 2 π ) d (2\pi)^3 \to(2\pi)^d ( 2 π ) 3 → ( 2 π ) d and C = C d α d r s C = C_d\alpha_d r_s C = C d α d r s . The Fock shift of the excitation energy and the vertex share the factor Δ p Δ k C v ( p − k ) \Delta_p\Delta_k C v(\mathbf{p}-\mathbf{k}) Δ p Δ k C v ( p − k ) . The fixed-density counterterm (the shift δ μ x \delta\mu_x δ μ x of the chemical potential in Sec. S9.2) removes the occupation-surface terms , which come from the change of the occupation numbers at the Fermi surface. Symmetrization then gives
Π ( 1 ) ( q , i ν ) = C ∫ d d p d d k ( 2 π ) 2 d C v ( p − k ) Δ p Δ k Re [ ( g p − g k ) 2 ] , (68) \Pi^{(1)}(q,i\nu)=C\int\frac{\mathrm{d}^d p\,\mathrm{d}^d k}{(2\pi)^{2d}}\,C v(\mathbf{p}-\mathbf{k})\,\Delta_p\Delta_k\,\operatorname{Re}\left[(g_p-g_k)^2\right], \tag*{(68)} Π ( 1 ) ( q , i ν ) = C ∫ ( 2 π ) 2 d d d p d d k C v ( p − k ) Δ p Δ k Re [ ( g p − g k ) 2 ] , (68)
which for d = 3 d=3 d = 3 is Eq. (85) and for d = 2 d=2 d = 2 is the combined kernel of Eq. (11) of Ref. [36 ] continued to imaginary frequency.
Limit. In the long-wavelength limit of Sec. S1, on the Fermi surface,
Δ p Δ k → q 2 μ μ ′ , ( g p − g k ) 2 → ( b − b ′ ) 2 q 2 , v ( p − k ) → ∣ p ^ − p ^ ′ ∣ 1 − d , \Delta_p\Delta_k\to q^2\mu\mu',\qquad
(g_p-g_k)^2\to\frac{(b-b')^2}{q^2},\qquad
v(\mathbf{p}-\mathbf{k})\to\lvert\hat{\mathbf{p}}-\hat{\mathbf{p}}'\rvert^{1-d}, Δ p Δ k → q 2 μ μ ′ , ( g p − g k ) 2 → q 2 ( b − b ′ ) 2 , v ( p − k ) → ∣ p ^ − p ^ ′ ∣ 1 − d ,
and therefore
Π ( 1 ) → C 2 A d 2 ( 2 π ) 2 d ⟨ μ μ ′ Re [ ( b − b ′ ) 2 ] ∣ p ^ − p ^ ′ ∣ d − 1 ⟩ = − q T F d − 1 α d r s R d ( z ) , (69) \Pi^{(1)}\to C^2\frac{A_d^2}{(2\pi)^{2d}}\left\langle\frac{\mu\mu'\operatorname{Re}\left[(b-b')^2\right]}{\lvert\hat{\mathbf{p}}-\hat{\mathbf{p}}'\rvert^{d-1}}\right\rangle=-q_{\mathrm{TF}}^{d-1}\alpha_d r_s R_d(z), \tag*{(69)} Π ( 1 ) → C 2 ( 2 π ) 2 d A d 2 ⟨ ∣ p ^ − p ^ ′ ∣ d − 1 μ μ ′ Re [ ( b − b ′ ) 2 ] ⟩ = − q TF d − 1 α d r s R d ( z ) , (69)
with
q T F d − 1 α d r s = 2 C 2 A d ( 2 π ) d C d , C 2 A d 2 ( 2 π ) 2 d = B d 4 q T F d − 1 α d r s , q_{\mathrm{TF}}^{d-1}\alpha_d r_s=\frac{2C^2A_d}{(2\pi)^d C_d},\qquad
\frac{C^2A_d^2}{(2\pi)^{2d}}=\frac{B_d}{4}q_{\mathrm{TF}}^{d-1}\alpha_d r_s, q TF d − 1 α d r s = ( 2 π ) d C d 2 C 2 A d , ( 2 π ) 2 d C 2 A d 2 = 4 B d q TF d − 1 α d r s ,
which gives the prefactor − B d / 4 -B_d/4 − B d /4 of main-text Eq. (33) . Since ( b − b ′ ) 2 = ( μ − μ ′ ) 2 b 2 b ′ 2 (b-b')^2=(\mu-\mu')^2b^2b'^2 ( b − b ′ ) 2 = ( μ − μ ′ ) 2 b 2 b ′2 and ∣ p ^ − p ^ ′ ∣ 2 ≥ ( μ − μ ′ ) 2 \lvert\hat{\mathbf{p}}-\hat{\mathbf{p}}'\rvert^2\ge(\mu-\mu')^2 ∣ p ^ − p ^ ′ ∣ 2 ≥ ( μ − μ ′ ) 2 , the integrand obeys
∣ μ μ ′ Re [ ( b − b ′ ) 2 ] ∣ p ^ − p ^ ′ ∣ d − 1 ∣ ≤ ∣ p ^ − p ^ ′ ∣ 3 − d ∣ μ μ ′ ∣ ∣ b ∣ 2 ∣ b ′ ∣ 2 ≤ ∣ p ^ − p ^ ′ ∣ 3 − d z − 4 , \left\lvert\frac{\mu\mu'\operatorname{Re}\left[(b-b')^2\right]}{\lvert\hat{\mathbf{p}}-\hat{\mathbf{p}}'\rvert^{d-1}}\right\rvert\le\lvert\hat{\mathbf{p}}-\hat{\mathbf{p}}'\rvert^{3-d}\lvert\mu\mu'\rvert\lvert b\rvert^2\lvert b'\rvert^2\le\lvert\hat{\mathbf{p}}-\hat{\mathbf{p}}'\rvert^{3-d}z^{-4}, ∣ p ^ − p ^ ′ ∣ d − 1 μ μ ′ Re [ ( b − b ′ ) 2 ] ≤ ∣ p ^ − p ^ ′ ∣ 3 − d ∣ μ μ ′ ∣ ∣ b ∣ 2 ∣ b ′ ∣ 2 ≤ ∣ p ^ − p ^ ′ ∣ 3 − d z − 4 ,
which is integrable over pairs of directions in every dimension, uniformly in z ≥ z 0 > 0 z\ge z_0>0 z ≥ z 0 > 0 . The static limit z → 0 z\to0 z → 0 exists as a principal value in μ \mu μ , since the terms singular at μ = 0 \mu=0 μ = 0 are odd in μ \mu μ . Taken separately, the vertex and self-energy parts contain the nonintegrable factor ∣ p ^ − p ^ ′ ∣ 1 − d \lvert\hat{\mathbf{p}}-\hat{\mathbf{p}}'\rvert^{1-d} ∣ p ^ − p ^ ′ ∣ 1 − d near p ^ = p ^ ′ \hat{\mathbf{p}}=\hat{\mathbf{p}}' p ^ = p ^ ′ . Cut off at angles of order q q q at finite q q q , it produces their Fermi-surface factors log ( 1 / q ) \log(1/q) log ( 1/ q ) .
Three dimensions. At fixed μ , μ ′ \mu,\mu' μ , μ ′ the average over the relative azimuth is
⟨ 1 ∣ p ^ − p ^ ′ ∣ 2 ⟩ = ⟨ 1 2 ( 1 − cos Θ ) ⟩ = 1 2 ∣ μ − μ ′ ∣ , \left\langle\frac{1}{\lvert\hat{\mathbf{p}}-\hat{\mathbf{p}}'\rvert^2}\right\rangle=\left\langle\frac{1}{2(1-\cos\Theta)}\right\rangle=\frac{1}{2\lvert\mu-\mu'\rvert}, ⟨ ∣ p ^ − p ^ ′ ∣ 2 1 ⟩ = ⟨ 2 ( 1 − cos Θ ) 1 ⟩ = 2 ∣ μ − μ ′ ∣ 1 ,
and with B 3 = 4 / π B_3=4/\pi B 3 = 4/ π
R 3 ( z ) = − 1 8 π ∫ − 1 1 d μ ∫ − 1 1 d μ ′ μ μ ′ ∣ μ − μ ′ ∣ Re 1 ( i z − μ ) 2 ( i z − μ ′ ) 2 = R ( 1 ) ( z ) 4 π . (70) R_3(z)=-\frac{1}{8\pi}\int_{-1}^{1}\mathrm{d}\mu\int_{-1}^{1}\mathrm{d}\mu'\,\mu\mu'\lvert\mu-\mu'\rvert\operatorname{Re}\frac{1}{(iz-\mu)^2(iz-\mu')^2}=\frac{R^{(1)}(z)}{4\pi}. \tag*{(70)} R 3 ( z ) = − 8 π 1 ∫ − 1 1 d μ ∫ − 1 1 d μ ′ μ μ ′ ∣ μ − μ ′ ∣ Re ( i z − μ ) 2 ( i z − μ ′ ) 2 1 = 4 π R ( 1 ) ( z ) . (70)
Quadrature on the triangle μ ′ < μ \mu' < \mu μ ′ < μ gives
R 3 ( 0.1 ) = 0.23049451763139286 , R 3 ( 0.3 ) = 0.11645170899574007 , R_3(0.1)=0.23049451763139286,\qquad R_3(0.3)=0.11645170899574007, R 3 ( 0.1 ) = 0.23049451763139286 , R 3 ( 0.3 ) = 0.11645170899574007 ,
R 3 ( 1 ) = 0.011619772367581344 , R 3 ( 3 ) = 0.0002451548733833656 , R_3(1)=0.011619772367581344,\qquad R_3(3)=0.0002451548733833656, R 3 ( 1 ) = 0.011619772367581344 , R 3 ( 3 ) = 0.0002451548733833656 ,
equal to R ( 1 ) / ( 4 π ) R^{(1)}/(4\pi) R ( 1 ) / ( 4 π ) of main-text Eq. (35) within 2.5 × 10 − 17 2.5 \times10^{-17} 2.5 × 1 0 − 17 . With
∫ − ∞ ∞ h R 3 d z = π 2 − 12 log 2 12 , ∫ − ∞ ∞ h 2 g 3 d z = − π ( π 2 − 6 ) 288 , \int_{-\infty}^{\infty} hR_3\,\mathrm{d}z=\frac{\pi^2-12\log2}{12},\qquad
\int_{-\infty}^{\infty} h^2g_3\,\mathrm{d}z=-\frac{\pi(\pi^2-6)}{288}, ∫ − ∞ ∞ h R 3 d z = 12 π 2 − 12 log 2 , ∫ − ∞ ∞ h 2 g 3 d z = − 288 π ( π 2 − 6 ) ,
the two parts of the Carr–Maradudin coefficient and their sum are
α L ( π 2 − 12 log 2 ) 2 π 3 + α L ( π 2 − 6 ) 12 π 3 = 0.0184584246606864 … R y . \frac{\alpha_L(\pi^2-12\log2)}{2\pi^3}+\frac{\alpha_L(\pi^2-6)}{12\pi^3}=0.0184584246606864\ldots\ \mathrm{Ry}. 2 π 3 α L ( π 2 − 12 log 2 ) + 12 π 3 α L ( π 2 − 6 ) = 0.0184584246606864 … Ry .
The sum agrees with the closed form of the coefficient C C C of main-text Eq. (1) to 60 digits.
Closed form of R 2 R_2 R 2 and the log ( 1 / q ) \log(1/q) log ( 1/ q ) terms of the vertex and self-energy parts in two dimensions
Fourier evaluation. For d = 2 d=2 d = 2 , B 2 = 2 B_2=2 B 2 = 2 and ∣ p ^ − p ^ ′ ∣ = 2 ∣ sin ( ϕ / 2 ) ∣ |\hat{\mathbf p}-\hat{\mathbf p}'|=2|\sin(\phi/2)| ∣ p ^ − p ^ ′ ∣ = 2∣ sin ( ϕ /2 ) ∣ , with θ \theta θ and θ ′ = θ + ϕ \theta'=\theta+\phi θ ′ = θ + ϕ the polar angles of p ^ \hat{\mathbf p} p ^ and p ^ ′ \hat{\mathbf p}' p ^ ′ relative to q \mathbf q q . Let
T ( ϕ ) = ⟨ cos θ cos θ ′ Re [ ( b − b ′ ) 2 ] ⟩ θ = ∑ m T m e i m ϕ , R 2 = − 1 2 ⟨ T ( ϕ ) 2 ∣ sin ( ϕ / 2 ) ∣ ⟩ ϕ . T(\phi)=\left\langle\cos\theta\cos\theta'\operatorname{Re}[(b-b')^2]\right\rangle_\theta=\sum_m T_m e^{im\phi},\qquad
R_2=-\frac{1}{2}\left\langle\frac{T(\phi)}{2|\sin(\phi/2)|}\right\rangle_\phi. T ( ϕ ) = ⟨ cos θ cos θ ′ Re [( b − b ′ ) 2 ] ⟩ θ = m ∑ T m e im ϕ , R 2 = − 2 1 ⟨ 2∣ sin ( ϕ /2 ) ∣ T ( ϕ ) ⟩ ϕ .
The Coulomb factor has the Fourier coefficients
⟨ e i m ϕ 2 ∣ sin ϕ 2 ∣ ⟩ ϕ = K 0 − 2 π O ∣ m ∣ , O m = ∑ j = 1 m 1 2 j − 1 , \left\langle\frac{e^{im\phi}}{2|\sin\frac{\phi}{2}|}\right\rangle_\phi=K_0-\frac{2}{\pi}O_{|m|},\qquad
O_m=\sum_{j=1}^{m}\frac{1}{2j-1}, ⟨ 2∣ sin 2 ϕ ∣ e im ϕ ⟩ ϕ = K 0 − π 2 O ∣ m ∣ , O m = j = 1 ∑ m 2 j − 1 1 ,
with the divergent constant K 0 K_0 K 0 . Since T ( 0 ) = ∑ m T m = 0 T(0)=\sum_m T_m=0 T ( 0 ) = ∑ m T m = 0 , K 0 K_0 K 0 drops out (with an angular cutoff ϵ \epsilon ϵ the constant is finite and multiplies T ( 0 ) = 0 T(0)=0 T ( 0 ) = 0 ), and R 2 = π − 1 ∑ m O ∣ m ∣ T m R_2=\pi^{-1}\sum_m O_{|m|}T_m R 2 = π − 1 ∑ m O ∣ m ∣ T m . With s z = 1 + z 2 s_z=\sqrt{1+z^2} s z = 1 + z 2 and r = s z − z r=s_z-z r = s z − z ,
b ( θ ) = ( i z − cos θ ) − 1 = ( i s z ) − 1 ∑ m ( − i r ) ∣ m ∣ e i m θ . b(\theta)=(iz-\cos\theta)^{-1}=(is_z)^{-1}\sum_m(-ir)^{|m|}e^{im\theta}. b ( θ ) = ( i z − cos θ ) − 1 = ( i s z ) − 1 m ∑ ( − i r ) ∣ m ∣ e im θ .
The functions F 0 = cos θ F_0=\cos\theta F 0 = cos θ , F 1 = cos θ b F_1=\cos\theta b F 1 = cos θ b and F 2 = cos θ b 2 F_2=\cos\theta b^2 F 2 = cos θ b 2 have Fourier coefficients even in the index,
F 0 , ± 1 = 1 2 , F 1 , m = ( − i r ) m − 1 ( 1 − r 2 ) 2 i s z ( m ≥ 1 ) , F 2 , ± 1 = ⟨ cos 2 θ b 2 ⟩ . F_{0,\pm1}=\frac{1}{2},\qquad
F_{1,m}=\frac{(-ir)^{m-1}(1-r^2)}{2is_z}\quad(m\ge1),\qquad
F_{2,\pm1}=\langle\cos^2\theta\,b^2\rangle. F 0 , ± 1 = 2 1 , F 1 , m = 2 i s z ( − i r ) m − 1 ( 1 − r 2 ) ( m ≥ 1 ) , F 2 , ± 1 = ⟨ cos 2 θ b 2 ⟩ .
Then T m = 2 Re ( F 2 , m F 0 , m ) − 2 Re ( F 1 , m 2 ) T_m=2\operatorname{Re}(F_{2,m}F_{0,m})-2\operatorname{Re}(F_{1,m}^2) T m = 2 Re ( F 2 , m F 0 , m ) − 2 Re ( F 1 , m 2 ) , where the first term comes from the self-energy terms b 2 + b ′ 2 b^2+b'^2 b 2 + b ′2 and the second from the vertex term − 2 b b ′ -2bb' − 2 b b ′ . With (S2),
F 1 , m 2 = ( − 1 ) m r 2 ∣ m ∣ a 2 , Re F 2 , ± 1 = ⟨ cos 2 θ Re b 2 ⟩ = h 2 + z h 2 ′ , F_{1,m}^2=(-1)^m r^{2|m|}a^2,\qquad
\operatorname{Re}F_{2,\pm1}=\langle\cos^2\theta\operatorname{Re}b^2\rangle=h_2+zh_2', F 1 , m 2 = ( − 1 ) m r 2∣ m ∣ a 2 , Re F 2 , ± 1 = ⟨ cos 2 θ Re b 2 ⟩ = h 2 + z h 2 ′ ,
hence, for m ≠ 0 m\ne0 m = 0 ,
T m = δ ∣ m ∣ , 1 ( h 2 + z h 2 ′ ) − 2 ( − 1 ) m r 2 ∣ m ∣ a 2 , R 2 = 2 π [ h 2 + z h 2 ′ − 2 a 2 ∑ m ≥ 1 O m ( − r 2 ) m ] . T_m=\delta_{|m|,1}(h_2+zh_2')-2(-1)^m r^{2|m|}a^2,\qquad
R_2=\frac{2}{\pi}\left[h_2+zh_2'-2a^2\sum_{m\ge1}O_m(-r^2)^m\right]. T m = δ ∣ m ∣ , 1 ( h 2 + z h 2 ′ ) − 2 ( − 1 ) m r 2∣ m ∣ a 2 , R 2 = π 2 [ h 2 + z h 2 ′ − 2 a 2 m ≥ 1 ∑ O m ( − r 2 ) m ] .
The generating function at x = − r 2 x=-r^{2} x = − r 2 gives
∑ m ≥ 1 O m x m = x artanh x 1 − x , ∑ m ≥ 1 O m ( − r 2 ) m = − r arctan r 1 + r 2 , \sum_{m\ge1} O_m x^m=\frac{\sqrt{x}\operatorname{artanh}\sqrt{x}}{1-x},\qquad\sum_{m\ge1}O_m(-r^2)^m=-\frac{r\arctan r}{1+r^2}, m ≥ 1 ∑ O m x m = 1 − x x artanh x , m ≥ 1 ∑ O m ( − r 2 ) m = − 1 + r 2 r arctan r ,
and
R 2 = 2 π [ h 2 + z h 2 ′ + 2 a 2 r arctan r 1 + r 2 ] , h 2 + z h 2 ′ = h 2 2 − 2 a 2 r 2 1 + r 2 , (71) R_2=\frac{2}{\pi}\left[h_2+zh_2'+\frac{2a^2r\arctan r}{1+r^2}\right],\qquad h_2+zh_2'=h_2^2-\frac{2a^2r^2}{1+r^2}, \tag*{(71)} R 2 = π 2 [ h 2 + z h 2 ′ + 1 + r 2 2 a 2 r arctan r ] , h 2 + z h 2 ′ = h 2 2 − 1 + r 2 2 a 2 r 2 , (71)
which is main-text Eq. (37) . The second equality follows from (S2) and
h 2 ′ = − s z − 3 , h 2 + z h 2 ′ = 1 − z s z − z s z 3 . h_2'=-s_z^{-3},\qquad h_2+zh_2'=1-\frac{z}{s_z}-\frac{z}{s_z^3}. h 2 ′ = − s z − 3 , h 2 + z h 2 ′ = 1 − s z z − s z 3 z .
Checks. With μ = cos θ \mu=\cos\theta μ = cos θ , μ ′ = cos θ ′ \mu'=\cos\theta' μ ′ = cos θ ′ , σ = ( θ + θ ′ ) / 2 \sigma=(\theta+\theta')/2 σ = ( θ + θ ′ ) /2 and δ = ( θ − θ ′ ) / 2 \delta=(\theta-\theta')/2 δ = ( θ − θ ′ ) /2 , ( μ − μ ′ ) 2 = 4 sin 2 σ sin 2 δ (\mu-\mu')^2=4\sin^2\sigma\sin^2\delta ( μ − μ ′ ) 2 = 4 sin 2 σ sin 2 δ , so the angular integrand is smooth in ( σ , δ ) (\sigma,\delta) ( σ , δ ) . Gauss–Legendre quadrature in these variables, independent of the Fourier sum, gives R 2 ( 1 ) = 0.04977088289807023 R_2(1)=0.04977088289807023 R 2 ( 1 ) = 0.04977088289807023 , within 7 × 10 − 18 7\times10^{-18} 7 × 1 0 − 18 of (S9), and agrees with it to 10 − 13 10^{-13} 1 0 − 13 at z = 0.3 z=0.3 z = 0.3 and 3 3 3 . At z = 0 z=0 z = 0 , R 2 = 2 / π R_2=2/\pi R 2 = 2/ π . At large z z z , (S9) with
h 2 = 1 2 z 2 − 3 8 z 4 + ⋯ , r = 1 2 z + ⋯ h_2=\frac{1}{2z^2}-\frac{3}{8z^4}+\cdots,\qquad r=\frac{1}{2z}+\cdots h 2 = 2 z 2 1 − 8 z 4 3 + ⋯ , r = 2 z 1 + ⋯
gives
R 2 = 5 12 π z 4 − 11 20 π z 6 + 1283 2240 π z 8 + O ( z − 10 ) . R_2=\frac{5}{12\pi z^4}-\frac{11}{20\pi z^6}+\frac{1283}{2240\pi z^8}+O(z^{-10}). R 2 = 12 π z 4 5 − 20 π z 6 11 + 2240 π z 8 1283 + O ( z − 10 ) .
Factor log ( 1 / q ) \log(1/q) log ( 1/ q ) of the vertex part. The vertex part alone is
− 1 2 ⟨ μ μ ′ Re ( − 2 b b ′ ) 2 ∣ sin ϕ 2 ∣ ⟩ . -\frac{1}{2}\left\langle\frac{\mu\mu'\operatorname{Re}(-2bb')}{2\lvert\sin\frac{\phi}{2}\rvert}\right\rangle. − 2 1 ⟨ 2 ∣ sin 2 ϕ ∣ μ μ ′ Re ( − 2 b b ′ ) ⟩ .
With an angular cutoff ∣ ϕ ∣ > ϵ \lvert\phi\rvert>\epsilon ∣ ϕ ∣ > ϵ , the average of the Coulomb factor is
⟨ 1 2 ∣ sin ϕ 2 ∣ ⟩ ∣ ϕ ∣ > ϵ = 1 π log 1 ϵ + O ( 1 ) . \left\langle\frac{1}{2\lvert\sin\frac{\phi}{2}\rvert}\right\rangle_{\lvert\phi\rvert>\epsilon}=\frac{1}{\pi}\log\frac{1}{\epsilon}+O(1). ⟨ 2 ∣ sin 2 ϕ ∣ 1 ⟩ ∣ ϕ ∣ > ϵ = π 1 log ϵ 1 + O ( 1 ) .
Near ϕ = 0 \phi=0 ϕ = 0 , μ μ ′ Re ( b b ′ ) → cos 2 θ Re b 2 \mu\mu'\operatorname{Re}(bb')\to\cos^2\theta\operatorname{Re}b^2 μ μ ′ Re ( b b ′ ) → cos 2 θ Re b 2 , and therefore the vertex part grows as
1 π ⟨ cos 2 θ Re b 2 ⟩ log 1 ϵ = 1 π ( h 2 + z h 2 ′ ) log 1 ϵ . \frac{1}{\pi}\langle\cos^2\theta\operatorname{Re}b^2\rangle\log\frac{1}{\epsilon}=\frac{1}{\pi}(h_2+zh_2')\log\frac{1}{\epsilon}. π 1 ⟨ cos 2 θ Re b 2 ⟩ log ϵ 1 = π 1 ( h 2 + z h 2 ′ ) log ϵ 1 .
The self-energy part has the same factor log ( 1 / ϵ ) \log(1/\epsilon) log ( 1/ ϵ ) with the opposite sign, since the sum of the two parts is finite. At finite q q q the shell width cuts off the angle at ϵ ∼ q \epsilon\sim q ϵ ∼ q , which gives the coefficient of log ( 1 / q ) \log(1/q) log ( 1/ q ) in main-text property (iii) of Sec. 3.1. Between ϵ = 10 − 5 \epsilon=10^{-5} ϵ = 1 0 − 5 and 10 − 7 10^{-7} 1 0 − 7 , a quadrature with a symmetric cutoff gives the slopes per unit of log ( 1 / ϵ ) \log(1/\epsilon) log ( 1/ ϵ ) .
Table 5.
At z = 0 z=0 z = 0 the coefficient 1 / π 1/\pi 1/ π reproduces the terms ∓ ( r s / 2 π ) log ( k F / q ) \mp(r_s/\sqrt{2}\pi)\log(k_F/q) ∓ ( r s / 2 π ) log ( k F / q ) of Eqs. (21) and (22) of Ref. [36 ] . In three dimensions the same argument gives main-text Eq. (59) , since q T F 2 α L r s / π = q T F 4 / 4 q_{\mathrm{TF}}^2\alpha_Lr_s/\pi=q_{\mathrm{TF}}^4/4 q TF 2 α L r s / π = q TF 4 /4 .
Isolated vertex in the ring chain. Inserted alone into main-text Eq. (32) , the vertex part has at small q q q the kernel q q T F d − 1 α d r s π − 1 ( h d + z h d ′ ) log ( 1 / q ) q q_{\mathrm{TF}}^{d-1}\alpha_d r_s\pi^{-1}(h_d+zh_d')\log(1/q) q q TF d − 1 α d r s π − 1 ( h d + z h d ′ ) log ( 1/ q ) . With the tail coefficient γ N ∗ = ( − h d ) N ∗ \gamma_{N_*}=(-h_d)^{N_*} γ N ∗ = ( − h d ) N ∗ , Eq. (140) with m = 1 m=1 m = 1 gives the triple-pole term of the energy e v t x e_{\mathrm{vtx}} e vtx of this series,
e v t x ⊃ d 8 π E 0 α d r s q T F d + 1 ( − 1 ) N ∗ 2 π ( d − 1 ) 2 ∫ − ∞ ∞ h d N ∗ ( h d + z h d ′ ) d z log 2 r s . (72) e_{\mathrm{vtx}}\supset\frac{d}{8\pi}E_0\alpha_d r_s q_{\mathrm{TF}}^{d+1}\frac{(-1)^{N_*}}{2\pi(d-1)^2}\int_{-\infty}^{\infty}h_d^{N_*}(h_d+zh_d')\,\mathrm{d}z\log^2 r_s. \tag*{(72)} e vtx ⊃ 8 π d E 0 α d r s q TF d + 1 2 π ( d − 1 ) 2 ( − 1 ) N ∗ ∫ − ∞ ∞ h d N ∗ ( h d + z h d ′ ) d z log 2 r s . (72)
Integration by parts, with z h d N ∗ + 1 → 0 zh_{d}^{N_*+1} \to0 z h d N ∗ + 1 → 0 at both ends, gives
∫ z h d N ∗ h d ′ d z = ∫ z d ( h d N ∗ + 1 ) N ∗ + 1 = − ∫ h d N ∗ + 1 d z N ∗ + 1 , ∫ h d N ∗ ( h d + z h d ′ ) d z = N ∗ N ∗ + 1 ∫ h d N ∗ + 1 d z . \int zh_{d}^{N_*}h_{d}'\,dz=\int\frac{z\,d\left(h_{d}^{N_*+1}\right)}{N_*+1}=-\frac{\int h_{d}^{N_*+1}\,dz}{N_*+1},\qquad\int h_{d}^{N_*}\left(h_{d}+zh_{d}'\right)\,dz=\frac{N_*}{N_*+1}\int h_{d}^{N_*+1}\,dz. ∫ z h d N ∗ h d ′ d z = ∫ N ∗ + 1 z d ( h d N ∗ + 1 ) = − N ∗ + 1 ∫ h d N ∗ + 1 d z , ∫ h d N ∗ ( h d + z h d ′ ) d z = N ∗ + 1 N ∗ ∫ h d N ∗ + 1 d z .
With table S3,
d = 2 : d 8 π E 0 α 2 r s q T F 3 = 2 r s 2 π R y , e v t x ⊃ 2 3 π 2 ( 10 − 3 π ) r s 2 log 2 r s , d=2:\qquad\frac{d}{8\pi}E_{0}\alpha_{2}r_{s}q_{\mathrm{TF}}^{3}=\frac{2r_{s}^{2}}{\pi} \,\mathrm{Ry},\qquad e_{\mathrm{vtx}}\supset\frac{2}{3\pi^{2}}(10-3\pi)r_{s}^{2}\log^{2}r_{s}, d = 2 : 8 π d E 0 α 2 r s q TF 3 = π 2 r s 2 Ry , e vtx ⊃ 3 π 2 2 ( 10 − 3 π ) r s 2 log 2 r s ,
d = 3 : 3 8 π E 0 α L r s q T F 4 = 12 α L r s π 3 , e v t x ⊃ − α L ( 1 − log 2 ) 2 π 3 r s log 2 r s = 3 c h d r s log 2 r s . d=3:\qquad\frac{3}{8\pi}E_{0}\alpha_{L}r_{s}q_{\mathrm{TF}}^{4}=\frac{12\alpha_{L}r_{s}}{\pi^{3}},\qquad e_{\mathrm{vtx}}\supset-\frac{\alpha_{L}(1-\log2)}{2\pi^{3}}r_{s}\log^{2}r_{s}=3c_{\mathrm{hd}}r_{s}\log^{2}r_{s}. d = 3 : 8 π 3 E 0 α L r s q TF 4 = π 3 12 α L r s , e vtx ⊃ − 2 π 3 α L ( 1 − log 2 ) r s log 2 r s = 3 c hd r s log 2 r s .
The first is main-text Eq. (61) . The second is the chain-family coefficient of main-text Eq. (50) , which checks the normalization.
The coefficient a 2 ℓ a_{2\ell} a 2 ℓ
Residue. For d = 2 d=2 d = 2 the kernel of main-text Eq. (32) at fixed z z z is q ⋅ q T F ( r s / 2 ) R 2 ( z ) q\cdot q_{\mathrm{TF}}(r_{s}/\sqrt{2})R_{2}(z) q ⋅ q TF ( r s / 2 ) R 2 ( z ) at small q q q , and the scaling function 1 / ( 1 + c ) − 1 1/(1+c)-1 1/ ( 1 + c ) − 1 has the term h 2 2 t − 2 h_{2}^{2}t^{-2} h 2 2 t − 2 (N ∗ = 2 N_{*}=2 N ∗ = 2 ). By Eq. (140) with m = 0 m=0 m = 0 and, in its generic notation, ε = q T F \varepsilon=q_{\mathrm{TF}} ε = q TF , κ = q T F ( r s / 2 ) R 2 \kappa=q_{\mathrm{TF}}(r_{s}/\sqrt{2})R_{2} κ = q TF ( r s / 2 ) R 2 , γ = h 2 2 \gamma=h_{2}^{2} γ = h 2 2 and β = 2 \beta=2 β = 2 , the double pole contributes to the q q q integral
− q T F r s 2 R 2 h 2 2 q T F 2 log q T F . -q_{\mathrm{TF}}\frac{r_{s}}{\sqrt{2}}R_{2}h_{2}^{2}q_{\mathrm{TF}}^{2}\log q_{\mathrm{TF}}. − q TF 2 r s R 2 h 2 2 q TF 2 log q TF .
With log q T F = log r s + O ( 1 ) \log q_{\mathrm{TF}}=\log r_{s}+O(1) log q TF = log r s + O ( 1 ) and the prefactor ( 1 / 4 π ) E 0 = 1 / ( π r s 2 ) R y (1/4\pi)E_{0}=1/(\pi r_{s}^{2})\,\mathrm{Ry} ( 1/4 π ) E 0 = 1/ ( π r s 2 ) Ry this gives main-text Eq. (38) ,
e Π ( 1 ) ⊃ − 1 π r s 2 q T F 3 r s 2 ∫ h 2 2 R 2 d z log r s = − 2 π ∫ h 2 2 R 2 d z r s 2 log r s . e_{\Pi^{(1)}}\supset-\frac{1}{\pi r_{s}^{2}}q_{\mathrm{TF}}^{3}\frac{r_{s}}{\sqrt{2}}\int h_{2}^{2}R_{2}\,dz\log r_{s}=-\frac{2}{\pi}\int h_{2}^{2}R_{2}\,dz\,r_{s}^{2}\log r_{s}. e Π ( 1 ) ⊃ − π r s 2 1 q TF 3 2 r s ∫ h 2 2 R 2 d z log r s = − π 2 ∫ h 2 2 R 2 d z r s 2 log r s .
The corrections to the limiting forms are controlled as for d = 3 d=3 d = 3 in § 25.6 . By Eq. (66) , replacing L \mathcal{L} L by h 2 h_{2} h 2 changes c c c by O ( q T F q ) O(q_{\mathrm{TF}}q) O ( q TF q ) . The region q > 1 q>1 q > 1 contributes only at O ( r s ) O(r_{s}) O ( r s ) and O ( r s 2 ) O(r_{s}^{2}) O ( r s 2 ) , without log r s \log r_{s} log r s . The corrections of order q 2 log q q^{2}\log q q 2 log q to R 2 R_{2} R 2 (§ 17 ) enter at O ( r s 4 log 2 r s ) O(r_{s}^{4}\log^{2}r_{s}) O ( r s 4 log 2 r s ) .
Closed form. In Hartree, with the first form of main-text Eq. (37) and Eq. (S2),
a 2 ℓ 2 = − 1 π ∫ − ∞ ∞ h 2 2 R 2 d z = − 8 π 2 ∫ 0 1 r 2 1 + r 2 [ h 2 2 + 2 a 2 r arctan r − r 2 1 + r 2 ] d r , \frac{a_{2\ell}}{2}=-\frac{1}{\pi}\int_{-\infty}^{\infty}h_{2}^{2}R_{2}\,dz=-\frac{8}{\pi^{2}}\int_{0}^{1}\frac{r^{2}}{1+r^{2}}\left[h_{2}^{2}+2a^{2}\frac{r\arctan r-r^{2}}{1+r^{2}}\right]dr, 2 a 2 ℓ = − π 1 ∫ − ∞ ∞ h 2 2 R 2 d z = − π 2 8 ∫ 0 1 1 + r 2 r 2 [ h 2 2 + 2 a 2 1 + r 2 r arctan r − r 2 ] d r ,
and therefore
a 2 ℓ 2 = − 32 A + 16 B − 16 C π 2 , A = ∫ 0 1 r 6 d r ( 1 + r 2 ) 3 , \frac{a_{2\ell}}{2}=\frac{-32A+16B-16C}{\pi^{2}},\qquad A=\int_{0}^{1}\frac{r^{6}\,dr}{(1+r^{2})^{3}}, 2 a 2 ℓ = π 2 − 32 A + 16 B − 16 C , A = ∫ 0 1 ( 1 + r 2 ) 3 r 6 d r ,
B = ∫ 0 1 r 4 ( 1 − r 2 ) 2 d r ( 1 + r 2 ) 4 , C = ∫ 0 1 r 3 ( 1 − r 2 ) 2 arctan r ( 1 + r 2 ) 4 d r . (73) B=\int_{0}^{1}\frac{r^{4}(1-r^{2})^{2}\,dr}{(1+r^{2})^{4}},\qquad C=\int_{0}^{1}\frac{r^{3}(1-r^{2})^{2}\arctan r}{(1+r^{2})^{4}}\,dr. \tag*{(73)} B = ∫ 0 1 ( 1 + r 2 ) 4 r 4 ( 1 − r 2 ) 2 d r , C = ∫ 0 1 ( 1 + r 2 ) 4 r 3 ( 1 − r 2 ) 2 arctan r d r . (73)
The first two are integrals of rational functions,
A = 3 2 − 15 π 32 , B = 13 6 − 11 π 16 . A=\frac{3}{2}-\frac{15\pi}{32},\qquad B=\frac{13}{6}-\frac{11\pi}{16}. A = 2 3 − 32 15 π , B = 6 13 − 16 11 π .
For C C C , with u = 1 + r 2 u=1+r^{2} u = 1 + r 2 , an antiderivative of the rational factor is
Φ = 1 2 log u + 5 2 u − 2 u 2 + 2 3 u 3 , Φ ∣ r = 1 = 1 2 log 2 + 5 6 . \Phi=\frac{1}{2}\log u+\frac{5}{2u}-\frac{2}{u^{2}}+\frac{2}{3u^{3}},\qquad\left.\Phi\right|_{r=1}=\frac{1}{2}\log2+\frac{5}{6}. Φ = 2 1 log u + 2 u 5 − u 2 2 + 3 u 3 2 , Φ ∣ r = 1 = 2 1 log 2 + 6 5 .
One integration by parts gives
C = − π 4 Φ ∣ r = 1 − ∫ 0 1 Φ d r 1 + r 2 , ∫ 0 1 Φ d r 1 + r 2 = 1 2 ∫ 0 1 log ( 1 + r 2 ) 1 + r 2 d r + 5 18 + 17 π 96 . C=\left.-\frac{\pi}{4}\Phi\right|_{r=1}-\int_{0}^{1}\frac{\Phi\,dr}{1+r^{2}},\qquad\int_{0}^{1}\frac{\Phi\,dr}{1+r^{2}}=\frac{1}{2}\int_{0}^{1}\frac{\log(1+r^{2})}{1+r^{2}}\,dr+\frac{5}{18}+\frac{17\pi}{96}. C = − 4 π Φ r = 1 − ∫ 0 1 1 + r 2 Φ d r , ∫ 0 1 1 + r 2 Φ d r = 2 1 ∫ 0 1 1 + r 2 log ( 1 + r 2 ) d r + 18 5 + 96 17 π .
It leaves the integral ∫ 0 1 log ( 1 + r 2 ) / ( 1 + r 2 ) d r = π 8 log 2 − G \int_{0}^{1} \log(1+r^{2})/(1+r^{2})\,\mathrm{d}r = \frac{\pi}{8}\log2-G ∫ 0 1 log ( 1 + r 2 ) / ( 1 + r 2 ) d r = 8 π log 2 − G , with G G G Catalan’s constant, and gives
C = − 5 18 − π log 2 8 + π 32 + G 2 . C=-\frac{5}{18}-\frac{\pi\log2}{8}+\frac{\pi}{32}+\frac{G}{2}. C = − 18 5 − 8 π log 2 + 32 π + 2 G .
Collecting terms gives main-text Eq. (39) ,
a 2 ℓ = 2 π 2 ( − 32 A + 16 B − 16 C ) = 7 π + 4 log 2 π − 160 9 π 2 − 16 G π 2 . a_{2\ell}=\frac{2}{\pi^{2}}(-32A+16B-16C)=\frac{7}{\pi}+\frac{4\log2}{\pi}-\frac{160}{9\pi^{2}}-\frac{16G}{\pi^{2}}. a 2 ℓ = π 2 2 ( − 32 A + 16 B − 16 C ) = π 7 + π 4 log 2 − 9 π 2 160 − π 2 16 G .
Multiprecision quadrature of (S11) in r r r agrees with it to 10 − 41 10^{-41} 1 0 − 41 (40 digits) and to 3 × 10 − 61 3\times10^{-61} 3 × 1 0 − 61 (60 digits), and quadrature of (38) in z z z to 100 digits. An integer-relation search (PSLQ) on π 2 \pi^{2} π 2 times this value over { 1 , π , G , π log 2 } \{1,\pi,G,\pi\log2\} { 1 , π , G , π log 2 } returns the same combination.
Finite wave number in two dimensions
Method. At finite q q q we evaluate (S6) for d = 2 d=2 d = 2 directly. With k F = 1 k_{\mathrm{F}}=1 k F = 1 and q \mathbf{q} q along the x x x axis, Δ p ≠ 0 \Delta_{p}\ne0 Δ p = 0 only on the two crescents between the unit circle and its translate. We use the coordinates
ρ = ρ ∗ ( φ ) + ( 1 − ρ ∗ ( φ ) ) t , 0 < t < 1 , ρ ∗ ( φ ) = − q cos φ + ( 1 − q 2 sin 2 φ ) 1 / 2 . \rho=\rho_{*}(\varphi)+(1-\rho_{*}(\varphi))t,\qquad0<t<1,\qquad\rho_{*}(\varphi)=-q\cos\varphi+(1-q^{2}\sin^{2}\varphi)^{1/2}. ρ = ρ ∗ ( φ ) + ( 1 − ρ ∗ ( φ )) t , 0 < t < 1 , ρ ∗ ( φ ) = − q cos φ + ( 1 − q 2 sin 2 φ ) 1/2 .
with ρ ∗ \rho_{*} ρ ∗ the polar radius of the translated circle. The measure is signed and covers both signs of Δ p \Delta_{p} Δ p . The Gauss–Legendre nodes in φ \varphi φ are split at cos φ = − q / 2 \cos\varphi=-q/2 cos φ = − q /2 , where the crescents close. The scaled polarization at finite wave number is R ( q , z ) = − Π ( 1 ) / [ q T F ( r s / 2 ) ] R(q,z)=-\Pi^{(1)}/[q_{\mathrm{TF}}(r_{s}/\sqrt{2})] R ( q , z ) = − Π ( 1 ) / [ q TF ( r s / 2 )] .
Results. Table S4 lists R ( q , z ) − R 2 ( z ) R(q,z)-R_{2}(z) R ( q , z ) − R 2 ( z ) . At z = 0.3 z=0.3 z = 0.3 the differences decrease by factors of about 3.6 when q q q is halved, as for a correction q 2 ( a log q + b ) q^{2}(a\log q+b) q 2 ( a log q + b ) and not for an O ( q ) O(q) O ( q ) term, which would give a factor 2. At z = 1 z=1 z = 1 they are below 10 − 5 10^{-5} 1 0 − 5 in magnitude and change sign. Refining the nodes changes the tabulated values by less than 10 − 13 10^{-13} 1 0 − 13 . At z = 0 z=0 z = 0 we compare Δ R / R ( 0 ) \Delta R/R(0) Δ R / R ( 0 ) with the values of Δ P / P ( 0 ) \Delta P/P(0) Δ P / P ( 0 ) in Table I of [36 ] , which gives the small-x x x form (its Eq. (25) divided by P ( 0 ) = − 2 / π 2 P(0)=-2/\pi^{2} P ( 0 ) = − 2/ π 2 )
Δ P P ( 0 ) = x 2 ( log x + β ) 96 , β = 6.6441 , x = q / k F . \frac{\Delta P}{P(0)}=\frac{x^{2}(\log x+\beta)}{96},\qquad\beta=6.6441,\qquad x=q/k_{\mathrm{F}}. P ( 0 ) Δ P = 96 x 2 ( log x + β ) , β = 6.6441 , x = q / k F .
Our values and those of that table are
We attribute the differences of 1–4% of Δ R \Delta R Δ R to our quadrature, which is not adapted to the singularity of the static integrand at coincident momenta at z = 0 z=0 z = 0 .
Table S4. Finite-q q q first-order exchange polarization of the two-dimensional gas: R ( q , z ) − R 2 ( z ) R(q,z)-R_{2}(z) R ( q , z ) − R 2 ( z ) , with R 2 ( 0.3 ) = 0.3180676 R_{2}(0.3)=0.3180676 R 2 ( 0.3 ) = 0.3180676 and R 2 ( 1 ) = 0.0497709 R_{2}(1)=0.0497709 R 2 ( 1 ) = 0.0497709 .
Order counting and completeness
One soft line. Let an insertion of order r s j r_s^j r s j have the small-q q q form
Π ( j ) ( q , i q z ) ≃ − q T F d − 1 ( α d r s ) j q m F ( z ) . \Pi^{(j)}(\mathbf{q},iqz) \simeq-q_{\mathrm{TF}}^{d-1}(\alpha_d r_s)^j q^m F(z). Π ( j ) ( q , i q z ) ≃ − q TF d − 1 ( α d r s ) j q m F ( z ) .
In main-text Eq. (32) its kernel is q 1 + m q^{1+m} q 1 + m , with a pole at s = 2 + m s=2+m s = 2 + m , and the screening poles lie at s = N ( d − 1 ) s=N(d-1) s = N ( d − 1 ) . With E 0 ∝ r s − 2 E_0 \propto r_s^{-2} E 0 ∝ r s − 2 and q T F ∝ r s 1 / ( d − 1 ) q_{\mathrm{TF}} \propto r_s^{1/(d-1)} q TF ∝ r s 1/ ( d − 1 ) , a coincidence gives
N ( d − 1 ) = 2 + m : E 0 r s j q T F d + 1 + m log r s ∝ r s j + ( 3 − d + m ) / ( d − 1 ) log r s . N(d-1)=2+m:\qquad E_0 r_s^j q_{\mathrm{TF}}^{d+1+m}\log r_s \propto r_s^{j+(3-d+m)/(d-1)}\log r_s. N ( d − 1 ) = 2 + m : E 0 r s j q TF d + 1 + m log r s ∝ r s j + ( 3 − d + m ) / ( d − 1 ) log r s .
The ring series (j = 0 j=0 j = 0 , m = 0 m=0 m = 0 , with its own kernel q d q^d q d , Sec. S2) gives the same exponent ( 3 − d ) / ( d − 1 ) (3-d)/(d-1) ( 3 − d ) / ( d − 1 ) . The q 2 q^2 q 2 term of the Lindhard function is the case j = 0 j=0 j = 0 , m = 2 m=2 m = 2 . The exponent is j + m / 2 j+m/2 j + m /2 for d = 3 d=3 d = 3 and j + 1 + m j+1+m j + 1 + m for d = 2 d=2 d = 2 , and the orders listed in main-text Sec. 3.1 follow.
Several soft lines. After the two-point loops (free bubbles) are summed into the RPA line, the soft density fluctuations have nonlinear vertices Γ l \Gamma_l Γ l with l ≥ 3 l\ge3 l ≥ 3 legs. These are the closed fermion loops with three or more density vertices, symmetrized over the vertex orderings. For soft momenta Q = τ k F Q=\tau k_F Q = τ k F and imaginary frequencies ω ∼ v F Q \omega\sim v_F Q ω ∼ v F Q , the symmetrized loops remain bounded as Q → 0 Q\to0 Q → 0 . Neumayr and Metzner showed this in two dimensions [37 Sec. 6, Eq. (73)] . Kopper and Magnen proved it with Ward identities for d ≤ 3 d\le3 d ≤ 3 , for the quadratic dispersion and with momenta and frequencies scaled to zero together [38 ] . Their bound requires every partial sum of the external frequencies of a loop to be at least a fixed fraction η \eta η of v F Q v_F Q v F Q , and its constant grows as a power of 1 / η 1/\eta 1/ η . In physical units it reads Γ l = O ( g F k F − 2 ( l − 2 ) ) \Gamma_l=O(g_F k_F^{-2(l-2)}) Γ l = O ( g F k F − 2 ( l − 2 ) ) , with g F g_F g F the density of states at the Fermi level. Equation (74) of Ref. [37 ] refers to the dynamical limit at fixed frequency and does not apply here.
A connected vacuum diagram with V l V_l V l vertices of l l l legs, I I I soft lines and L L L soft loops has
L = I − ∑ l V l + 1 , ∑ l l V l = 2 I . L=I-\sum_l V_l+1,\qquad\sum_l lV_l=2I. L = I − l ∑ V l + 1 , l ∑ l V l = 2 I .
With each soft line of order 1 / g F 1/g_F 1/ g F and each loop integral of order k F Q d + 1 k_FQ^{d+1} k F Q d + 1 , the energy per electron scales as
k F − d [ k F Q d + 1 ] L [ k F 2 − d ] I ∏ l [ k F d + 2 − 2 l ] V l = k F − d + ( d + 2 ) ( L − I + ∑ l V l ) τ L ( d + 1 ) = k F 2 τ L ( d + 1 ) . k_F^{-d}[k_FQ^{d+1}]^L[k_F^{2-d}]^I\prod_l[k_F^{d+2-2l}]^{V_l}=k_F^{-d+(d+2)(L-I+\sum_l V_l)}\tau^{L(d+1)}=k_F^2\tau^{L(d+1)}. k F − d [ k F Q d + 1 ] L [ k F 2 − d ] I l ∏ [ k F d + 2 − 2 l ] V l = k F − d + ( d + 2 ) ( L − I + ∑ l V l ) τ L ( d + 1 ) = k F 2 τ L ( d + 1 ) .
With τ = q T F ∝ r s 1 / ( d − 1 ) \tau=q_{\mathrm{TF}}\propto r_s^{1/(d-1)} τ = q TF ∝ r s 1/ ( d − 1 ) and k F 2 ∝ r s − 2 k_F^2\propto r_s^{-2} k F 2 ∝ r s − 2 ,
k F 2 τ L ( d + 1 ) ∝ r s L ( d + 1 ) / ( d − 1 ) − 2 : r s 3 L − 2 ( d = 2 ) , r s 2 L − 2 ( d = 3 ) . k_F^2\tau^{L(d+1)}\propto r_s^{L(d+1)/(d-1)-2}:\qquad r_s^{3L-2}\ (d=2),\qquad r_s^{2L-2}\ (d=3). k F 2 τ L ( d + 1 ) ∝ r s L ( d + 1 ) / ( d − 1 ) − 2 : r s 3 L − 2 ( d = 2 ) , r s 2 L − 2 ( d = 3 ) .
A log r s \log r_s log r s term at these orders also requires a matching between soft and hard regions. Without the loop cancellation of Ref. [37 ] (main-text Sec. 3.1), each unsymmetrized three-point vertex would be larger by τ − 1 \tau^{-1} τ − 1 and each four-point vertex by τ − 2 \tau^{-2} τ − 2 . The two-loop diagrams with two three-point vertices or one four-point vertex would then scale as
k F 2 τ 2 ( d + 1 ) − 2 ∝ r s 2 / ( d − 1 ) : r s 2 ( d = 2 ) , r s ( d = 3 ) , k_F^2\tau^{2(d+1)-2}\propto r_s^{2/(d-1)}:\qquad r_s^2\ (d=2),\qquad r_s\ (d=3), k F 2 τ 2 ( d + 1 ) − 2 ∝ r s 2/ ( d − 1 ) : r s 2 ( d = 2 ) , r s ( d = 3 ) ,
the orders of a 2 ℓ a_{2\ell} a 2 ℓ and of the Carr–Maradudin coefficient. The bound of Refs. [37 , 38 ] excludes this enhancement. Hypotheses (H1) and (H2) of main-text Sec. 3.1 concern what this counting does not control: the uniformity of the bound, the degenerate regions (where the constant of the bound grows), the mixed regions with hard subdiagrams, the absence of log r s \log r_s log r s terms from the hard regions, and the regularity of the higher-order polarization at fixed z z z .
Proposition S1 (Completeness). *Under hypotheses (H1) and (H2) of main-text Sec. 3.1, the coefficient of r s 2 log r s r_s^2 \log r_s r s 2 log r s in the correlation energy of the paramagnetic two-dimensional electron gas is a 2 ℓ a_{2\ell} a 2 ℓ of main-text Eq. (39) .
Proof sketch. By (H2) and the one-line counting, the only term of order r s 2 log r s r_s^2 \log r_s r s 2 log r s with one soft line is the residue of § 16 . By the bound of Refs. [37 , 38 ] and (H1), the diagrams with L ≥ 2 L \ge2 L ≥ 2 soft loops are O ( r s 4 ∣ log r s ∣ p ) O(r_s^4 \lvert\log r_s\rvert^p) O ( r s 4 ∣ log r s ∣ p ) , and the hard regions give powers of r s r_s r s without log r s \log r_s log r s . The coefficient of r s 2 log r s r_s^2 \log r_s r s 2 log r s is therefore a 2 ℓ a_{2\ell} a 2 ℓ .
Screened internal exchange line. Replacing the bare exchange line in (S6) by the Thomas–Fermi form 2 π / ( ∣ p − k ∣ + κ s ) 2\pi/(\lvert\mathbf{p}-\mathbf{k}\rvert+\kappa_s) 2 π / (∣ p − k ∣ + κ s ) for both the vertex and the self-energy parts changes R 2 ( 1 ) R_2(1) R 2 ( 1 ) . For the screening wave numbers κ s = 10 − 1 , … , 10 − 5 \kappa_s=10^{-1},\ldots,10^{-5} κ s = 1 0 − 1 , … , 1 0 − 5 (in units of k F k_F k F ),
R 2 ( 1 ; κ s ) − R 2 ( 1 ) κ s = − 0.02951 , − 0.03180 , − 0.03212 , − 0.03216 , − 0.03217 , \frac{R_2(1;\kappa_s)-R_2(1)}{\kappa_s}=-0.02951,\ -0.03180,\ -0.03212,\ -0.03216,\ -0.03217, κ s R 2 ( 1 ; κ s ) − R 2 ( 1 ) = − 0.02951 , − 0.03180 , − 0.03212 , − 0.03216 , − 0.03217 ,
approaching the value − 0.0321699 -0.0321699 − 0.0321699 obtained from the integrand at κ s → 0 \kappa_s\to0 κ s → 0 . The change is thus O ( κ s ) O(\kappa_s) O ( κ s ) , without log κ s \log\kappa_s log κ s . The angles ϕ < κ s \phi<\kappa_s ϕ < κ s contribute O ( κ s 2 ) O(\kappa_s^2) O ( κ s 2 ) , and the remainder after the linear term is O ( κ s 2 log κ s ) O(\kappa_s^2\log\kappa_s) O ( κ s 2 log κ s ) , from κ s ≲ ϕ ≪ 1 \kappa_s\lesssim\phi\ll1 κ s ≲ ϕ ≪ 1 . With κ s ∝ r s \kappa_s\propto r_s κ s ∝ r s , the screening of the internal line shifts the Π ( 1 ) \Pi^{(1)} Π ( 1 ) series by O ( r s 3 log r s ) O(r_s^3\log r_s) O ( r s 3 log r s ) .
Three dimensions. The one-line terms Π ( 1 ) \Pi^{(1)} Π ( 1 ) and g 3 g_3 g 3 alone reproduce the complete coefficient C C C of main-text Eq. (1) (§ 14 ), and the counting places the two-loop diagrams at r s 2 L − 2 = r s 2 r_s^{2L-2}=r_s^2 r s 2 L − 2 = r s 2 .
Spin polarization in two dimensions
For spin polarization ζ \zeta ζ in two dimensions, with r s r_s r s referring to the total density,
k F σ = k F x σ , x ± = ( 1 ± ζ ) 1 / 2 . k_{F\sigma}=k_F x_\sigma,\qquad x_\pm=(1\pm\zeta)^{1/2}. k F σ = k F x σ , x ± = ( 1 ± ζ ) 1/2 .
The density of states per spin is independent of k F σ k_{F\sigma} k F σ in two dimensions. Hence the free response of spin σ \sigma σ is h 2 ( z / x σ ) h_2(z/x_\sigma) h 2 ( z / x σ ) times its paramagnetic value per spin, and the corresponding first-order exchange polarization, with coupling e 2 / ( ℏ v F σ ) e^2/(\hbar v_{F\sigma}) e 2 / ( ℏ v F σ ) , is R 2 ( z / x σ ) / x σ R_2(z/x_\sigma)/x_\sigma R 2 ( z / x σ ) / x σ . The coefficients then scale as
a ℓ ( ζ ) a ℓ ( 0 ) = ∫ [ ∑ σ h 2 ( z / x σ ) ] 3 d z 8 ∫ h 2 3 d z , a 2 ℓ ( ζ ) a 2 ℓ ( 0 ) = ∫ [ ∑ σ R 2 ( z / x σ ) / x σ ] [ ∑ σ h 2 ( z / x σ ) ] 2 d z 8 ∫ R 2 h 2 2 d z . (74) \frac{a_\ell(\zeta)}{a_\ell(0)}=\frac{\int[\sum_\sigma h_2(z/x_\sigma)]^3\,\mathrm{d}z}{8\int h_2^3\,\mathrm{d}z},\qquad
\frac{a_{2\ell}(\zeta)}{a_{2\ell}(0)}=\frac{\int[\sum_\sigma R_2(z/x_\sigma)/x_\sigma][\sum_\sigma h_2(z/x_\sigma)]^2\,\mathrm{d}z}{8\int R_2h_2^2\,\mathrm{d}z}. \tag*{(74)} a ℓ ( 0 ) a ℓ ( ζ ) = 8 ∫ h 2 3 d z ∫ [ ∑ σ h 2 ( z / x σ ) ] 3 d z , a 2 ℓ ( 0 ) a 2 ℓ ( ζ ) = 8 ∫ R 2 h 2 2 d z ∫ [ ∑ σ R 2 ( z / x σ ) / x σ ] [ ∑ σ h 2 ( z / x σ ) ] 2 d z . (74)
Checks. For ζ = 1 \zeta=1 ζ = 1 only x + = 2 x_+=\sqrt{2} x + = 2 remains, and the substitution z → 2 z z\to\sqrt{2}z z → 2 z gives
a ℓ ( 1 ) a ℓ ( 0 ) = 2 ∫ h 2 3 d z 8 ∫ h 2 3 d z = 1 4 2 , a 2 ℓ ( 1 ) a 2 ℓ ( 0 ) = ∫ R 2 h 2 2 d z 8 ∫ R 2 h 2 2 d z = 1 8 . \frac{a_\ell(1)}{a_\ell(0)}=\frac{\sqrt{2}\int h_2^3\,\mathrm{d}z}{8\int h_2^3\,\mathrm{d}z}=\frac{1}{4\sqrt{2}},\qquad
\frac{a_{2\ell}(1)}{a_{2\ell}(0)}=\frac{\int R_2h_2^2\,\mathrm{d}z}{8\int R_2h_2^2\,\mathrm{d}z}=\frac{1}{8}. a ℓ ( 0 ) a ℓ ( 1 ) = 8 ∫ h 2 3 d z 2 ∫ h 2 3 d z = 4 2 1 , a 2 ℓ ( 0 ) a 2 ℓ ( 1 ) = 8 ∫ R 2 h 2 2 d z ∫ R 2 h 2 2 d z = 8 1 .
The first is the known ratio [34 , 35 , 39 ] , which checks (S12). Table S5 lists intermediate values. At small ζ \zeta ζ ,
a 2 ℓ ( ζ ) a 2 ℓ ( 0 ) ≃ 1 − 0.0355 ζ 2 , a ℓ ( ζ ) a ℓ ( 0 ) ≃ 1 − 0.289 ζ 2 . \frac{a_{2\ell}(\zeta)}{a_{2\ell}(0)}\simeq1-0.0355\zeta^2,\qquad
\frac{a_\ell(\zeta)}{a_\ell(0)}\simeq1-0.289\zeta^2. a 2 ℓ ( 0 ) a 2 ℓ ( ζ ) ≃ 1 − 0.0355 ζ 2 , a ℓ ( 0 ) a ℓ ( ζ ) ≃ 1 − 0.289 ζ 2 .
Limit ζ → 1 \zeta\to1 ζ → 1 . The ratio for a ℓ a_\ell a ℓ tends to 1 / ( 4 2 ) 1/(4\sqrt{2}) 1/ ( 4 2 ) , with corrections of order ( 1 − ζ ) 1 / 2 (1-\zeta)^{1/2} ( 1 − ζ ) 1/2 . However, the ratio for a 2 ℓ a_{2\ell} a 2 ℓ does not tend to its value at ζ = 1 \zeta=1 ζ = 1 . In the region z ≲ x − z\lesssim x_- z ≲ x − , the minority response R 2 ( z / x − ) / x − R_2(z/x_-)/x_- R 2 ( z / x − ) / x − is of order 1 / x − 1/x_- 1/ x − on an interval of width x − x_- x − . With z = x − u z=x_-u z = x − u and h 2 ( 0 ) = 1 h_2(0)=1 h 2 ( 0 ) = 1 it leaves a finite contribution,
∫ R 2 ( z / x − ) x − [ ∑ σ h 2 ( z / x σ ) ] 2 d z = ∫ R 2 ( u ) [ h 2 ( x − u / x + ) + h 2 ( u ) ] 2 d u → ∫ R 2 ( u ) [ 1 + h 2 ( u ) ] 2 d u . \int\frac{R_2(z/x_-)}{x_-}\left[\sum_\sigma h_2(z/x_\sigma)\right]^2\,\mathrm{d}z
=\int R_2(u)[h_2(x_-u/x_+)+h_2(u)]^2\,\mathrm{d}u
\to\int R_2(u)[1+h_2(u)]^2\,\mathrm{d}u. ∫ x − R 2 ( z / x − ) [ σ ∑ h 2 ( z / x σ ) ] 2 d z = ∫ R 2 ( u ) [ h 2 ( x − u / x + ) + h 2 ( u ) ] 2 d u → ∫ R 2 ( u ) [ 1 + h 2 ( u ) ] 2 d u .
Table S5. Spin dependence of the coefficients of the two log r s \log r_s log r s terms of the two-dimensional gas, (S12).
Hence
a 2 ℓ ( ζ → 1 ) a 2 ℓ ( 0 ) = 1 8 + ∫ R 2 ( 1 + h 2 ) 2 d u 8 ∫ R 2 h 2 2 d u = 0.81649 , \frac{a_{2\ell}(\zeta\to1)}{a_{2\ell}(0)}=\frac{1}{8}+\frac{\int R_2(1+h_2)^2\,du}{8\int R_2h_2^2\,du}=0.81649, a 2 ℓ ( 0 ) a 2 ℓ ( ζ → 1 ) = 8 1 + 8 ∫ R 2 h 2 2 d u ∫ R 2 ( 1 + h 2 ) 2 d u = 0.81649 ,
a limit approached as 0.87373, 0.82392, 0.81727 and 0.81657 at 1 − ζ = 10 − 2 1-\zeta=10^{-2} 1 − ζ = 1 0 − 2 , 10 − 4 10^{-4} 1 0 − 4 , 10 − 6 10^{-6} 1 0 − 6 and 10 − 8 10^{-8} 1 0 − 8 .
Order of limits. The minority spins have the coupling e 2 / ( ℏ v F ↓ ) ∝ r s ( 1 − ζ ) − 1 / 2 e^2/(\hbar v_{F\downarrow})\propto r_s(1-\zeta)^{-1/2} e 2 / ( ℏ v F ↓ ) ∝ r s ( 1 − ζ ) − 1/2 . The expansion in r s r_s r s at fixed ζ \zeta ζ is therefore uniform only for r s ≪ ( 1 − ζ ) 1 / 2 r_s\ll(1-\zeta)^{1/2} r s ≪ ( 1 − ζ ) 1/2 , and the limits r s → 0 r_s\to0 r s → 0 and ζ → 1 \zeta\to1 ζ → 1 do not commute. In three dimensions, the minority density of states vanishes in proportion to k F ↓ k_{F\downarrow} k F ↓ , and the same counting gives a contribution of order k F ↓ / k F = ( 1 − ζ ) 1 / 3 k_{F\downarrow}/k_F=(1-\zeta)^{1/3} k F ↓ / k F = ( 1 − ζ ) 1/3 that vanishes as ζ → 1 \zeta\to1 ζ → 1 .
Definitions and first-order exchange polarization
This section derives the normalization of the kite functional of main-text Eq. (40) ,
K [ f ] = A 0 ∫ d 3 q ∫ 0 ∞ d ν π f ( q , i ν ) X ( q , ν ) , A 0 = 3 32 π 5 , \mathcal{K}[f]=A_0\int d^3q\int_0^\infty\frac{d\nu}{\pi}f(q,i\nu)X(q,\nu),\qquad A_0=\frac{3}{32\pi^5}, K [ f ] = A 0 ∫ d 3 q ∫ 0 ∞ π d ν f ( q , i ν ) X ( q , ν ) , A 0 = 32 π 5 3 ,
and defines the vertex part and the fixed-density self-energy part of the first-order exchange polarization.
Units and kite normalization
Conventions. For N N N electrons in a volume V \mathcal{V} V , the density n e n_e n e (not n n n , which counts the interactions), the Bohr radius, r s r_s r s , the coupling-scaled Hamiltonian and the Coulomb interaction are
n e = N / V , a B = ℏ 2 / ( m e e 2 ) , 4 π n e ( r s a B ) 3 / 3 = 1 , H ^ λ = T ^ + λ V ^ , v ( q ) = 4 π e 2 / q 2 . n_e=N/\mathcal{V},\qquad a_B=\hbar^2/(m_e e^2),\qquad4\pi n_e(r_sa_B)^3/3=1,\qquad\widehat{H}_{\lambda}=\widehat{T}+\lambda\widehat{V},\qquad v(q)=4\pi e^2/q^2. n e = N / V , a B = ℏ 2 / ( m e e 2 ) , 4 π n e ( r s a B ) 3 /3 = 1 , H λ = T + λ V , v ( q ) = 4 π e 2 / q 2 .
Relative to Fetter and Walecka [1 Chap. 3, Sec. 9] , the Green function is energy-normalized, with G F W = ℏ G G_{\mathrm{FW}}=\hbar G G FW = ℏ G and Σ F W = Σ / ℏ \Sigma_{\mathrm{FW}}=\Sigma/\hbar Σ FW = Σ/ℏ . The internal measure is d k ~ = d 3 k d ( ℏ k 0 ) d\tilde{k}=d^3k\,d(\hbar k^0) d k ~ = d 3 k d ( ℏ k 0 ) , with d k ~ / ( 2 π ) 4 d\tilde{k}/(2\pi)^4 d k ~ / ( 2 π ) 4 per independent integration. A diagram with n n n interactions and n L n_L n L fermion loops carries i n ( − 1 ) n L i^n(-1)^{n_L} i n ( − 1 ) n L .
Self-energy trace. Coupling integration gives the ground-state energy as a self-energy trace Eq. (5) of Ref. [2 ] , BRM below. BRM write the left side of that equation as ⟨ Ψ 0 ∣ V ^ ∣ Ψ 0 ⟩ \langle\Psi_0|\widehat{V}|\Psi_0\rangle ⟨ Ψ 0 ∣ V ∣ Ψ 0 ⟩ , but with the measure d λ / λ d\lambda/\lambda d λ / λ the right side is the coupling-integrated interaction energy,
E ( 1 ) − E ( 0 ) = ∫ 0 1 d λ ⟨ V ^ λ ⟩ λ = − i V 2 ∫ 0 1 d λ λ ∑ σ ∫ d k ~ ( 2 π ) 4 e i k 0 t 0 Σ σ λ ( k ) G σ λ ( k ) , (75) E(1)-E(0)=\int_0^1d\lambda\,\langle\widehat{V}_{\lambda}\rangle_\lambda=-\frac{i\mathcal{V}}{2}\int_0^1\frac{d\lambda}{\lambda}\sum_\sigma\int\frac{d\tilde{k}}{(2\pi)^4}e^{ik^0t_0}\Sigma^\lambda_\sigma(k)G^\lambda_\sigma(k), \tag*{(75)} E ( 1 ) − E ( 0 ) = ∫ 0 1 d λ ⟨ V λ ⟩ λ = − 2 i V ∫ 0 1 λ d λ σ ∑ ∫ ( 2 π ) 4 d k ~ e i k 0 t 0 Σ σ λ ( k ) G σ λ ( k ) , (75)
with E ( λ ) E(\lambda) E ( λ ) the ground-state energy of H ^ λ \widehat{H}_{\lambda} H λ , ⟨ ⋅ ⟩ λ \langle\cdot\rangle_\lambda ⟨ ⋅ ⟩ λ its ground-state expectation value, σ \sigma σ the spin index, t 0 → 0 + t_0\to0^+ t 0 → 0 + , k = ( k , k 0 ) k=(\mathbf{k},k^0) k = ( k , k 0 ) , Σ σ λ \Sigma^\lambda_\sigma Σ σ λ the proper self-energy and G σ λ G^\lambda_\sigma G σ λ the interacting propagator. With k F , σ k_{F,\sigma} k F , σ the Fermi wave number of spin σ \sigma σ , the free propagator is
G 0 , σ ( k , k 0 ) = Θ ( k − k F , σ ) ℏ k 0 − ϵ k + i 0 + Θ ( k F , σ − k ) ℏ k 0 − ϵ k − i 0 , ϵ k = ℏ 2 k 2 2 m e . G_{0,\sigma}(k,k^0)=\frac{\Theta(k-k_{F,\sigma})}{\hbar k^0-\epsilon_k+i0}+\frac{\Theta(k_{F,\sigma}-k)}{\hbar k^0-\epsilon_k-i0},\qquad\epsilon_k=\frac{\hbar^2k^2}{2m_e}. G 0 , σ ( k , k 0 ) = ℏ k 0 − ϵ k + i 0 Θ ( k − k F , σ ) + ℏ k 0 − ϵ k − i 0 Θ ( k F , σ − k ) , ϵ k = 2 m e ℏ 2 k 2 .
Kite. The proper second-order exchange self-energy with one line replaced by W λ W_\lambda W λ carries i 2 = − 1 i^2=-1 i 2 = − 1 . With the trace factor − i / 2 -i/2 − i /2 of (S13) the prefactor is ( − i / 2 ) i 2 = i / 2 (-i/2)i^2=i/2 ( − i /2 ) i 2 = i /2 , and the measure d λ / λ d\lambda/\lambda d λ / λ absorbs the λ \lambda λ of the bare line. With d q i = d 3 q i d ( ℏ q i 0 ) d q_i=d^3q_i\,d(\hbar q_i^0) d q i = d 3 q i d ( ℏ q i 0 ) , the kite contribution E k i t e E_{\mathrm{kite}} E kite to (S13) is
E k i t e = i V 2 ∫ 0 1 d λ ∫ d ℏ q 1 d q 1 d q 2 ( 2 π ) 12 e i k 0 t 0 v ( k − q 1 − q 2 ) W λ ( q 1 , q 1 0 ) G 4 ( k , q 1 , q 2 ) , (76) E_{\mathrm{kite}}=\frac{i\mathcal{V}}{2}\int_0^1d\lambda\int\frac{d\hbar q_1\,d q_1\,d q_2}{(2\pi)^{12}}e^{i k^0t_0}v(\mathbf{k}-\mathbf{q}_1-\mathbf{q}_2)W_\lambda(q_1,q_1^0)G_4(k,q_1,q_2), \tag*{(76)} E kite = 2 i V ∫ 0 1 d λ ∫ ( 2 π ) 12 d ℏ q 1 d q 1 d q 2 e i k 0 t 0 v ( k − q 1 − q 2 ) W λ ( q 1 , q 1 0 ) G 4 ( k , q 1 , q 2 ) , (76)
where the exchange loop is
G 4 = ∑ σ G 0 , σ ( q 1 + q 2 ) G 0 , σ ( q 2 ) G 0 , σ ( k − q 1 ) G 0 , σ ( k ) . G_4=\sum_\sigma G_{0,\sigma}(q_1+q_2)G_{0,\sigma}(q_2)G_{0,\sigma}(k-q_1)G_{0,\sigma}(k). G 4 = σ ∑ G 0 , σ ( q 1 + q 2 ) G 0 , σ ( q 2 ) G 0 , σ ( k − q 1 ) G 0 , σ ( k ) .
For a time-ordered line f f f whose causal continuation is analytic in quadrants I and III of the complex q 1 0 q_1^0 q 1 0 plane and bounded at large frequency, define
K [ f ] : = i V 4 N ∫ d ℏ q 1 d q 1 d q 2 ( 2 π ) 12 e i k 0 t 0 v ( k − q 1 − q 2 ) f ( q 1 , q 1 0 ) G 4 ( k , q 1 , q 2 ) , (77) \mathcal{K}[f]:=\frac{i\mathcal{V}}{4N}\int\frac{d\hbar q_1\,d q_1\,d q_2}{(2\pi)^{12}}e^{i k^0t_0}v(\mathbf{k}-\mathbf{q}_1-\mathbf{q}_2)f(q_1,q_1^0)G_4(k,q_1,q_2), \tag*{(77)} K [ f ] := 4 N i V ∫ ( 2 π ) 12 d ℏ q 1 d q 1 d q 2 e i k 0 t 0 v ( k − q 1 − q 2 ) f ( q 1 , q 1 0 ) G 4 ( k , q 1 , q 2 ) , (77)
e k i t e = E k i t e N = 2 ∫ 0 1 d λ K [ W λ ] = 2 K [ W ‾ ] , K [ v ] = e 2 x . (78) e_{\mathrm{kite}}=\frac{E_{\mathrm{kite}}}{N}=2\int_0^1d\lambda\,\mathcal{K}[W_\lambda]=2\mathcal{K}[\overline{W}],\qquad\mathcal{K}[v]=e_{2x}. \tag*{(78)} e kite = N E kite = 2 ∫ 0 1 d λ K [ W λ ] = 2 K [ W ] , K [ v ] = e 2 x . (78)
The first relation is main-text Eq. (43) . The factor 1 / 4 1/4 1/4 in (S15) is the bare closed-diagram coefficient 2 / ( 2 ! 2 2 ) 2/(2!2^2) 2/ ( 2 ! 2 2 ) .
Frequency integrals. The k 0 k^0 k 0 and q 2 0 q_2^0 q 2 0 integrations each give a factor
i Δ ℏ q 1 0 − δ ± i 0 , Δ = Δ p ( q ) = n p − n p + q , δ = δ p ( q ) = ε p + q − ε p , n p = Θ ( k F , σ − p ) , \frac{i\Delta}{\hbar q_1^0-\delta\pm i0},\qquad\Delta=\Delta_p(q)=n_p-n_{p+q},\qquad\delta=\delta_p(q)=\varepsilon_{p+q}-\varepsilon_p,\qquad n_p=\Theta(k_{F,\sigma}-p), ℏ q 1 0 − δ ± i 0 i Δ , Δ = Δ p ( q ) = n p − n p + q , δ = δ p ( q ) = ε p + q − ε p , n p = Θ ( k F , σ − p ) ,
where q = q 1 \mathbf{q}=\mathbf{q}_1 q = q 1 and p \mathbf{p} p stands for q 2 \mathbf{q}_2 q 2 or, after the shift k ↦ k + q 1 \mathbf{k}\mapsto\mathbf{k}+\mathbf{q}_1 k ↦ k + q 1 below, for k \mathbf{k} k . The poles in q 1 0 q_1^0 q 1 0 lie in quadrants IV (Δ = 1 \Delta=1 Δ = 1 ) and II (Δ = − 1 \Delta=-1 Δ = − 1 ). Hence the contour rotation q 1 0 = i ν q_1^0=i\nu q 1 0 = i ν crosses no singularity. The large arcs vanish by the O ( ∣ q 1 0 ∣ − 2 ) O(\lvert q_1^0\rvert^{-2}) O (∣ q 1 0 ∣ − 2 ) decay of the two denominators. Combining the two halves of the imaginary axis gives
∫ − ∞ ∞ d ν 2 π ⟶ ∫ 0 ∞ d ν π \int_{-\infty}^{\infty}\frac{d\nu}{2\pi}\longrightarrow\int_0^\infty\frac{d\nu}{\pi} ∫ − ∞ ∞ 2 π d ν ⟶ ∫ 0 ∞ π d ν
with the real part taken.
After the shift we obtain
K [ f ] = V 4 N ∑ σ ∫ d 3 q 1 ( 2 π ) 3 ∫ 0 ∞ d ( ℏ ν ) π f ( q 1 , i ν ) × ∫ d 3 k d 3 q 2 ( 2 π ) 6 v ( k − q 2 ) Re Δ q 2 ( q 1 ) Δ k ( q 1 ) [ i ℏ ν − δ q 2 ( q 1 ) ] [ i ℏ ν − δ k ( q 1 ) ] . (79) \mathcal{K}[f]=\frac{\mathcal{V}}{4N}\sum_\sigma\int\frac{d^3q_1}{(2\pi)^3}\int_0^\infty\frac{d(\hbar\nu)}{\pi}f(q_1,i\nu)\times\int\frac{d^3k\,d^3q_2}{(2\pi)^6}v(\mathbf{k}-\mathbf{q}_2)\operatorname{Re}\frac{\Delta_{q_2}(q_1)\Delta_k(q_1)}{[i\hbar\nu-\delta_{q_2}(q_1)][i\hbar\nu-\delta_k(q_1)]}. \tag*{(79)} K [ f ] = 4 N V σ ∑ ∫ ( 2 π ) 3 d 3 q 1 ∫ 0 ∞ π d ( ℏ ν ) f ( q 1 , i ν ) × ∫ ( 2 π ) 6 d 3 k d 3 q 2 v ( k − q 2 ) Re [ i ℏ ν − δ q 2 ( q 1 )] [ i ℏ ν − δ k ( q 1 )] Δ q 2 ( q 1 ) Δ k ( q 1 ) . (79)
where the prefactor is positive because the phases (prefactor of K [ f ] \mathcal{K}[f] K [ f ] , the k 0 k^0 k 0 and q 2 0 q_2^0 q 2 0 integrations, d q 1 0 = i d ν d q_1^0=i\,d\nu d q 1 0 = i d ν ) combine as i ⋅ i 2 ⋅ i = + 1 i\cdot i^2\cdot i=+1 i ⋅ i 2 ⋅ i = + 1 .
Prefactor. After rescaling to the units of main-text Eq. (4) (q 1 = q q_1=q q 1 = q , q 2 = p q_2=p q 2 = p ), the momentum measures supply k F 9 k_F^9 k F 9 , the two interactions v 0 2 v_0^2 v 0 2 and the frequency integral with two denominators E 0 − 1 E_0^{-1} E 0 − 1 . The prefactor A 0 A_0 A 0 of main-text Eq. (40) is
A 0 R y = 2 4 n e k F 9 ( 2 π ) 9 v 0 2 E 0 = 3 m e e 4 64 π 5 ℏ 2 = 3 32 π 5 R y = 2 H a → R y ( 1 2 ) d i a g r a m ( 1 2 ) s p i n ( 1 2 ) n o r m ( 4 π ) − 2 ( 2 π ) − 9 ( 3 π 2 ) 2 R y , (80) \begin{aligned}
A_0\,\mathrm{Ry}=\frac{2}{4n_e}\frac{k_F^9}{(2\pi)^9}\frac{v_0^2}{E_0}=\frac{3m_e e^4}{64\pi^5\hbar^2}=\frac{3}{32\pi^5}\,\mathrm{Ry} \\
&=2\mathrm{Ha}\to\mathrm{Ry}\left(\frac{1}{2}\right)_{\mathrm{diagram}}\left(\frac{1}{2}\right)_{\mathrm{spin}}\left(\frac{1}{2}\right)_{\mathrm{norm}}(4\pi)^{-2}(2\pi)^{-9}(3\pi^2)^2\,\mathrm{Ry}, \tag*{(80)}
\end{aligned} A 0 Ry = 4 n e 2 ( 2 π ) 9 k F 9 E 0 v 0 2 = 64 π 5 ℏ 2 3 m e e 4 = 32 π 5 3 Ry = 2 Ha → Ry ( 2 1 ) diagram ( 2 1 ) spin ( 2 1 ) norm ( 4 π ) − 2 ( 2 π ) − 9 ( 3 π 2 ) 2 Ry , (80)
using
1 n e = V N = 3 π 2 k F 3 , v 0 = 4 π e 2 k F 2 , E 0 = ℏ 2 k F 2 m e , m e e 4 ℏ 2 = 2 R y . \frac{1}{n_e}=\frac{\mathcal{V}}{N}=\frac{3\pi^2}{k_F^3},\qquad v_0=\frac{4\pi e^2}{k_F^2},\qquad E_0=\frac{\hbar^2k_F^2}{m_e},\qquad\frac{m_e e^4}{\hbar^2}=2\,\mathrm{Ry}. n e 1 = N V = k F 3 3 π 2 , v 0 = k F 2 4 π e 2 , E 0 = m e ℏ 2 k F 2 , ℏ 2 m e e 4 = 2 Ry .
The factors ( 1 2 ) d i a g r a m \left(\frac{1}{2}\right)_{\mathrm{diagram}} ( 2 1 ) diagram from (S13) and ( 1 2 ) n o r m \left(\frac{1}{2}\right)_{\mathrm{norm}} ( 2 1 ) norm from (S14)–(S15) give the bare coefficient 1 / 4 1/4 1/4 . All four propagators of the exchange loop carry the same spin, and the spin sum gives 2 s p i n 2_{\mathrm{spin}} 2 spin . With the dimensionless occupation difference Δ p ( q ) = Θ ( 1 − ∣ p ∣ ) − Θ ( 1 − ∣ p + q ∣ ) \Delta_p(q)=\Theta(1-|p|)-\Theta(1-|p+q|) Δ p ( q ) = Θ ( 1 − ∣ p ∣ ) − Θ ( 1 − ∣ p + q ∣ ) this gives main-text Eq. (40) . The known value K [ v ] = e 2 x \mathcal{K}[v]=e_{2x} K [ v ] = e 2 x checks the normalization without fitting (§ 30 ).
General diagram. The same steps give main-text Eq. (11) for any closed diagram D D D with one selected line f f f . The trace (S13) and the factor i n ( − 1 ) n L i^n(-1)^{n_L} i n ( − 1 ) n L give the closed diagram, and its coefficient 1 / S D 1/S_D 1/ S D replaces the bare kite coefficient 1 / 4 1/4 1/4 of (S15). Carrying out all internal frequency integrals except that of the selected line leaves f f f times the time-ordered insertion Π D ( q , q 0 ) \Pi_D(q,q^0) Π D ( q , q 0 ) . As for Δ = ± 1 \Delta=\pm1 Δ = ± 1 above, each of its time-ordered terms has q 0 q^0 q 0 -dependent denominators
ℏ q 0 − E I + i 0 or ℏ q 0 + E I − i 0 , \hbar q^0-E_I+i0 \quad\text{or} \quad\hbar q^0+E_I-i0, ℏ q 0 − E I + i 0 or ℏ q 0 + E I − i 0 ,
with excitation energies E I > 0 E_I>0 E I > 0 of the intermediate states. Its poles therefore lie in quadrants IV and II. The two halves of the imaginary axis combine to ∫ 0 ∞ d ν / π \int_0^\infty d\nu/\pi ∫ 0 ∞ d ν / π of the real part, because f f f is real and even in ν \nu ν on the imaginary axis and the real numerators give Π D ( q , − i ν ) = Π D ( q , i ν ) ∗ \Pi_D(q,-i\nu)=\Pi_D(q,i\nu)^* Π D ( q , − i ν ) = Π D ( q , i ν ) ∗ . For the kite and the self-energy pair, every term has at least two q 0 q^0 q 0 -dependent denominators [(S17) and (S21)], and therefore the large arcs vanish. For a general D D D we assume this decay. Comparing (S17) with (S20) shows that the closed diagram per electron is, in dimensional form,
− V N 1 S D ∫ d 3 q ( 2 π ) 3 ∫ 0 ∞ d ( ℏ ν ) π f Π D . -\frac{\mathcal{V}}{N}\frac{1}{S_D}\int\frac{d^3q}{(2\pi)^3}\int_0^\infty\frac{d(\hbar\nu)}{\pi}\,f\Pi_D. − N V S D 1 ∫ ( 2 π ) 3 d 3 q ∫ 0 ∞ π d ( ℏ ν ) f Π D .
In the units of Sec. S9.2, Π D \Pi_D Π D carries one factor C = π 2 q T F 2 C=\pi^2q_{\mathrm{TF}}^2 C = π 2 q TF 2 for each independent loop momentum of the insertion, that is, one for each bare interaction of D D D including the selected line. With V / N = 3 π 2 / k F 3 \mathcal{V}/N=3\pi^2/k_F^3 V / N = 3 π 2 / k F 3 and E 0 = 32 / ( π 2 q T F 4 ) E_0=32/(\pi^2q_{\mathrm{TF}}^4) E 0 = 32/ ( π 2 q TF 4 ) Ry, the prefactor becomes
− 3 π 2 k F 3 k F 3 ( 2 π ) 3 E 0 S D = − 3 E 0 8 π S D = − 12 S D π 3 q T F 4 R y . -\frac{3\pi^2}{k_F^3}\frac{k_F^3}{(2\pi)^3}\frac{E_0}{S_D}
=-\frac{3E_0}{8\pi S_D}
=-\frac{12}{S_D\pi^3q_{\mathrm{TF}}^4}\ \mathrm{Ry}. − k F 3 3 π 2 ( 2 π ) 3 k F 3 S D E 0 = − 8 π S D 3 E 0 = − S D π 3 q TF 4 12 Ry .
This gives Y D = − 12 Π D / ( S D π 3 q T F 4 ) Y_D=-12\Pi_D/(S_D\pi^3q_{\mathrm{TF}}^4) Y D = − 12 Π D / ( S D π 3 q TF 4 ) of main-text Sec. 2.3. For the kite, S D = 4 S_D=4 S D = 4 and Π D = Π v t x ( 1 ) \Pi_D=\Pi_{\mathrm{vtx}}^{(1)} Π D = Π vtx ( 1 ) of (S20) give main-text Eq. (40) ,
Y D = − 3 π 3 q T F 4 ( − q T F 4 32 π 2 X ) = 3 32 π 5 X = A 0 X . Y_D=-\frac{3}{\pi^3q_{\mathrm{TF}}^4}\left(-\frac{q_{\mathrm{TF}}^4}{32\pi^2}X\right)=\frac{3}{32\pi^5}X=A_0X. Y D = − π 3 q TF 4 3 ( − 32 π 2 q TF 4 X ) = 32 π 5 3 X = A 0 X .
Vertex and Fock self-energy corrections
This subsection supplies the responses that enter main-text Eqs. (52) and (59) . With (S20) below, the prefactor of main-text Eq. (52) gives K Π v t x ( 1 ) = K \mathcal{K}_{\Pi_{\mathrm{vtx}}^{(1)}}=\mathcal{K} K Π vtx ( 1 ) = K of main-text Eq. (40) . Responses are measured in v 0 − 1 v_0^{-1} v 0 − 1 , hence Π 0 = χ 0 = − q T F 2 L \Pi_0=\chi_0=-q_{\mathrm{TF}}^2\mathcal{L} Π 0 = χ 0 = − q TF 2 L . With the loop factor C C C of Sec. S1 and the spin factor 2, the free response is
Π 0 ( q , i ν ) = 2 C ∫ d 3 p ( 2 π ) 3 Δ p g p , g p = ( i ν − δ p ) − 1 . \Pi_0(q,i\nu)=2C\int\frac{d^3p}{(2\pi)^3}\Delta_p g_p,\qquad g_p=(i\nu-\delta_p)^{-1}. Π 0 ( q , i ν ) = 2 C ∫ ( 2 π ) 3 d 3 p Δ p g p , g p = ( i ν − δ p ) − 1 .
For the first-order exchange polarization [3 , 4 ] we use the dimensionless definitions, with the argument q q q of Δ p ( q ) \Delta_p(q) Δ p ( q ) and δ p ( q ) \delta_p(q) δ p ( q ) suppressed,
Δ p = Δ p ( q ) , δ p = p ⋅ q + q 2 2 , q T F 2 = a T F r s , a T F = 4 α L π , α L = ( 4 9 π ) 1 / 3 , Σ x ( k ) = − q T F 2 2 f L ( k ) , f L ( k ) = 1 2 + 1 − k 2 4 k log ∣ 1 + k 1 − k ∣ , C = π 2 q T F 2 , v ( q ) = q − 2 . (81) \begin{aligned}
\Delta_p&=\Delta_p(q),\qquad\delta_p=p\cdot q+\frac{q^2}{2},\\
q_{\mathrm{TF}}^2&=a_{\mathrm{TF}}r_s,\qquad a_{\mathrm{TF}}=\frac{4\alpha_L}{\pi},\qquad\alpha_L=\left(\frac{4}{9\pi}\right)^{1/3},\\
\Sigma_x(k)&=-\frac{q_{\mathrm{TF}}^2}{2}f_L(k),\qquad f_L(k)=\frac{1}{2}+\frac{1-k^2}{4k}\log\left|\frac{1+k}{1-k}\right|,\\
C&=\pi^2q_{\mathrm{TF}}^2,\qquad v(q)=q^{-2}. \tag*{(81)}
\end{aligned} Δ p q TF 2 Σ x ( k ) C = Δ p ( q ) , δ p = p ⋅ q + 2 q 2 , = a TF r s , a TF = π 4 α L , α L = ( 9 π 4 ) 1/3 , = − 2 q TF 2 f L ( k ) , f L ( k ) = 2 1 + 4 k 1 − k 2 log 1 − k 1 + k , = π 2 q TF 2 , v ( q ) = q − 2 . (81)
Here Σ x \Sigma_x Σ x is the Fock self-energy, and ∫ ∣ k ∣ < 1 ∣ p − k ∣ − 2 d 3 k = 4 π f L ( p ) \int_{|\mathbf{k}|<1}|\mathbf{p}-\mathbf{k}|^{-2}\,\mathrm{d}^3k=4\pi f_L(p) ∫ ∣ k ∣ < 1 ∣ p − k ∣ − 2 d 3 k = 4 π f L ( p ) gives the form above,
Σ x ( p ) = − ∫ ∣ k ∣ < 1 d 3 k ( 2 π ) 3 C v ( p − k ) = − 4 π C ( 2 π ) 3 f L ( p ) = − q T F 2 2 f L ( p ) . \Sigma_x(p)=-\int_{|\mathbf{k}|<1}\frac{\mathrm{d}^3k}{(2\pi)^3}\,C v(\mathbf{p}-\mathbf{k})=-\frac{4\pi C}{(2\pi)^3}f_L(p)=-\frac{q_{\mathrm{TF}}^2}{2}f_L(p). Σ x ( p ) = − ∫ ∣ k ∣ < 1 ( 2 π ) 3 d 3 k C v ( p − k ) = − ( 2 π ) 3 4 π C f L ( p ) = − 2 q TF 2 f L ( p ) .
The vertex part Π v t x ( 1 ) ( q , i ν ) \Pi_{\mathrm{vtx}}^{(1)}(q,i\nu) Π vtx ( 1 ) ( q , i ν ) of the first-order exchange polarization Π ( 1 ) \Pi^{(1)} Π ( 1 ) (main-text § 3.4 ) is, with X X X of main-text Eq. (41) ,
Π v t x ( 1 ) = − 2 C ∫ d 3 p d 3 k ( 2 π ) 6 C v ( p − k ) Δ p Δ k Re 1 ( i ν − δ p ) ( i ν − δ k ) = − q T F 4 32 π 2 X . (82) \Pi_{\mathrm{vtx}}^{(1)}=-2C\int\frac{\mathrm{d}^3p\,\mathrm{d}^3k}{(2\pi)^6}Cv(\mathbf{p}-\mathbf{k})\Delta_p\Delta_k\operatorname{Re}\frac{1}{(i\nu-\delta_p)(i\nu-\delta_k)}=-\frac{q_{\mathrm{TF}}^4}{32\pi^2}X. \tag*{(82)} Π vtx ( 1 ) = − 2 C ∫ ( 2 π ) 6 d 3 p d 3 k C v ( p − k ) Δ p Δ k Re ( i ν − δ p ) ( i ν − δ k ) 1 = − 32 π 2 q TF 4 X . (82)
Self-energy part. The Fock self-energy insertions shift δ p \delta_p δ p by Σ x ( p + q ) − Σ x ( p ) \Sigma_x(p+q)-\Sigma_x(p) Σ x ( p + q ) − Σ x ( p ) , which to first order gives
g p → g p + [ Σ x ( p + q ) − Σ x ( p ) ] g p 2 . g_p\to g_p+[\Sigma_x(p+q)-\Sigma_x(p)]g_p^2. g p → g p + [ Σ x ( p + q ) − Σ x ( p )] g p 2 .
The vertex and self-energy corrections use the same bare exchange line, δ Σ x ( p ) / δ n k = − C v ( p − k ) \delta\Sigma_x(p)/\delta n_k=-Cv(\mathbf{p}-\mathbf{k}) δ Σ x ( p ) / δ n k = − C v ( p − k ) . The fixed-density shift δ μ x = Σ x ( 1 ) = − q T F 2 / 4 \delta\mu_x=\Sigma_x(1)=-q_{\mathrm{TF}}^2/4 δ μ x = Σ x ( 1 ) = − q TF 2 /4 cancels the occupation-surface terms of the self-energy part Π S E ( 1 ) ( q , i ν ) \Pi_{\mathrm{SE}}^{(1)}(q,i\nu) Π SE ( 1 ) ( q , i ν ) of Π ( 1 ) \Pi^{(1)} Π ( 1 ) , leaving
Π S E ( 1 ) = 2 C ∫ d 3 p ( 2 π ) 3 Δ p Re Σ x ( p + q ) − Σ x ( p ) ( i ν − δ p ) 2 , (83) \Pi_{\mathrm{SE}}^{(1)}=2C\int\frac{\mathrm{d}^3p}{(2\pi)^3}\Delta_p\operatorname{Re}\frac{\Sigma_x(p+q)-\Sigma_x(p)}{(i\nu-\delta_p)^2}, \tag*{(83)} Π SE ( 1 ) = 2 C ∫ ( 2 π ) 3 d 3 p Δ p Re ( i ν − δ p ) 2 Σ x ( p + q ) − Σ x ( p ) , (83)
with
Σ x ( p + q ) − Σ x ( p ) = ∫ d 3 k ( 2 π ) 3 C v ( p − k ) Δ k . (84) \Sigma_x(p+q)-\Sigma_x(p)=\int\frac{\mathrm{d}^3k}{(2\pi)^3}Cv(\mathbf{p}-\mathbf{k})\Delta_k. \tag*{(84)} Σ x ( p + q ) − Σ x ( p ) = ∫ ( 2 π ) 3 d 3 k C v ( p − k ) Δ k . (84)
Sum. Symmetrized in p ↔ k \mathbf{p}\leftrightarrow\mathbf{k} p ↔ k , the two parts read
Π v t x ( 1 ) = C ∫ d 3 p d 3 k ( 2 π ) 6 C v ( p − k ) Δ p Δ k Re ( − 2 g p g k ) , Π S E ( 1 ) = C ∫ ⋯ Re ( g p 2 + g k 2 ) . \Pi_{\mathrm{vtx}}^{(1)}=C\int\frac{\mathrm{d}^3p\,\mathrm{d}^3k}{(2\pi)^6}Cv(\mathbf{p}-\mathbf{k})\Delta_p\Delta_k\operatorname{Re}(-2g_pg_k),\qquad\Pi_{\mathrm{SE}}^{(1)}=C\int\cdots\operatorname{Re}(g_p^2+g_k^2). Π vtx ( 1 ) = C ∫ ( 2 π ) 6 d 3 p d 3 k C v ( p − k ) Δ p Δ k Re ( − 2 g p g k ) , Π SE ( 1 ) = C ∫ ⋯ Re ( g p 2 + g k 2 ) .
Adding the vertex and self-energy parts gives
Π ( 1 ) ( q , i ν ) : = Π v t x ( 1 ) + Π S E ( 1 ) = C ∫ d 3 p d 3 k ( 2 π ) 6 C v ( p − k ) Δ p Δ k Re [ ( g p − g k ) 2 ] . (85) \Pi^{(1)}(q,i\nu):=\Pi_{\mathrm{vtx}}^{(1)}+\Pi_{\mathrm{SE}}^{(1)}=C\int\frac{\mathrm{d}^3p\,\mathrm{d}^3k}{(2\pi)^6}Cv(\mathbf{p}-\mathbf{k})\Delta_p\Delta_k\operatorname{Re}[(g_p-g_k)^2]. \tag*{(85)} Π ( 1 ) ( q , i ν ) := Π vtx ( 1 ) + Π SE ( 1 ) = C ∫ ( 2 π ) 6 d 3 p d 3 k C v ( p − k ) Δ p Δ k Re [( g p − g k ) 2 ] . (85)
This is Eq. (4) of Ref. [4 ] , symmetrized in p ↔ k \mathbf{p}\leftrightarrow\mathbf{k} p ↔ k and continued to imaginary frequency. The high-frequency ν − 2 \nu^{-2} ν − 2 terms of the two parts cancel (Sec. S10.2).
High-density expansion and cancellation by the Fock self-energy insertions
We derive the r s log 2 r s r_s\log^2r_s r s log 2 r s term of main-text Eq. (50) and its cancellation against the self-energy pair. The frequency-integrated shell bounds are in § 27 .
Fermi-surface origin and the integrated remainder
This subsection derives main-text Eq. (47) and, through Eqs. (91) –(94) , the coefficient c h d c_{\mathrm{hd}} c hd of main-text Eq. (50) . It also gives the small-q q q form of e Λ e_\Lambda e Λ stated in main-text § 3.2 . The three results are
X ( q , q z ) = a x ( z ) log 1 q + b x ( z ) + o ( 1 ) , e Λ ( q ) = κ ( Λ ) q log 1 q + O ( q ) , e k i t e − e 2 x ⊃ c h d r s log 2 r s , X(q,q_z)=a_x(z)\log\frac{1}{q}+b_x(z)+o(1),\qquad e_\Lambda(q)=\kappa(\Lambda)q\log\frac{1}{q}+O(q),\qquad e_{\mathrm{kite}}-e_{2x}\supset c_{\mathrm{hd}}\,r_s\log^2r_s, X ( q , q z ) = a x ( z ) log q 1 + b x ( z ) + o ( 1 ) , e Λ ( q ) = κ ( Λ ) q log q 1 + O ( q ) , e kite − e 2 x ⊃ c hd r s log 2 r s ,
with a x a_x a x , b x b_x b x as in main-text Eq. (47) . The shells at the Fermi surface are integrated first, at fixed z = ν / q > 0 z=\nu/q>0 z = ν / q > 0 .
Thin shells. As q → 0 q\to0 q → 0 at fixed z z z ,
δ p = q c p + O ( q 2 ) , c p = p ^ ⋅ q ^ , p ^ = p / p . \delta_p=qc_p+O(q^2),\qquad c_p=\hat{p}\cdot\hat{q},\qquad\hat{p}=p/p. δ p = q c p + O ( q 2 ) , c p = p ^ ⋅ q ^ , p ^ = p / p .
The two transition regions, where Δ p = ± 1 \Delta_p=\pm1 Δ p = ± 1 , are at small q q q the shells of width O ( q ) O(q) O ( q ) at the Fermi surface (main-text § 3.2 ). Integrating X X X of main-text Eq. (41) across them and over the directions d Ω p , d Ω k d\Omega_p,d\Omega_k d Ω p , d Ω k gives
X ( q , q z ) = ∫ d Ω p d Ω k c p c k R ( c p , c k ; z ) F p 1 ∣ p ^ − k ^ ∣ 2 + 2 π ∫ d Ω p c p 2 R ( c p , c p ; z ) [ log 1 q − log ∣ c p ∣ + 3 2 ] + o ( 1 ) , (86) \begin{aligned}
X(q,qz)={}&\int d\Omega_p\,d\Omega_k\,c_pc_k\,\mathcal{R}(c_p,c_k;z)\,\mathrm{F_p}\,\frac{1}{\lvert\hat{p}-\hat{k}\rvert^2}\\
&+2\pi\int d\Omega_p\,c_p^2\mathcal{R}(c_p,c_p;z)\left[\log\frac{1}{q}-\log\lvert c_p\rvert+\frac{3}{2}\right]\\
&+o(1), \tag*{(86)}
\end{aligned} X ( q , q z ) = ∫ d Ω p d Ω k c p c k R ( c p , c k ; z ) F p ∣ p ^ − k ^ ∣ 2 1 + 2 π ∫ d Ω p c p 2 R ( c p , c p ; z ) [ log q 1 − log ∣ c p ∣ + 2 3 ] + o ( 1 ) , (86)
where
R ( a , b ; z ) = a b − z 2 ( a b − z 2 ) 2 + z 2 ( a + b ) 2 , (87) \mathcal{R}(a,b;z)=\frac{ab-z^2}{(ab-z^2)^2+z^2(a+b)^2}, \tag*{(87)} R ( a , b ; z ) = ( ab − z 2 ) 2 + z 2 ( a + b ) 2 ab − z 2 , (87)
and Fp is the finite part. For an angular integrand Φ \Phi Φ ,
F p ∫ d Ω p d Ω k Φ ( p ^ , k ^ ) ∣ p ^ − k ^ ∣ 2 = lim ϵ → 0 [ ∫ ∣ p ^ − k ^ ∣ > ϵ d Ω p d Ω k Φ ( p ^ , k ^ ) ∣ p ^ − k ^ ∣ 2 − 2 π log 1 ϵ ∫ d Ω p Φ ( p ^ , p ^ ) ] , \mathrm{Fp}\int d\Omega_p\,d\Omega_k\,\frac{\Phi(\hat{p},\hat{k})}{\lvert\hat{p}-\hat{k}\rvert^2}
=\lim_{\epsilon\to0}\left[\int_{\lvert\hat{p}-\hat{k}\rvert>\epsilon}d\Omega_p\,d\Omega_k\,\frac{\Phi(\hat{p},\hat{k})}{\lvert\hat{p}-\hat{k}\rvert^2}-2\pi\log\frac{1}{\epsilon}\int d\Omega_p\,\Phi(\hat{p},\hat{p})\right], Fp ∫ d Ω p d Ω k ∣ p ^ − k ^ ∣ 2 Φ ( p ^ , k ^ ) = ϵ → 0 lim [ ∫ ∣ p ^ − k ^ ∣ > ϵ d Ω p d Ω k ∣ p ^ − k ^ ∣ 2 Φ ( p ^ , k ^ ) − 2 π log ϵ 1 ∫ d Ω p Φ ( p ^ , p ^ ) ] ,
with Φ = c p c k R ( c p , c k ; z ) \Phi=c_pc_k\mathcal{R}(c_p,c_k;z) Φ = c p c k R ( c p , c k ; z ) here and Φ = c p c k Λ 2 J ( Λ ; c p , c k ) \Phi=c_pc_k\Lambda^2J(\Lambda;c_p,c_k) Φ = c p c k Λ 2 J ( Λ ; c p , c k ) in (S27) below. The subtracted term is 2 π log ( 1 / ϵ ) 2\pi\log(1/\epsilon) 2 π log ( 1/ ϵ ) times the diagonal integral.
Derivation. To derive (S24), write p = ( 1 + q x ) p ^ p=(1+qx)\hat{p} p = ( 1 + q x ) p ^ and k = ( 1 + q y ) k ^ k=(1+qy)\hat{k} k = ( 1 + q y ) k ^ . At fixed z z z the occupation difference restricts x x x to an interval of length ∣ c p ∣ \lvert c_p\rvert ∣ c p ∣ . Each shell therefore contributes the radial measure q ∣ c p ∣ q\lvert c_p\rvert q ∣ c p ∣ and the electron–hole factor ( i q z − q c p ) − 1 (iqz-qc_p)^{-1} ( i q z − q c p ) − 1 . The factors q q q cancel, and with the signs Δ p = sgn c p \Delta_p=\operatorname{sgn}c_p Δ p = sgn c p of the transition regions the real part gives
Δ p Δ k ( q ∣ c p ∣ ) ( q ∣ c k ∣ ) Re 1 ( i q z − q c p ) ( i q z − q c k ) = c p c k R ( c p , c k ; z ) . \Delta_p\Delta_k(q\lvert c_p\rvert)(q\lvert c_k\rvert)\operatorname{Re}\frac{1}{(iqz-qc_p)(iqz-qc_k)}
=c_pc_k\mathcal{R}(c_p,c_k;z). Δ p Δ k ( q ∣ c p ∣) ( q ∣ c k ∣) Re ( i q z − q c p ) ( i q z − q c k ) 1 = c p c k R ( c p , c k ; z ) .
For p ^ ≠ k ^ \hat{p}\ne\hat{k} p ^ = k ^ the Coulomb factor tends to ∣ p ^ − k ^ ∣ − 2 \lvert\hat{p}-\hat{k}\rvert^{-2} ∣ p ^ − k ^ ∣ − 2 , which gives the first line. Near p ^ = k ^ \hat{p}=\hat{k} p ^ = k ^ , the separation of the two momenta has the tangential component d d d and the normal component q ( x − y ) q(x-y) q ( x − y ) . With x = ∣ c p ∣ u x=\lvert c_p\rvert u x = ∣ c p ∣ u , y = ∣ c p ∣ v y=\lvert c_p\rvert v y = ∣ c p ∣ v ,
∫ ∣ d ∣ < ϵ d 2 d d 2 + q 2 ( x − y ) 2 = 2 π log ϵ q ∣ x − y ∣ + o ( 1 ) , ∫ 0 1 ∫ 0 1 log 1 q ∣ x − y ∣ d u d v = log 1 q − log ∣ c p ∣ + 3 2 , \int_{\lvert d\rvert<\epsilon}\frac{d^2d}{d^2+q^2(x-y)^2}=2\pi\log\frac{\epsilon}{q\lvert x-y\rvert}+o(1),\qquad
\int_0^1\int_0^1\log\frac{1}{q\lvert x-y\rvert}\,du\,dv=\log\frac{1}{q}-\log\lvert c_p\rvert+\frac{3}{2}, ∫ ∣ d ∣ < ϵ d 2 + q 2 ( x − y ) 2 d 2 d = 2 π log q ∣ x − y ∣ ϵ + o ( 1 ) , ∫ 0 1 ∫ 0 1 log q ∣ x − y ∣ 1 d u d v = log q 1 − log ∣ c p ∣ + 2 3 ,
since ∫ 0 1 ∫ 0 1 log ∣ u − v ∣ d u d v = − 3 / 2 \int_0^1\int_0^1\log\lvert u-v\rvert\,du\,dv=-3/2 ∫ 0 1 ∫ 0 1 log ∣ u − v ∣ d u d v = − 3/2 . This average over the two shells gives the second line. Section S16 justifies these steps after the frequency integration.
Radial kernel. For fixed Λ \Lambda Λ , integrating the pointwise expansion (S24) term by term gives the radial kernel e Λ ( q ) e_\Lambda(q) e Λ ( q ) of main-text Eq. (46) in the form
e Λ ( q ) = κ ( Λ ) q log 1 q + β ( Λ ) q + o ( q ) , e_\Lambda(q)=\kappa(\Lambda)q\log\frac{1}{q}+\beta(\Lambda)q+o(q), e Λ ( q ) = κ ( Λ ) q log q 1 + β ( Λ ) q + o ( q ) ,
κ ( Λ ) = 3 2 π 2 Λ ∫ 0 1 c p 2 d c p ( c p + Λ ) 2 = 3 Λ 2 π 2 [ 1 − 2 Λ log ( 1 + 1 Λ ) + Λ 1 + Λ ] . (88) \kappa(\Lambda)=\frac{3}{2\pi^2}\Lambda\int_0^1\frac{c_p^2\,dc_p}{(c_p+\Lambda)^2}
=\frac{3\Lambda}{2\pi^2}\left[1-2\Lambda\log\left(1+\frac{1}{\Lambda}\right)+\frac{\Lambda}{1+\Lambda}\right]. \tag*{(88)} κ ( Λ ) = 2 π 2 3 Λ ∫ 0 1 ( c p + Λ ) 2 c p 2 d c p = 2 π 2 3Λ [ 1 − 2Λ log ( 1 + Λ 1 ) + 1 + Λ Λ ] . (88)
with
κ ( Λ ) ∼ 3 Λ 2 π 2 ( Λ → 0 ) , κ ( Λ ) ∼ 1 2 π 2 Λ ( Λ → ∞ ) . \kappa(\Lambda) \sim\frac{3\Lambda}{2\pi^2}\quad(\Lambda\to0),\qquad\kappa(\Lambda) \sim\frac{1}{2\pi^2\Lambda}\quad(\Lambda\to\infty). κ ( Λ ) ∼ 2 π 2 3Λ ( Λ → 0 ) , κ ( Λ ) ∼ 2 π 2 Λ 1 ( Λ → ∞ ) .
The coefficient κ \kappa κ follows from the diagonal line of (S24), with R ( c , c ; z ) = ( c 2 − z 2 ) / ( c 2 + z 2 ) 2 \mathcal{R}(c,c;z)=(c^2-z^2)/(c^2+z^2)^2 R ( c , c ; z ) = ( c 2 − z 2 ) / ( c 2 + z 2 ) 2 . The integrand is even in c p c_p c p , hence ∫ d Ω p = 4 π ∫ 0 1 d ∣ c p ∣ \int d\Omega_p=4\pi\int_0^1d|c_p| ∫ d Ω p = 4 π ∫ 0 1 d ∣ c p ∣ , and (S90) applies with c = ∣ c p ∣ c=|c_p| c = ∣ c p ∣ . With the prefactor 4 π A 0 4\pi A_0 4 π A 0 , the Lorentzian weight of main-text Eq. (46) and d ν = q d z d\nu=q\,dz d ν = q d z ,
κ = 4 π A 0 ⋅ 2 π ∫ d Ω p c p 2 ∫ 0 ∞ d z π Λ 2 ( c p 2 − z 2 ) ( Λ 2 + z 2 ) ( c p 2 + z 2 ) 2 = 16 π 3 A 0 Λ ∫ 0 1 c 2 d c ( c + Λ ) 2 , 16 π 3 A 0 = 3 2 π 2 . \kappa=4\pi A_0\cdot2\pi\int d\Omega_p\,c_p^2\int_0^\infty\frac{dz}{\pi}\frac{\Lambda^2(c_p^2-z^2)}{(\Lambda^2+z^2)(c_p^2+z^2)^2}=16\pi^3A_0\Lambda\int_0^1\frac{c^2\,dc}{(c+\Lambda)^2},\qquad16\pi^3A_0=\frac{3}{2\pi^2}. κ = 4 π A 0 ⋅ 2 π ∫ d Ω p c p 2 ∫ 0 ∞ π d z ( Λ 2 + z 2 ) ( c p 2 + z 2 ) 2 Λ 2 ( c p 2 − z 2 ) = 16 π 3 A 0 Λ ∫ 0 1 ( c + Λ ) 2 c 2 d c , 16 π 3 A 0 = 2 π 2 3 .
The finite part is
β ( Λ ) = 4 π A 0 ∫ d Ω p d Ω k c p c k Λ 2 J ( Λ ; c p , c k ) F p ∣ p ^ − k ^ ∣ − 2 + 8 π 2 A 0 ∫ d Ω p c p 2 Λ 2 J ( Λ ; c p , c p ) [ 3 2 − log ∣ c p ∣ ] , (89) \begin{aligned}
\beta(\Lambda)={}&4\pi A_0\int d\Omega_p d\Omega_k\,c_pc_k\Lambda^2J(\Lambda;c_p,c_k)\,F_p|\hat p-\hat k|^{-2}\\
&+8\pi^2A_0\int d\Omega_p\,c_p^2\Lambda^2J(\Lambda;c_p,c_p)\left[\frac{3}{2}-\log|c_p|\right], \tag*{(89)}
\end{aligned} β ( Λ ) = 4 π A 0 ∫ d Ω p d Ω k c p c k Λ 2 J ( Λ ; c p , c k ) F p ∣ p ^ − k ^ ∣ − 2 + 8 π 2 A 0 ∫ d Ω p c p 2 Λ 2 J ( Λ ; c p , c p ) [ 2 3 − log ∣ c p ∣ ] , (89)
with J J J from (S58). It follows by inserting (S24) into main-text Eq. (46) with ν = q z \nu=qz ν = q z and using
( a b − z 2 ) 2 + z 2 ( a + b ) 2 = ( a 2 + z 2 ) ( b 2 + z 2 ) , ∫ 0 ∞ d z π Λ 2 Λ 2 + z 2 R ( a , b ; z ) = Λ 2 J ( Λ ; a , b ) . (ab-z^2)^2+z^2(a+b)^2=(a^2+z^2)(b^2+z^2),\qquad\int_0^\infty\frac{dz}{\pi}\frac{\Lambda^2}{\Lambda^2+z^2}\mathcal{R}(a,b;z)=\Lambda^2J(\Lambda;a,b). ( ab − z 2 ) 2 + z 2 ( a + b ) 2 = ( a 2 + z 2 ) ( b 2 + z 2 ) , ∫ 0 ∞ π d z Λ 2 + z 2 Λ 2 R ( a , b ; z ) = Λ 2 J ( Λ ; a , b ) .
The first relation turns R \mathcal{R} R into the integrand of (S58), which gives the second. The bounds of § 27 control the integrated remainder of (S26) only to O ( q ) O(q) O ( q ) . For finite Λ \Lambda Λ , only the form e Λ ( q ) = κ ( Λ ) q log ( 1 / q ) + O ( q ) e_\Lambda(q)=\kappa(\Lambda)q\log(1/q)+O(q) e Λ ( q ) = κ ( Λ ) q log ( 1/ q ) + O ( q ) is therefore proved, and β ( Λ ) \beta(\Lambda) β ( Λ ) is the term-by-term coefficient. The constants in these bounds do not depend on Λ \Lambda Λ (§ 27 ). The same form therefore holds at Λ = ∞ \Lambda=\infty Λ = ∞ , where κ ( ∞ ) = 0 \kappa(\infty)=0 κ ( ∞ ) = 0 . The main text uses only the form with O ( q ) O(q) O ( q ) .
Window. The coupling average of the screened line, main-text Eqs. (6) and (19) , is
2 W ‾ = 2 ∫ 0 1 λ v 1 + λ c d λ = v ρ ( c ) , ρ ( c ) = 2 [ c − log ( 1 + c ) ] c 2 , c = q T F 2 L ( q , ν ) q 2 , (90) 2\overline{W}=2\int_0^1\frac{\lambda v}{1+\lambda c}\,d\lambda=v\rho(c),\qquad\rho(c)=\frac{2[c-\log(1+c)]}{c^2},\qquad c=\frac{q_{\mathrm{TF}}^2\mathcal{L}(q,\nu)}{q^2}, \tag*{(90)} 2 W = 2 ∫ 0 1 1 + λ c λ v d λ = v ρ ( c ) , ρ ( c ) = c 2 2 [ c − log ( 1 + c )] , c = q 2 q TF 2 L ( q , ν ) , (90)
L ( 0 , 0 ) = 1 , ρ ( c ) − 1 = − 2 3 c + O ( c 2 ) . \mathcal{L}(0,0)=1,\qquad\rho(c)-1=-\frac{2}{3}c+O(c^2). L ( 0 , 0 ) = 1 , ρ ( c ) − 1 = − 3 2 c + O ( c 2 ) .
In the window q T F ≪ q ≪ 1 q_{\mathrm{TF}}\ll q\ll1 q TF ≪ q ≪ 1 , the screening ratio c c c is small. By main-text Eqs. (42) and (43) , and with ρ − 1 ≃ − 2 3 c \rho-1\simeq-\frac{2}{3}c ρ − 1 ≃ − 3 2 c in main-text Eq. (40) ,
e k i t e − e 2 x = K [ v ( ρ − 1 ) ] , v ( ρ − 1 ) ≃ − 2 3 q T F 2 L ( q , ν ) q 4 , d 3 q = 4 π q 2 d q , e_{\mathrm{kite}}-e_{2x}=\mathcal{K}[v(\rho-1)],\qquad v(\rho-1)\simeq-\frac{2}{3}\frac{q_{\mathrm{TF}}^2\mathcal{L}(q,\nu)}{q^4},\qquad d^3q=4\pi q^2\,dq, e kite − e 2 x = K [ v ( ρ − 1 )] , v ( ρ − 1 ) ≃ − 3 2 q 4 q TF 2 L ( q , ν ) , d 3 q = 4 π q 2 d q ,
and therefore
e k i t e − e 2 x = − 8 π A 0 q T F 2 3 ∫ q T F 1 d q q 2 ∫ 0 ∞ d ν π L ( q , ν ) X ( q , ν ) + O ( q T F 2 log ( 1 / q T F 2 ) ) , (91) e_{\mathrm{kite}}-e_{2x}=-\frac{8\pi A_0q_{\mathrm{TF}}^2}{3}\int_{q_{\mathrm{TF}}}^1\frac{dq}{q^2}\int_0^\infty\frac{d\nu}{\pi}\mathcal{L}(q,\nu)X(q,\nu)+O\left(q_{\mathrm{TF}}^2\log(1/q_{\mathrm{TF}}^2)\right), \tag*{(91)} e kite − e 2 x = − 3 8 π A 0 q TF 2 ∫ q TF 1 q 2 d q ∫ 0 ∞ π d ν L ( q , ν ) X ( q , ν ) + O ( q TF 2 log ( 1/ q TF 2 ) ) , (91)
where the neglected regions are bounded as follows. For q > 1 q>1 q > 1 the bound uses (S57), and for q < q T F q<q_{\mathrm{TF}} q < q TF it uses § 27 and Eq. (S57),
q > 1 : ∣ ρ − 1 ∣ ≤ 2 c , ∫ 0 ∞ ∣ X ∣ d ν = O ( q − 2 ) ⇒ O ( q T F 2 ) , q>1:\quad|\rho-1|\leq\frac{2}{c},\qquad\int_0^\infty|X|\,d\nu=O(q^{-2})\ \Rightarrow\ O(q_{\mathrm{TF}}^2), q > 1 : ∣ ρ − 1∣ ≤ c 2 , ∫ 0 ∞ ∣ X ∣ d ν = O ( q − 2 ) ⇒ O ( q TF 2 ) ,
q < q T F : ∣ ρ − 1 ∣ ≤ 1 , ∫ ∣ X ( q , q z ) ∣ d z = O ( log ( e / q ) ) ⇒ O ( q T F 2 log ( 1 / q T F 2 ) ) . q<q_{\mathrm{TF}}:\quad|\rho-1|\leq1,\qquad\int|X(q,qz)|\,dz=O(\log(e/q))\ \Rightarrow\ O\left(q_{\mathrm{TF}}^2\log(1/q_{\mathrm{TF}}^2)\right). q < q TF : ∣ ρ − 1∣ ≤ 1 , ∫ ∣ X ( q , q z ) ∣ d z = O ( log ( e / q )) ⇒ O ( q TF 2 log ( 1/ q TF 2 ) ) .
By the bound ∫ ∣ X ( q , q z ) ∣ d z = O ( log ( e / q ) ) \int|X(q,qz)|\,dz=O(\log(e/q)) ∫ ∣ X ( q , q z ) ∣ d z = O ( log ( e / q )) , the O ( c 2 ) O(c^2) O ( c 2 ) error of ρ − 1 ≃ − 2 3 c \rho-1\simeq-\frac{2}{3}c ρ − 1 ≃ − 3 2 c inside the window also contributes O ( q T F 2 log ( 1 / q T F 2 ) ) O(q_{\mathrm{TF}}^2\log(1/q_{\mathrm{TF}}^2)) O ( q TF 2 log ( 1/ q TF 2 )) .
Coefficient a x a_x a x . Within the window, X ( q , q z ) X(q,qz) X ( q , q z ) has the expansion of main-text Eq. (47) . The second line of (S24), d Ω p = 2 π d c p \mathrm{d}\Omega_p=2\pi\,\mathrm{d}c_p d Ω p = 2 π d c p and (S34) give
a x ( z ) = 2 π ∫ d Ω p c p 2 c p 2 − z 2 ( c p 2 + z 2 ) 2 = 8 π 2 [ 1 − 2 z arctan 1 z + z 2 1 + z 2 ] = 8 π 2 [ h ( z ) + z h ′ ( z ) ] , (92) a_x(z)=2\pi\int\mathrm{d}\Omega_p\,c_p^2\frac{c_p^2-z^2}{(c_p^2+z^2)^2}=8\pi^2\left[1-2z\arctan\frac{1}{z}+\frac{z^2}{1+z^2}\right]=8\pi^2\left[h(z)+zh'(z)\right], \tag*{(92)} a x ( z ) = 2 π ∫ d Ω p c p 2 ( c p 2 + z 2 ) 2 c p 2 − z 2 = 8 π 2 [ 1 − 2 z arctan z 1 + 1 + z 2 z 2 ] = 8 π 2 [ h ( z ) + z h ′ ( z ) ] , (92)
and therefore ∫ 0 ∞ h ( z ) X ( q , q z ) d z / π = A log ( 1 / q ) + O ( 1 ) \int_0^\infty h(z)X(q,qz)\,\mathrm{d}z/\pi=\mathcal{A}\log(1/q)+O(1) ∫ 0 ∞ h ( z ) X ( q , q z ) d z / π = A log ( 1/ q ) + O ( 1 ) by (S86). The replacement of h h h by L ( q , q z ) \mathcal{L}(q,qz) L ( q , q z ) costs only O ( q log ( e / q ) ) O(q\log(e/q)) O ( q log ( e / q )) by (S88). The integral of main-text Eq. (47) is
A = ∫ 0 ∞ d z π h ( z ) a x ( z ) = 8 π ∫ 0 ∞ [ h 2 + 1 2 z ( h 2 ) ′ ] d z = 4 π ∫ 0 ∞ h ( z ) 2 d z = 4 π 2 3 ( 1 − log 2 ) , (93) \mathcal{A}=\int_0^\infty\frac{\mathrm{d}z}{\pi}\,h(z)a_x(z)=8\pi\int_0^\infty\left[h^2+\frac{1}{2}z(h^2)'\right]\mathrm{d}z=4\pi\int_0^\infty h(z)^2\mathrm{d}z=\frac{4\pi^2}{3}(1-\log2), \tag*{(93)} A = ∫ 0 ∞ π d z h ( z ) a x ( z ) = 8 π ∫ 0 ∞ [ h 2 + 2 1 z ( h 2 ) ′ ] d z = 4 π ∫ 0 ∞ h ( z ) 2 d z = 3 4 π 2 ( 1 − log 2 ) , (93)
where the third equality integrates 1 2 z ( h 2 ) ′ \frac{1}{2}z(h^2)' 2 1 z ( h 2 ) ′ by parts, since z h 2 → 0 zh^2\to0 z h 2 → 0 at both ends (h ≃ 1 / ( 3 z 2 ) h\simeq1/(3z^2) h ≃ 1/ ( 3 z 2 ) at large z z z ). The last equality uses ∫ 0 ∞ h ( z ) 2 d z = π ( 1 − log 2 ) / 3 \int_0^\infty h(z)^2\mathrm{d}z=\pi(1-\log2)/3 ∫ 0 ∞ h ( z ) 2 d z = π ( 1 − log 2 ) /3 [5 ] . With h ( z ) = ∫ 0 1 a 2 d a / ( a 2 + z 2 ) h(z)=\int_0^1a^2\mathrm{d}a/(a^2+z^2) h ( z ) = ∫ 0 1 a 2 d a / ( a 2 + z 2 ) and ∫ 0 ∞ d z / [ ( z 2 + a 2 ) ( z 2 + b 2 ) ] = π / [ 2 a b ( a + b ) ] \int_0^\infty\mathrm{d}z/[(z^2+a^2)(z^2+b^2)]=\pi/[2ab(a+b)] ∫ 0 ∞ d z / [( z 2 + a 2 ) ( z 2 + b 2 )] = π / [ 2 ab ( a + b )] ,
∫ 0 ∞ h 2 d z = π 2 ∫ 0 1 ∫ 0 1 a b a + b d a d b = π 2 ⋅ 2 3 ( 1 − log 2 ) . \int_0^\infty h^2\mathrm{d}z=\frac{\pi}{2}\int_0^1\int_0^1\frac{ab}{a+b}\,\mathrm{d}a\,\mathrm{d}b=\frac{\pi}{2}\cdot\frac{2}{3}(1-\log2). ∫ 0 ∞ h 2 d z = 2 π ∫ 0 1 ∫ 0 1 a + b ab d a d b = 2 π ⋅ 3 2 ( 1 − log 2 ) .
Coefficient c h d c_{\mathrm{hd}} c hd . Insert ∫ 0 ∞ L X d ν / π = q [ A log ( 1 / q ) + O ( 1 ) ] \int_0^\infty\mathcal{L}X\,\mathrm{d}\nu/\pi=q[\mathcal{A}\log(1/q)+O(1)] ∫ 0 ∞ L X d ν / π = q [ A log ( 1/ q ) + O ( 1 )] into (S29), with q T F 2 = a T F r s q_{\mathrm{TF}}^2=a_{\mathrm{TF}}r_s q TF 2 = a TF r s and log 2 ( q T F 2 ) = log 2 r s + O ( ∣ log r s ∣ ) \log^2(q_{\mathrm{TF}}^2)=\log^2r_s+O(\lvert\log r_s\rvert) log 2 ( q TF 2 ) = log 2 r s + O (∣ log r s ∣) . The log 2 \log^2 log 2 term is
e k i t e − e 2 x ⊃ − 8 π A 0 q T F 2 3 A ∫ q T F 1 log ( 1 / q ) q d q ⊃ c h d r s log 2 r s , e_{\mathrm{kite}}-e_{2x}\supset-\frac{8\pi A_0q_{\mathrm{TF}}^2}{3}\mathcal{A}\int_{q_{\mathrm{TF}}}^1\frac{\log(1/q)}{q}\,\mathrm{d}q\supset c_{\mathrm{hd}}r_s\log^2r_s, e kite − e 2 x ⊃ − 3 8 π A 0 q TF 2 A ∫ q TF 1 q log ( 1/ q ) d q ⊃ c hd r s log 2 r s ,
and main-text Eq. (50) follows from
∫ q T F 1 log ( 1 / q ) q d q = 1 8 log 2 ( q T F 2 ) , c h d = − π A 0 a T F 3 A = − α L ( 1 − log 2 ) 6 π 3 . (94) \int_{q_{\mathrm{TF}}}^1\frac{\log(1/q)}{q}\,\mathrm{d}q=\frac{1}{8}\log^2(q_{\mathrm{TF}}^2),\qquad c_{\mathrm{hd}}=-\frac{\pi A_0a_{\mathrm{TF}}}{3}\mathcal{A}=-\frac{\alpha_L(1-\log2)}{6\pi^3}. \tag*{(94)} ∫ q TF 1 q log ( 1/ q ) d q = 8 1 log 2 ( q TF 2 ) , c hd = − 3 π A 0 a TF A = − 6 π 3 α L ( 1 − log 2 ) . (94)
Cancellation by the Fock self-energy insertions
This subsection derives the self-energy factor log ( 1 / q ) \log(1/q) log ( 1/ q ) of the kernel in main-text Eq. (59) , its cancellation after the frequency integration, the tail condition of Lemma S1, and the high-density result for e k i t e + e S E e_{\mathrm{kite}}+e_{\mathrm{SE}} e kite + e SE of main-text § 3.4 ,
e k i t e + e S E = e 2 x + O ( r s ∣ log r s ∣ ) . e_{\mathrm{kite}}+e_{\mathrm{SE}}=e_{2x}+O(r_s\lvert\log r_s\rvert). e kite + e SE = e 2 x + O ( r s ∣ log r s ∣) .
Shell expansion of Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) . The cusp of the Fock self-energy Σ x = − q T F 2 f L / 2 \Sigma_x=-q_{\mathrm{TF}}^2f_L/2 Σ x = − q TF 2 f L /2 at the Fermi surface is
f L ( 1 + u ) − f L ( 1 ) = − u 2 log 2 ∣ u ∣ + O ( u 2 ∣ log ∣ u ∣ ∣ ) , p = 1 + q η s h , ∣ p + q ∣ − 1 = q ( η s h + c p ) + O ( q 2 ) , f_L(1+u)-f_L(1)=-\frac{u}{2}\log\frac{2}{\lvert u\rvert}+O(u^2\lvert\log\lvert u\rvert\rvert),\qquad p=1+q\eta_{\mathrm{sh}},\qquad\lvert p+q\rvert-1=q(\eta_{\mathrm{sh}}+c_p)+O(q^2), f L ( 1 + u ) − f L ( 1 ) = − 2 u log ∣ u ∣ 2 + O ( u 2 ∣ log ∣ u ∣∣) , p = 1 + q η sh , ∣ p + q ∣ − 1 = q ( η sh + c p ) + O ( q 2 ) ,
which gives
Σ x ( p + q ) − Σ x ( p ) = q T F 2 q 4 [ c p log 1 q + H s h ( η s h , c p ) ] + O ( q T F 2 q 2 ∣ log q ∣ ) , (95) \Sigma_x(p+q)-\Sigma_x(p)=\frac{q_{\mathrm{TF}}^2q}{4}\left[c_p\log\frac{1}{q}+\mathcal{H}_{\mathrm{sh}}(\eta_{\mathrm{sh}},c_p)\right]+O(q_{\mathrm{TF}}^2q^2\lvert\log q\rvert), \tag*{(95)} Σ x ( p + q ) − Σ x ( p ) = 4 q TF 2 q [ c p log q 1 + H sh ( η sh , c p ) ] + O ( q TF 2 q 2 ∣ log q ∣) , (95)
where H s h \mathcal{H}_{\mathrm{sh}} H sh denotes the part without log ( 1 / q ) \log(1/q) log ( 1/ q ) and is bounded on the shell. Integration across the signed shell supplies the factor q c p qc_p q c p , and
∫ − 1 1 c p 2 ( c p 2 − z 2 ) ( c p 2 + z 2 ) 2 d c p = 2 [ h ( z ) + z h ′ ( z ) ] , z = ν / q . (96) \int_{-1}^{1}\frac{c_p^2(c_p^2-z^2)}{(c_p^2+z^2)^2}\,\mathrm{d}c_p=2[h(z)+zh'(z)],\qquad z=\nu/q. \tag*{(96)} ∫ − 1 1 ( c p 2 + z 2 ) 2 c p 2 ( c p 2 − z 2 ) d c p = 2 [ h ( z ) + z h ′ ( z )] , z = ν / q . (96)
With 2 C / ( 2 π ) 3 = q T F 2 / ( 4 π ) 2\mathcal{C}/(2\pi)^3=q_{\mathrm{TF}}^2/(4\pi) 2 C / ( 2 π ) 3 = q TF 2 / ( 4 π ) , the shell factor q c p q c_p q c p , ( i ν − δ p ) − 2 = q − 2 ( i z − c p ) − 2 (i\nu-\delta_p)^{-2}=q^{-2}(iz-c_p)^{-2} ( i ν − δ p ) − 2 = q − 2 ( i z − c p ) − 2 and Re ( i z − c p ) − 2 = ( c p 2 − z 2 ) / ( c p 2 + z 2 ) 2 \operatorname{Re}(iz-c_p)^{-2}=(c_p^2-z^2)/(c_p^2+z^2)^2 Re ( i z − c p ) − 2 = ( c p 2 − z 2 ) / ( c p 2 + z 2 ) 2 , Eqs. (83) , (95) and (96) give
Π S E ( 1 ) ( q , i q z ) = q T F 2 4 π ∫ d Ω p q c p ⋅ q T F 2 q c p q log 1 q ⋅ c p 2 − z 2 q 2 ( c p 2 + z 2 ) 2 + O ( q T F 4 ) = 1 4 q T F 4 [ h + z h ′ ] log ( 1 / q ) + O ( q T F 4 ) . \begin{aligned}
\Pi_{\mathrm{SE}}^{(1)}(q,iqz)=\frac{q_{\mathrm{TF}}^2}{4\pi}\int d\Omega_p\,q c_p\cdot\frac{q_{\mathrm{TF}}^2 q c_p}{q}\log\frac{1}{q}\cdot\frac{c_p^2-z^2}{q^2(c_p^2+z^2)^2}+O(q_{\mathrm{TF}}^4) \\
&=\frac{1}{4}q_{\mathrm{TF}}^4[h+zh']\log(1/q)+O(q_{\mathrm{TF}}^4).
\end{aligned} Π SE ( 1 ) ( q , i q z ) = 4 π q TF 2 ∫ d Ω p q c p ⋅ q q TF 2 q c p log q 1 ⋅ q 2 ( c p 2 + z 2 ) 2 c p 2 − z 2 + O ( q TF 4 ) = 4 1 q TF 4 [ h + z h ′ ] log ( 1/ q ) + O ( q TF 4 ) .
The two parts therefore have terms in log ( 1 / q ) \log(1/q) log ( 1/ q ) of opposite sign, main-text Eq. (59) .
Frequency integration. The cancellation survives the frequency integration. For a = ∣ δ p ∣ / q a=|\delta_p|/q a = ∣ δ p ∣/ q ,
∫ 0 ∞ h ( z ) Re ( i z − a ) − 2 d z π = 1 2 [ log ( 1 + 1 / a ) − ( 1 + a ) − 1 ] , (97) \int_0^\infty h(z)\operatorname{Re}(iz-a)^{-2}\frac{dz}{\pi}=\frac{1}{2}\left[\log(1+1/a)-(1+a)^{-1}\right], \tag*{(97)} ∫ 0 ∞ h ( z ) Re ( i z − a ) − 2 π d z = 2 1 [ log ( 1 + 1/ a ) − ( 1 + a ) − 1 ] , (97)
which, multiplied by the shell factor a 2 a^2 a 2 , is the diagonal weight
M h ( a , a ) = ∫ 0 ∞ d z π h ( z ) a 2 R ( a , a ; z ) = ∫ 0 1 M Λ ( a , a ) d Λ , M_h(a,a)=\int_0^\infty\frac{dz}{\pi}h(z)a^2R(a,a;z)=\int_0^1M_\Lambda(a,a)\,d\Lambda, M h ( a , a ) = ∫ 0 ∞ π d z h ( z ) a 2 R ( a , a ; z ) = ∫ 0 1 M Λ ( a , a ) d Λ ,
with M Λ M_\Lambda M Λ of (S85) and M h M_h M h defined after it. This diagonal weight carries the vertex factor log ( 1 / q ) \log(1/q) log ( 1/ q ) in (S86). The two frequency-integrated terms in log ( 1 / q ) \log(1/q) log ( 1/ q ) therefore cancel direction by direction.
Bounded part. The part 1 4 q T F 2 q H s h \frac{1}{4}q_{\mathrm{TF}}^2q\mathcal{H}_{\mathrm{sh}} 4 1 q TF 2 q H sh of (S33) is O ( q T F 2 q a [ 1 + ∣ log a ∣ ] ) O(q_{\mathrm{TF}}^2qa[1+|\log a|]) O ( q TF 2 q a [ 1 + ∣ log a ∣ ]) on the shell. By (S35) with d ν = q d z d\nu=q\,dz d ν = q d z and ( i ν − δ ) − 2 = q − 2 ( i z − a ) − 2 (i\nu-\delta)^{-2}=q^{-2}(iz-a)^{-2} ( i ν − δ ) − 2 = q − 2 ( i z − a ) − 2 , and with the radial width q a qa q a and the factor 2 C = 2 π q T F 2 2\mathcal{C}=2\pi q_{\mathrm{TF}}^2 2 C = 2 π q TF 2 of (S21),
∫ 0 ∞ d ν π h Re ( i ν − δ ) − 2 = O ( q − 1 [ 1 + ∣ log a ∣ ] ) , \int_0^\infty\frac{d\nu}{\pi}h\operatorname{Re}(i\nu-\delta)^{-2}=O\left(q^{-1}[1+|\log a|]\right), ∫ 0 ∞ π d ν h Re ( i ν − δ ) − 2 = O ( q − 1 [ 1 + ∣ log a ∣ ] ) ,
2 C ⋅ q a ⋅ O ( q T F 2 q a [ 1 + ∣ log a ∣ ] ) ⋅ O ( q − 1 [ 1 + ∣ log a ∣ ] ) = O ( q T F 4 q a 2 [ 1 + ∣ log a ∣ ] 2 ) 2\mathcal{C}\cdot qa\cdot O\left(q_{\mathrm{TF}}^2qa[1+|\log a|]\right)\cdot O\left(q^{-1}[1+|\log a|]\right)=O\left(q_{\mathrm{TF}}^4qa^2[1+|\log a|]^2\right) 2 C ⋅ q a ⋅ O ( q TF 2 q a [ 1 + ∣ log a ∣ ] ) ⋅ O ( q − 1 [ 1 + ∣ log a ∣ ] ) = O ( q TF 4 q a 2 [ 1 + ∣ log a ∣ ] 2 )
in ∫ 0 ∞ h Π S E ( 1 ) d ν / π \int_0^\infty h\Pi_{\mathrm{SE}}^{(1)}\,d\nu/\pi ∫ 0 ∞ h Π SE ( 1 ) d ν / π , that is, O ( q a 2 [ 1 + ∣ log a ∣ ] 2 ) O(qa^2[1+|\log a|]^2) O ( q a 2 [ 1 + ∣ log a ∣ ] 2 ) in the units of the kite integrand. This bound is integrable over the grazing directions a → 0 a\to0 a → 0 , perpendicular to the momentum transfer, where the shell width vanishes.
High-frequency tails. At large ν \nu ν ,
Re 1 ( i ν − δ p ) ( i ν − δ k ) ≃ Re 1 ( i ν − δ p ) 2 ≃ 1 ν 2 . \operatorname{Re}\frac{1}{(i\nu-\delta_p)(i\nu-\delta_k)}\simeq\operatorname{Re}\frac{1}{(i\nu-\delta_p)^2}\simeq\frac{1}{\nu^2}. Re ( i ν − δ p ) ( i ν − δ k ) 1 ≃ Re ( i ν − δ p ) 2 1 ≃ ν 2 1 .
Then (S20) gives Π v t x ( 1 ) ≃ q T F 4 Z / ( 32 π 2 ν 2 ) \Pi_{\mathrm{vtx}}^{(1)}\simeq q_{\mathrm{TF}}^4Z/(32\pi^2\nu^2) Π vtx ( 1 ) ≃ q TF 4 Z / ( 32 π 2 ν 2 ) , with Z Z Z defined below, and (S21) with (S19) gives a tail proportional to ∫ d 3 p Δ p [ f L ( ∣ p + q ∣ ) − f L ( p ) ] \int d^3p\,\Delta_p[f_L(|p+q|)-f_L(p)] ∫ d 3 p Δ p [ f L ( ∣ p + q ∣ ) − f L ( p )] . With I B ( p ) = ∫ B ∣ p − k ∣ − 2 d 3 k = 4 π f L ( p ) I_{\mathrm{B}}(p)=\int_{\mathrm{B}}|p-k|^{-2}d^3k=4\pi f_L(p) I B ( p ) = ∫ B ∣ p − k ∣ − 2 d 3 k = 4 π f L ( p ) , the shift k → k + q k\to k+q k → k + q gives
I B ( ∣ p + q ∣ ) = ∫ ∣ k + q ∣ < 1 d 3 k ∣ p − k ∣ 2 , I B ( ∣ p + q ∣ ) − I B ( p ) = − ∫ Δ k d 3 k ∣ p − k ∣ 2 , I_{\mathrm{B}}(|p+q|)=\int_{|k+q|<1}\frac{d^3k}{|p-k|^2},\qquad I_{\mathrm{B}}(|p+q|)-I_{\mathrm{B}}(p)=-\int_{\Delta_k}\frac{d^3k}{|p-k|^2}, I B ( ∣ p + q ∣ ) = ∫ ∣ k + q ∣ < 1 ∣ p − k ∣ 2 d 3 k , I B ( ∣ p + q ∣ ) − I B ( p ) = − ∫ Δ k ∣ p − k ∣ 2 d 3 k ,
and therefore
∫ d 3 p Δ p [ f L ( ∣ p + q ∣ ) − f L ( p ) ] = − Z ( q ) 4 π , Z ( q ) = ∬ Δ p Δ k ∣ p − k ∣ 2 d 3 p d 3 k , (98) \int d^3p\,\Delta_p[f_L(|p+q|)-f_L(p)]=-\frac{Z(q)}{4\pi},\qquad Z(q)=\iint\frac{\Delta_p\Delta_k}{|p-k|^2}\,d^3p\,d^3k, \tag*{(98)} ∫ d 3 p Δ p [ f L ( ∣ p + q ∣ ) − f L ( p )] = − 4 π Z ( q ) , Z ( q ) = ∬ ∣ p − k ∣ 2 Δ p Δ k d 3 p d 3 k , (98)
which gives Π S E ( 1 ) ≃ − q T F 4 Z / ( 32 π 2 ν 2 ) \Pi_{\mathrm{SE}}^{(1)}\simeq-q_{\mathrm{TF}}^4Z/(32\pi^2\nu^2) Π SE ( 1 ) ≃ − q TF 4 Z / ( 32 π 2 ν 2 ) , and the ν − 2 \nu^{-2} ν − 2 terms cancel for every q q q . This is the tail condition of Lemma S1.
Energies. With K Π S E ( 1 ) \mathcal{K}_{\Pi_{\mathrm{SE}}^{(1)}} K Π SE ( 1 ) of main-text Eq. (52) ,
e k i t e = 2 ∫ 0 1 K [ W λ ] d λ , e S E = 2 ∫ 0 1 K Π S E ( 1 ) [ W λ ] d λ , W λ − λ v ≃ − λ 2 v c , ∫ 0 1 2 λ 2 d λ = 2 3 . e_{\mathrm{kite}}=2\int_0^1\mathcal{K}[W_\lambda]\,d\lambda,\qquad e_{\mathrm{SE}}=2\int_0^1\mathcal{K}_{\Pi_{\mathrm{SE}}^{(1)}}[W_\lambda]\,d\lambda,\qquad W_\lambda-\lambda v\simeq-\lambda^2v_c,\qquad\int_0^1 2\lambda^2\,d\lambda=\frac{2}{3}. e kite = 2 ∫ 0 1 K [ W λ ] d λ , e SE = 2 ∫ 0 1 K Π SE ( 1 ) [ W λ ] d λ , W λ − λ v ≃ − λ 2 v c , ∫ 0 1 2 λ 2 d λ = 3 2 .
The coupling weight 2, integrated against the factor λ 2 \lambda^{2} λ 2 of the first-order screening term, gives the same 2/3 for both energies. The coefficients of r s log 2 r s r_s \log^{2} r_s r s log 2 r s in e k i t e e_{\mathrm{kite}} e kite and in e S E e_{\mathrm{SE}} e SE are therefore c h d c_{\mathrm{hd}} c hd and − c h d -c_{\mathrm{hd}} − c hd . The r s log r s r_s \log r_s r s log r s terms require the finite shell parts [6 , 7 , 8 ] . For e k i t e e_{\mathrm{kite}} e kite the coefficient is c 1 = 0.01150 c_1 = 0.01150 c 1 = 0.01150 Ry [9 ] . For e S E e_{\mathrm{SE}} e SE and for the sum it is not evaluated here. The window estimate (S29) applies to e S E e_{\mathrm{SE}} e SE with the absolute bound (S69) in place of the shell bound of X X X , and the bounded part of (S33) is estimated above. Both r s log 2 r s r_s \log^{2} r_s r s log 2 r s coefficients therefore come with remainders O ( r s ∣ log r s ∣ ) O(r_s \lvert\log r_s\rvert) O ( r s ∣ log r s ∣) , and K Π S E ( 1 ) [ v ] = 0 K_{\Pi_{\mathrm{SE}}^{(1)}}[v] = 0 K Π SE ( 1 ) [ v ] = 0 [Eq. (146) , § 26 ] gives e k i t e + e S E = e 2 x + O ( r s ∣ log r s ∣ ) e_{\mathrm{kite}} + e_{\mathrm{SE}} = e_{2x} + O(r_s \lvert\log r_s\rvert) e kite + e SE = e 2 x + O ( r s ∣ log r s ∣) .
Relation to the vertex of the GWΓ \Gamma Γ scheme
The GWΓ \Gamma Γ scheme of Takada [10 , 11 ] constructs the density vertex as the ratio
Γ W I ( p , p + q ) = G ( p + q ) − 1 − G ( p ) − 1 G 0 ( p + q ) − 1 − G 0 ( p ) − 1 (99) \Gamma_{\mathrm{WI}}(p,p+q)=\frac{G(p+q)^{-1}-G(p)^{-1}}{G_0(p+q)^{-1}-G_0(p)^{-1}} \tag*{(99)} Γ WI ( p , p + q ) = G 0 ( p + q ) − 1 − G 0 ( p ) − 1 G ( p + q ) − 1 − G ( p ) − 1 (99)
at four-momenta p p p and p + q p+q p + q , with q = ( q , i ν ) q=(\mathbf{q},i\nu) q = ( q , i ν ) on the imaginary axis. The Ward identity constrains the density and current vertices together. Hence (S37) is consistent with it but is not the unique density vertex at finite q q q and ν \nu ν . With G 0 ( p + q ) − 1 − G 0 ( p ) − 1 = i ν − δ p G_0(p+q)^{-1}-G_0(p)^{-1}=i\nu-\delta_p G 0 ( p + q ) − 1 − G 0 ( p ) − 1 = i ν − δ p ,
G ( p ) G ( p + q ) Γ W I ( p , p + q ) = G ( p ) − G ( p + q ) i ν − δ p , G(p)G(p+q)\Gamma_{\mathrm{WI}}(p,p+q)=\frac{G(p)-G(p+q)}{i\nu-\delta_p}, G ( p ) G ( p + q ) Γ WI ( p , p + q ) = i ν − δ p G ( p ) − G ( p + q ) ,
so a polarization built with Γ W I \Gamma_{\mathrm{WI}} Γ WI depends on the self-energy only through the momentum distribution. The static Fock self-energy with the shift δ μ x \delta\mu_x δ μ x leaves this distribution unchanged. The first-order exchange correction built with Γ W I \Gamma_{\mathrm{WI}} Γ WI therefore vanishes identically. The first-order vertex of Γ W I \Gamma_{\mathrm{WI}} Γ WI and the split of the symmetrized exact first-order vertex are
− Δ Σ x ( p ) i ν − δ p ⟶ − Δ Σ x g p 2 , − 2 g p g k = − ( g p 2 + g k 2 ) + ( g p − g k ) 2 . -\frac{\Delta\Sigma_x(p)}{i\nu-\delta_p}\ \longrightarrow\ -\Delta\Sigma_x g_p^2,\qquad-2g_p g_k=-(g_p^2+g_k^2)+(g_p-g_k)^2. − i ν − δ p Δ Σ x ( p ) ⟶ − Δ Σ x g p 2 , − 2 g p g k = − ( g p 2 + g k 2 ) + ( g p − g k ) 2 .
The first cancels the self-energy insertion + Δ Σ x g p 2 +\Delta\Sigma_xg_p^2 + Δ Σ x g p 2 . After symmetrization this Ward part is − ( g p 2 + g k 2 ) -(g_p^2+g_k^2) − ( g p 2 + g k 2 ) in main-text Eq. (58) , and the remainder of the exact first-order vertex is ( g p − g k ) 2 (g_p-g_k)^2 ( g p − g k ) 2 . The terms in log ( 1 / q ) \log(1/q) log ( 1/ q ) of main-text Eq. (59) are the two halves of the vanishing part. The finite limit R ( 1 ) ( z ) R^{(1)}(z) R ( 1 ) ( z ) of − 16 Π ( 1 ) ( q , i q z ) / q T F 4 -16\Pi^{(1)}(q,iqz)/q_{\mathrm{TF}}^4 − 16 Π ( 1 ) ( q , i q z ) / q TF 4 as q → 0 q\to0 q → 0 [main-text Eq. (35) ] belongs to the remainder, which the G W Γ GW\Gamma G W Γ scheme represents by an electron–hole interaction I ˉ ( q ) \bar{I}(q) I ˉ ( q ) .
In the uniform gas, a Kohn–Sham reference with a local exchange–correlation potential shifts all orbital energies by a constant, whereas the Fock self-energy insertion Σ x ( k ) − δ μ x \Sigma_x(k)-\delta\mu_x Σ x ( k ) − δ μ x depends on k k k . Other SOSEX variants [12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 ] differ in the screened lines and the weights of the diagrams. Whether they retain the r s log 2 r s r_s\log^{2}r_s r s log 2 r s and r s − 1 / 4 r_s^{-1/4} r s − 1/4 terms depends on their frequency kernels. Singles (single-excitation) corrections [15 ] vanish on a Kohn–Sham reference by momentum conservation. Hartree–Fock orbital energies contain the insertions to all orders, together with the singularity ∝ u log ∣ u ∣ \propto u\log\lvert u\rvert ∝ u log ∣ u ∣ of the Hartree–Fock dispersion at the Fermi surface, where k = k F ( 1 + u ) k=k_{\mathrm{F}}(1+u) k = k F ( 1 + u ) . Yasuhara et al. [20 ] , whose exchange correction to the RPA starts from the Hartree–Fock approximation, observed a related cancellation in the electron liquid. In the contribution of first order in the dynamically screened interaction to the spin-parallel structure factor, two self-energy terms and one vertex term cancel at small momentum transfer. These authors argued that the self-energy part should be treated together with the other terms of the same order.
Correspondence with the BRM representation
This section transforms main-text Eqs. (40) and (43) into Eqs. (28) –(32) of Ref. [2 ] , starting from
e k i t e − e 2 x = 2 ∫ 0 1 d λ K [ W λ − λ v ] , K [ f ] = A 0 ∫ d 3 q ∫ 0 ∞ d ν π f ( q , i ν ) X ( q , ν ) , e_{\mathrm{kite}}-e_{2x}=2\int_{0}^{1}d\lambda\,\mathcal{K}[W_{\lambda}-\lambda v],\qquad\mathcal{K}[f]=A_{0}\int d^{3}q\int_{0}^{\infty}\frac{d\nu}{\pi}\,f(q,i\nu)X(q,\nu), e kite − e 2 x = 2 ∫ 0 1 d λ K [ W λ − λ v ] , K [ f ] = A 0 ∫ d 3 q ∫ 0 ∞ π d ν f ( q , i ν ) X ( q , ν ) ,
since 2 ∫ 0 1 K [ λ v ] d λ = K [ v ] = e 2 x 2\int_{0}^{1}\mathcal{K}[\lambda v]\,d\lambda=\mathcal{K}[v]=e_{2x} 2 ∫ 0 1 K [ λ v ] d λ = K [ v ] = e 2 x . The comparison fixes the normalization and the coupling weight in the BRM representation.
Sectors. The occupied-to-empty transitions form one of the two transition regions of § 21.1 ,
C q = { p : ∣ p ∣ ≤ 1 , ∣ p + q ∣ > 1 } , Δ p ( q ) = 1 , δ p ( q ) > 0. (100) C_{q}=\{\mathbf{p}:|\mathbf{p}|\leq1,\ |\mathbf{p}+\mathbf{q}|>1\},\qquad\Delta_{p}(q)=1,\qquad\delta_{p}(q)>0. \tag*{(100)} C q = { p : ∣ p ∣ ≤ 1 , ∣ p + q ∣ > 1 } , Δ p ( q ) = 1 , δ p ( q ) > 0. (100)
The reverse transitions map onto this region,
Δ = − 1 : p = − p ′ − q , p ′ ∈ C q , δ p ( q ) = − δ p ′ ( q ) . \Delta=-1:\quad\mathbf{p}=-\mathbf{p}^{\prime}-\mathbf{q},\qquad\mathbf{p}^{\prime}\in C_{q},\qquad\delta_{p}(q)=-\delta_{p^{\prime}}(q). Δ = − 1 : p = − p ′ − q , p ′ ∈ C q , δ p ( q ) = − δ p ′ ( q ) .
Reversing the signs of both excitation energies leaves the real part in main-text Eq. (41) unchanged. Hence, of the four sectors,
( Δ p , Δ k ) = ( 1 , 1 ) , ( − 1 , − 1 ) : equal , ( 1 , − 1 ) , ( − 1 , 1 ) : equal after exchanging p and k , (\Delta_{p},\Delta_{k})=(1,1),\ (-1,-1):\ \text{equal},\qquad(1,-1),\ (-1,1):\ \text{equal after exchanging }\mathbf{p}\text{ and }\mathbf{k}, ( Δ p , Δ k ) = ( 1 , 1 ) , ( − 1 , − 1 ) : equal , ( 1 , − 1 ) , ( − 1 , 1 ) : equal after exchanging p and k ,
and their sum is
X ( q , ν ) = 2 ∫ C q d 3 p d 3 k [ Re [ ( i ν − a ) ( i ν − b ) ] − 1 ∣ p − k ∣ 2 − Re [ ( i ν − a ) ( i ν + b ) ] − 1 ∣ p + k + q ∣ 2 ] , (101) X(q,\nu)=2\int_{C_{q}}d^{3}p\,d^{3}k\left[\frac{\operatorname{Re}[(i\nu-a)(i\nu-b)]^{-1}}{|\mathbf{p}-\mathbf{k}|^{2}}-\frac{\operatorname{Re}[(i\nu-a)(i\nu+b)]^{-1}}{|\mathbf{p}+\mathbf{k}+\mathbf{q}|^{2}}\right], \tag*{(101)} X ( q , ν ) = 2 ∫ C q d 3 p d 3 k [ ∣ p − k ∣ 2 Re [( i ν − a ) ( i ν − b ) ] − 1 − ∣ p + k + q ∣ 2 Re [( i ν − a ) ( i ν + b ) ] − 1 ] , (101)
a = δ p ( q ) , b = δ k ( q ) . a=\delta_{p}(q),\qquad b=\delta_{k}(q). a = δ p ( q ) , b = δ k ( q ) .
Here ∣ p + k + q ∣ 2 |\mathbf{p}+\mathbf{k}+\mathbf{q}|^{2} ∣ p + k + q ∣ 2 is the Coulomb factor ∣ p − k ∣ 2 |\mathbf{p}-\mathbf{k}|^{2} ∣ p − k ∣ 2 of the mixed sectors after the map is applied to k \mathbf{k} k .
Frequency integral. Partial fractions and Re ( i ν − a ) − 1 = − a / ( ν 2 + a 2 ) \operatorname{Re}(i\nu-a)^{-1}=-a/(\nu^{2}+a^{2}) Re ( i ν − a ) − 1 = − a / ( ν 2 + a 2 ) give sums of Lorentzians,
Re 1 ( i ν − a ) ( i ν ∓ b ) = 1 a ∓ b [ − a ν 2 + a 2 ∓ b ν 2 + b 2 ] , \operatorname{Re}\frac{1}{(i\nu-a)(i\nu\mp b)}=\frac{1}{a\mp b}\left[-\frac{a}{\nu^{2}+a^{2}}\mp\frac{b}{\nu^{2}+b^{2}}\right], Re ( i ν − a ) ( i ν ∓ b ) 1 = a ∓ b 1 [ − ν 2 + a 2 a ∓ ν 2 + b 2 b ] ,
with a = b a=b a = b as a continuous limit. For a > 0 a>0 a > 0 , ν = a tan u \nu=a\tan u ν = a tan u absorbs each Lorentzian into the measure (first identity, for any ϕ \phi ϕ for which the integral converges). For a line f ( q , i ν ) f(q,i\nu) f ( q , i ν ) as in main-text Eq. (40) , the second relation defines H ( a ) H(a) H ( a ) at fixed q q q :
∫ 0 ∞ a ϕ ( ν ) ν 2 + a 2 d ν = ∫ 0 π / 2 ϕ ( a tan u ) d u , H ( a ) = 1 π ∫ 0 π / 2 f ( q , i a tan u ) d u . (102) \int_{0}^{\infty}\frac{a\phi(\nu)}{\nu^{2}+a^{2}}\,d\nu=\int_{0}^{\pi/2}\phi(a\tan u)\,du,\qquad H(a)=\frac{1}{\pi}\int_{0}^{\pi/2}f(q,ia\tan u)\,du. \tag*{(102)} ∫ 0 ∞ ν 2 + a 2 a ϕ ( ν ) d ν = ∫ 0 π /2 ϕ ( a tan u ) d u , H ( a ) = π 1 ∫ 0 π /2 f ( q , ia tan u ) d u . (102)
Applied to Eq. (101) , these relations give
∫ 0 ∞ d ν π f ( q , i ν ) X ( q , ν ) = 2 ∫ C q d 3 p d 3 k [ H ( b ) − H ( a ) ( a − b ) ∣ p − k ∣ 2 + H ( a ) + H ( b ) ( a + b ) ∣ p + k + q ∣ 2 ] . (103) \int_{0}^{\infty}\frac{d\nu}{\pi}\,f(q,i\nu)X(q,\nu)=2\int_{C_{q}}d^{3}p\,d^{3}k\left[\frac{H(b)-H(a)}{(a-b)|\mathbf{p}-\mathbf{k}|^{2}}+\frac{H(a)+H(b)}{(a+b)|\mathbf{p}+\mathbf{k}+\mathbf{q}|^{2}}\right]. \tag*{(103)} ∫ 0 ∞ π d ν f ( q , i ν ) X ( q , ν ) = 2 ∫ C q d 3 p d 3 k [ ( a − b ) ∣ p − k ∣ 2 H ( b ) − H ( a ) + ( a + b ) ∣ p + k + q ∣ 2 H ( a ) + H ( b ) ] . (103)
Spin scaling. For spin polarization ζ = ( n ↑ − n ↓ ) / n e \zeta=(n_{\uparrow}-n_{\downarrow})/n_{e} ζ = ( n ↑ − n ↓ ) / n e , the species σ = ± 1 \sigma=\pm1 σ = ± 1 have Fermi momenta x σ k F x_{\sigma}k_{F} x σ k F with x σ = ( 1 + σ ζ ) 1 / 3 x_{\sigma}=(1+\sigma\zeta)^{1/3} x σ = ( 1 + σ ζ ) 1/3 . For ∣ ζ ∣ < 1 |\zeta|<1 ∣ ζ ∣ < 1 , rescaling the momenta by x σ x_{\sigma} x σ and the frequency by x σ 2 x_{\sigma}^{2} x σ 2 maps each spin contribution to a unit Fermi ball, with a factor x σ 3 x_{\sigma}^{3} x σ 3 from the measures and denominators. The fully spin-polarized case follows by continuity. In BRM notation, α = α L \alpha=\alpha_{L} α = α L and g ( q , ν ) = 2 L ( q , ν ) g(q,\nu)=2L(q,\nu) g ( q , ν ) = 2 L ( q , ν ) . Below, q q q and ν \nu ν are the rescaled variables for spin σ \sigma σ , and σ ′ \sigma' σ ′ runs over the screening responses of both species. Species σ ′ \sigma' σ ′ , with density of states at the Fermi surface proportional to x σ ′ x_{\sigma'} x σ ′ , contributes to Π 0 \Pi_0 Π 0
− 1 2 q T F 2 x σ ′ L ( q / x σ ′ , ν / x σ ′ 2 ) (unscaled) ⟶ − 1 2 q T F 2 x σ ′ L ( θ q , θ 2 ν ) (rescaled) . -\frac{1}{2}q_{\mathrm{TF}}^2x_{\sigma'}\mathcal{L}\left(q/x_{\sigma'},\nu/x_{\sigma'}^2\right)\ \text{(unscaled)}\ \longrightarrow\ -\frac{1}{2}q_{\mathrm{TF}}^2x_{\sigma'}\mathcal{L}\left(\theta q,\theta^2\nu\right)\ \text{(rescaled)}. − 2 1 q TF 2 x σ ′ L ( q / x σ ′ , ν / x σ ′ 2 ) (unscaled) ⟶ − 2 1 q TF 2 x σ ′ L ( θ q , θ 2 ν ) (rescaled) .
In the rescaled variables,
ϵ λ = 1 + λ α L r s S σ π x σ 2 q 2 , S σ = ∑ σ ′ x σ ′ g ( θ q , θ 2 ν ) , θ = x σ / x σ ′ . (104) \epsilon_\lambda=1+\frac{\lambda\alpha_Lr_sS_\sigma}{\pi x_\sigma^2q^2},\qquad S_\sigma=\sum_{\sigma'}x_{\sigma'}g\left(\theta q,\theta^2\nu\right),\qquad\theta=x_\sigma/x_{\sigma'}. \tag*{(104)} ϵ λ = 1 + π x σ 2 q 2 λ α L r s S σ , S σ = σ ′ ∑ x σ ′ g ( θ q , θ 2 ν ) , θ = x σ / x σ ′ . (104)
Here θ \theta θ and S σ S_\sigma S σ are local to this section. With 1 / ϵ λ − 1 = − ( ϵ λ − 1 ) / ϵ λ 1/\epsilon_\lambda-1=-(\epsilon_\lambda-1)/\epsilon_\lambda 1/ ϵ λ − 1 = − ( ϵ λ − 1 ) / ϵ λ , the screening correction to the selected line, with its spin scaling factor, is
λ x σ 3 1 q 2 ( 1 ϵ λ − 1 ) = − λ 2 α L r s π q 4 f λ σ ( q , ν ) , f λ σ = x σ 3 S σ x σ 2 + λ α L r s S σ / ( π q 2 ) . (105) \lambda x_\sigma^3\frac{1}{q^2}\left(\frac{1}{\epsilon_\lambda}-1\right)=-\frac{\lambda^2\alpha_Lr_s}{\pi q^4}f_\lambda^\sigma(q,\nu),\qquad f_\lambda^\sigma=\frac{x_\sigma^3S_\sigma}{x_\sigma^2+\lambda\alpha_Lr_sS_\sigma/(\pi q^2)}. \tag*{(105)} λ x σ 3 q 2 1 ( ϵ λ 1 − 1 ) = − π q 4 λ 2 α L r s f λ σ ( q , ν ) , f λ σ = x σ 2 + λ α L r s S σ / ( π q 2 ) x σ 3 S σ . (105)
Energy. For e k i t e − e 2 x e_{\mathrm{kite}}-e_{2x} e kite − e 2 x the line in Eq. (103) is W λ − λ v W_\lambda-\lambda v W λ − λ v . Each spin contribution has normalization A 0 / 2 A_0/2 A 0 /2 , the BRM energy uses the coupling integration 2 d λ 2\,d\lambda 2 d λ , and the two coupling factors of W λ − λ v W_\lambda-\lambda v W λ − λ v give the factor λ 2 \lambda^2 λ 2 :
e k i t e − e 2 x = 3 α L r s 16 π 7 ∑ σ ∫ 0 1 d λ λ 2 ∫ d 3 q q 4 ∫ C q 2 d 3 p d 3 k ∫ 0 π / 2 d u × [ f λ σ ( q , a tan u ) − f λ σ ( q , b tan u ) ( a − b ) ∣ p − k ∣ 2 − f λ σ ( q , a tan u ) + f λ σ ( q , b tan u ) ( a + b ) ∣ p + k + q ∣ 2 ] . (106) \begin{aligned}
e_{\mathrm{kite}}-e_{2x}=\frac{3\alpha_Lr_s}{16\pi^7}\sum_\sigma\int_0^1d\lambda\,\lambda^2\int\frac{d^3q}{q^4}\int_{C_q^2}d^3p\,d^3k\int_0^{\pi/2}du \\
&\times\left[\frac{f_\lambda^\sigma(q,a\tan u)-f_\lambda^\sigma(q,b\tan u)}{(a-b)\lvert p-k\rvert^2}-\frac{f_\lambda^\sigma(q,a\tan u)+f_\lambda^\sigma(q,b\tan u)}{(a+b)\lvert p+k+q\rvert^2}\right]. \tag*{(106)}
\end{aligned} e kite − e 2 x = 16 π 7 3 α L r s σ ∑ ∫ 0 1 d λ λ 2 ∫ q 4 d 3 q ∫ C q 2 d 3 p d 3 k ∫ 0 π /2 d u × [ ( a − b ) ∣ p − k ∣ 2 f λ σ ( q , a tan u ) − f λ σ ( q , b tan u ) − ( a + b ) ∣ p + k + q ∣ 2 f λ σ ( q , a tan u ) + f λ σ ( q , b tan u ) ] . (106)
The coupling, spin, sector, H and line factors give the prefactor,
2 ⋅ A 0 2 ⋅ 2 ⋅ 1 π ⋅ α L r s π = 2 A 0 α L r s π 2 = 3 α L r s 16 π 7 , A 0 = 3 32 π 5 , 2\cdot\frac{A_0}{2}\cdot2\cdot\frac{1}{\pi}\cdot\frac{\alpha_Lr_s}{\pi}=\frac{2A_0\alpha_Lr_s}{\pi^2}=\frac{3\alpha_Lr_s}{16\pi^7},\qquad A_0=\frac{3}{32\pi^5}, 2 ⋅ 2 A 0 ⋅ 2 ⋅ π 1 ⋅ π α L r s = π 2 2 A 0 α L r s = 16 π 7 3 α L r s , A 0 = 32 π 5 3 ,
and the minus sign of the line reverses both terms in the bracket. The relabeling ( q , p , k ) ↦ ( q 1 , q 2 , k ) (q,p,k)\mapsto(q_1,q_2,k) ( q , p , k ) ↦ ( q 1 , q 2 , k ) turns our a a a and b b b into the excitation energies a ( q 1 , q 2 ) a(q_1,q_2) a ( q 1 , q 2 ) and a ( q 1 , k ) a(q_1,k) a ( q 1 , k ) of BRM and completes the correspondence.
Diagram coefficients, coupling weights and energy expressions
This section derives the closed-diagram coefficients of the chain family of main-text § 3.2 , the weights assigned by the coupling average and the relations between the energy expressions.
Closed-diagram coefficients of the chain family
The closed-diagram coefficient of a completed diagram is its Wick-contraction coefficient in the connected closed-diagram expansion with fixed G 0 G_0 G 0 [1 Chap. 3] . The chain family contains the bare kite and every diagram with a nonempty chain of free bubbles on one interaction, the other interaction staying bare (Figure S1 ). Each unlabeled diagram appears once.
Figure S1. Closed-diagram coefficients in the connected closed-diagram expansion. (a) Bare exchange kite, coefficient 1/4. (b,c) One and two polarization bubbles on one interaction line, coefficient 1/2 each. (d) One bubble on each line, coefficient 1/4, outside the chain family e c l o s e d e_{\mathrm{closed}} e closed . Solid arrows are G 0 G_0 G 0 and wavy lines are bare interactions v v v . A crossing without a dot is not a vertex. The coefficients multiply complete diagrams, not individual bubbles, and refer to e c l o s e d e_{\mathrm{closed}} e closed , not to the BRM coupling average.
A vertex is an interaction end with one incoming and one outgoing fermion line. A symmetry is a bijection of the vertices preserving every directed fermion connection, every unordered interaction pair and any fixed physical label. It determines the map of every line.
Proposition S2 (Chain family). In the connected closed-diagram expansion with fixed G 0 G_0 G 0 , the bare kite has coefficient 1/4, whereas every diagram with a chain of m ≥ 1 m \ge1 m ≥ 1 polarization bubbles on one interaction has coefficient 1/2. With K [ f ] \mathcal{K}[f] K [ f ] retaining the bare coefficient 1/4, the energy per electron e c l o s e d e_{\mathrm{closed}} e closed of this family is
e c l o s e d = K [ v ] + 2 K [ W λ = 1 − v ] = 2 K [ W λ = 1 ] − K [ v ] . (107) e_{\mathrm{closed}} = \mathcal{K}[v] + 2\mathcal{K}[W_{\lambda=1} - v] = 2\mathcal{K}[W_{\lambda=1}] - \mathcal{K}[v]. \tag*{(107)} e closed = K [ v ] + 2 K [ W λ = 1 − v ] = 2 K [ W λ = 1 ] − K [ v ] . (107)
Proof. Bare kite. Number the core vertices along the fermion cycle 1 → 2 → 3 → 4 → 1 1 \to2 \to3 \to4 \to1 1 → 2 → 3 → 4 → 1 , with interactions { 1 , 3 } \{1,3\} { 1 , 3 } and { 2 , 4 } \{2,4\} { 2 , 4 } . The image of vertex 1 fixes the other three, and all four choices preserve the interaction pairs,
r k : i ↦ i + k ( m o d 4 ) , k = 0 , … , 3 , r 0 , r 2 : { 1 , 3 } ↦ { 1 , 3 } , r 1 , r 3 : { 1 , 3 } ↦ { 2 , 4 } . r_k : i \mapsto i + k \pmod4,\quad k = 0,\ldots,3,\qquad r_0,r_2 : \{1,3\} \mapsto\{1,3\},\qquad r_1,r_3 : \{1,3\} \mapsto\{2,4\}. r k : i ↦ i + k ( mod 4 ) , k = 0 , … , 3 , r 0 , r 2 : { 1 , 3 } ↦ { 1 , 3 } , r 1 , r 3 : { 1 , 3 } ↦ { 2 , 4 } .
The bare kite therefore has four symmetries and coefficient 1/4. In general, the 2 n n ! 2^n n! 2 n n ! relabelings of interaction labels and ends produce each distinct contraction as many times as the diagram has symmetries [21 Sec. 3.2] . With N s y m N_{\mathrm{sym}} N sym symmetries,
coefficient = # contractions 2 n n ! = 2 n n ! / N s y m 2 n n ! = 1 N s y m . \operatorname{coefficient} = \frac{\#\ \operatorname{contractions}}{2^n n!} = \frac{2^n n!/N_{\mathrm{sym}}}{2^n n!} = \frac{1}{N_{\mathrm{sym}}}. coefficient = 2 n n ! # contractions = 2 n n ! 2 n n ! / N sym = N sym 1 .
Chain. Insert a nonempty chain on { 1 , 3 } \{1,3\} { 1 , 3 } . The four-vertex cycle remains uniquely identifiable, since every added cycle has two vertices. A symmetry of the completed graph therefore induces a core symmetry preserving { 1 , 3 } \{1,3\} { 1 , 3 } , and only r 0 r_0 r 0 (the identity) and r 2 r_2 r 2 (1 ↔ 3 1 \leftrightarrow3 1 ↔ 3 , 2 ↔ 4 2 \leftrightarrow4 2 ↔ 4 ) qualify. Number the bubbles j = 1 , … , m j = 1,\ldots,m j = 1 , … , m along the chain from vertex 1, and let a j a_j a j (b j b_j b j ) be the vertex of bubble j j j whose interaction leads toward vertex 1 (vertex 3). Then
chain : { 1 , a 1 } , { b j , a j + 1 } ( 1 ≤ j < m ) , { b m , 3 } , bare : { 2 , 4 } , \operatorname{chain}: \{1,a_1\},\quad\{b_j,a_{j+1}\}\ (1 \le j < m),\quad\{b_m,3\},\qquad\operatorname{bare}: \{2,4\}, chain : { 1 , a 1 } , { b j , a j + 1 } ( 1 ≤ j < m ) , { b m , 3 } , bare : { 2 , 4 } ,
r 2 on the chain : a j ↦ b m + 1 − j , b j ↦ a m + 1 − j . r_2\text{ on the chain}: a_j \mapsto b_{m+1-j},\quad b_j \mapsto a_{m+1-j}. r 2 on the chain : a j ↦ b m + 1 − j , b j ↦ a m + 1 − j .
Each core symmetry extends uniquely to the chain. The second reverses the bubble order and exchanges the two vertices of each bubble. The completed graph therefore has two symmetries and coefficient 1/2 for every m ≥ 1 m \ge1 m ≥ 1 . Expanding W λ = 1 − v W_{\lambda=1}-v W λ = 1 − v into its nonempty chains and retaining the bare kite once gives (S45). □
The subtracted bare kite produces the constant that survives in e c l o s e d e_{\mathrm{closed}} e closed at low density. Table S6 checks the coefficients by enumerating the Wick contractions as permutations of the 2 n 2n 2 n interaction ends with one four-cycle for the kite loop and n − 2 n-2 n − 2 two-cycles for the bubbles. For one bubble, choose the bare interaction, the ends of the other two interactions that meet the loop, and the loop direction,
Table S6. Wick counts of the chain family at interaction order n n n . Dividing by 2 n n ! 2^n n! 2 n n ! gives the closed-diagram coefficient.
3 × 2 2 × 2 = 24 , 24 2 3 3 ! = 1 2 . 3 \times2^{2} \times2 = 24,\qquad\frac{24}{2^{3}3!} = \frac{1}{2}. 3 × 2 2 × 2 = 24 , 2 3 3 ! 24 = 2 1 .
These coefficient identities hold in the formal expansion with fixed G 0 G_0 G 0 at the free chemical potential μ 0 ( n e ) \mu_0(n_e) μ 0 ( n e ) . The individual chain terms diverge at small momentum transfer, so the energies are evaluated from the RPA-resummed integrals. The fixed-density energy also contains chemical-potential counterterm diagrams, which lie outside e c l o s e d e_{\mathrm{closed}} e closed .
Coupling-constant integration and coupling weights
This subsection derives the fraction of the closed-diagram coefficient that the coupling average assigns to a chain of m m m bubbles (j j j in the main text), stated after main-text Eq. (44) ,
\frac{\mathrm{BRM\ coefficient}}{\mathrm{closed\mathchar`-diagram\ coefficient}} = \frac{1}{m+2},\qquad m \ge1.
Marking. The BRM energy expression is a coupling-integrated self-energy trace [2 ] (§ 20 ). For diagrams containing a chain, its coefficients differ from those of e c l o s e d e_{\mathrm{closed}} e closed because the two expressions mark different interactions in the Hellmann–Feynman relation
E ( 1 ) − E ( 0 ) = ∫ 0 1 ⟨ V ^ ⟩ λ d λ . (108) E(1)-E(0)=\int_{0}^{1}\langle\hat{V}\rangle_{\lambda}\,\mathrm{d}\lambda. \tag*{(108)} E ( 1 ) − E ( 0 ) = ∫ 0 1 ⟨ V ^ ⟩ λ d λ . (108)
Here E ( λ ) E(\lambda) E ( λ ) is the ground-state energy of H ^ λ = T ^ + λ V ^ \hat{H}_{\lambda}=\hat{T}+\lambda\hat{V} H ^ λ = T ^ + λ V ^ (§ 20.1 ), and d E / d λ = ⟨ V ^ ⟩ λ \mathrm{d}E/\mathrm{d}\lambda=\langle\hat{V}\rangle_{\lambda} d E / d λ = ⟨ V ^ ⟩ λ holds for the normalized ground state (Hellmann–Feynman theorem [22 ] ). A closed diagram with n n n interactions is proportional to λ n \lambda^n λ n . Differentiation marks one interaction, and summing over the n n n markings and integrating over the coupling restores the coefficient,
d d λ λ n = n λ n − 1 , n ∫ 0 1 λ n − 1 d λ = 1. \frac{\mathrm{d}}{\mathrm{d}\lambda}\lambda^n=n\lambda^{n-1},\qquad n\int_{0}^{1}\lambda^{n-1}\,\mathrm{d}\lambda=1. d λ d λ n = n λ n − 1 , n ∫ 0 1 λ n − 1 d λ = 1.
An energy reconstructed from a response or self-energy expression counts only the interactions that this expression marks.
Chain series. The screened line expands as
W λ = ∑ m ≥ 0 λ m + 1 d m , d m = v ( Π 0 v ) m , K m = K [ d m ] , (109) W_{\lambda}=\sum_{m\ge0}\lambda^{m+1}d_m,\qquad d_m=v(\Pi_0v)^m,\qquad K_m=K[d_m], \tag*{(109)} W λ = m ≥ 0 ∑ λ m + 1 d m , d m = v ( Π 0 v ) m , K m = K [ d m ] , (109)
where m m m counts the bubbles, and the kite has m + 2 m+2 m + 2 interactions. The two energy expressions weight these contributions as
e k i t e = 2 ∫ 0 1 K [ W λ ] d λ = ∑ m ≥ 0 2 K m m + 2 , e c l o s e d = K 0 + 2 ∑ m ≥ 1 K m . (110) e_{\mathrm{kite}}=2\int_{0}^{1}K[W_{\lambda}]\,\mathrm{d}\lambda=\sum_{m\ge0}\frac{2K_m}{m+2},\qquad e_{\mathrm{closed}}=K_0+2\sum_{m\ge1}K_m. \tag*{(110)} e kite = 2 ∫ 0 1 K [ W λ ] d λ = m ≥ 0 ∑ m + 2 2 K m , e closed = K 0 + 2 m ≥ 1 ∑ K m . (110)
Both count K 0 = e 2 x K_0=e_{2x} K 0 = e 2 x once. For m ≥ 1 m\ge1 m ≥ 1 the BRM coefficient is 1 / ( m + 2 ) 1/(m+2) 1/ ( m + 2 ) times the closed-diagram coefficient 2 2 2 ,
2 ∫ 0 1 λ m + 1 d λ 2 = 2 / ( m + 2 ) 2 = 1 m + 2 . \frac{2\int_{0}^{1}\lambda^{m+1}\,\mathrm{d}\lambda}{2}=\frac{2/(m+2)}{2}=\frac{1}{m+2}. 2 2 ∫ 0 1 λ m + 1 d λ = 2 2/ ( m + 2 ) = m + 2 1 .
Restored coupling. To trace this ratio, the coupling is restored on every interaction in the family energy,
E c l o s e d ( λ ) = 2 λ K [ W λ ] − λ 2 K [ v ] , E c l o s e d ( 1 ) = e c l o s e d . (111) \mathcal{E}_{\mathrm{closed}}(\lambda)=2\lambda\mathcal{K}[W_\lambda]-\lambda^2\mathcal{K}[v],\qquad\mathcal{E}_{\mathrm{closed}}(1)=e_{\mathrm{closed}}. \tag*{(111)} E closed ( λ ) = 2 λ K [ W λ ] − λ 2 K [ v ] , E closed ( 1 ) = e closed . (111)
With G 0 G_0 G 0 and Π 0 \Pi_0 Π 0 fixed, ∂ λ W λ = v / ( 1 − λ v Π 0 ) 2 \partial_\lambda W_\lambda=v/(1-\lambda v\Pi_0)^2 ∂ λ W λ = v / ( 1 − λ v Π 0 ) 2 , so
E c l o s e d ′ ( λ ) = 2 K [ W λ + λ v ( 1 − λ v Π 0 ) 2 − λ v ] . (112) \mathcal{E}'_{\mathrm{closed}}(\lambda)=2\mathcal{K}\left[W_\lambda+\frac{\lambda v}{(1-\lambda v\Pi_0)^2}-\lambda v\right]. \tag*{(112)} E closed ′ ( λ ) = 2 K [ W λ + ( 1 − λ v Π 0 ) 2 λ v − λ v ] . (112)
The explicit λ \lambda λ marks the bare kite line and ∂ λ W λ \partial_\lambda W_\lambda ∂ λ W λ marks each interaction inside the screened line, so all m + 2 m+2 m + 2 positions are counted. Per m m m -bubble term (m ≥ 1 m\ge1 m ≥ 1 ), with arrows for the sum over m ≥ 1 m\ge1 m ≥ 1 ,
all m + 2 marks : 2 ( m + 2 ) λ m + 1 K m , two kite marks : 4 λ m + 1 K m → 4 K [ W λ − λ v ] , 2 2 ( m + 2 ) = 1 2 ⋅ 2 m + 2 = 1 m + 2 , BRM integrand : 2 λ m + 1 K m → 2 K [ W λ − λ v ] . \begin{aligned}
\text{all }m+2\text{ marks}:&\quad2(m+2)\lambda^{m+1}K_m,\\
\text{two kite marks}:&\quad4\lambda^{m+1}K_m\ \to\ 4\mathcal{K}[W_\lambda-\lambda v],\qquad\frac{2}{2(m+2)}=\frac{1}{2}\cdot\frac{2}{m+2}=\frac{1}{m+2},\\
\text{BRM integrand}:&\quad2\lambda^{m+1}K_m\ \to\ 2\mathcal{K}[W_\lambda-\lambda v].
\end{aligned} all m + 2 marks : two kite marks : BRM integrand : 2 ( m + 2 ) λ m + 1 K m , 4 λ m + 1 K m → 4 K [ W λ − λ v ] , 2 ( m + 2 ) 2 = 2 1 ⋅ m + 2 2 = m + 2 1 , 2 λ m + 1 K m → 2 K [ W λ − λ v ] .
The BRM integrand is half of the two-mark term, because K \mathcal{K} K retains the bare coefficient 1 / 4 1/4 1/4 in place of the closed-diagram coefficient 1 / 2 1/2 1/2 . It also marks none of the added interactions. The two factors 1 / 2 1/2 1/2 and 2 / ( m + 2 ) 2/(m+2) 2/ ( m + 2 ) give the ratio 1 / ( m + 2 ) 1/(m+2) 1/ ( m + 2 ) in (S48). Since F = K [ W λ = 1 ] F=\mathcal{K}[W_{\lambda=1}] F = K [ W λ = 1 ] , e k i t e e_{\mathrm{kite}} e kite and e c l o s e d e_{\mathrm{closed}} e closed are all expressed through the same functional K \mathcal{K} K , they obey main-text Eq. (45) .
Both kite lines screened
This subsection supports the remark after main-text Eq. (51) on the constant − e 2 x -e_{2x} − e 2 x of e c l o s e d e_{\mathrm{closed}} e closed at low density. Let K 2 [ f , g ] \mathcal{K}_2[f,g] K 2 [ f , g ] be the kite with its two interactions replaced by f f f and g g g and the bare coefficient 1 / 4 1/4 1/4 retained. Then K 2 \mathcal{K}_2 K 2 is bilinear and symmetric, and K [ f ] = K 2 [ f , v ] \mathcal{K}[f]=\mathcal{K}_2[f,v] K [ f ] = K 2 [ f , v ] . A diagram with m 1 m_1 m 1 and m 2 m_2 m 2 bubbles on the two lines appears in K 2 [ W λ = 1 , W λ = 1 ] \mathcal{K}_2[W_{\lambda=1},W_{\lambda=1}] K 2 [ W λ = 1 , W λ = 1 ] once for m 1 = m 2 m_1=m_2 m 1 = m 2 and twice for m 1 ≠ m 2 m_1\ne m_2 m 1 = m 2 . The proof of Proposition S2 gives the two symmetries for m 1 ≠ m 2 m_1\ne m_2 m 1 = m 2 . For m 1 = m 2 m_1=m_2 m 1 = m 2 , the core rotation 1 → 2 → 3 → 4 → 1 1\to2\to3\to4\to1 1 → 2 → 3 → 4 → 1 , which exchanges the two lines, also extends to the chains and gives four symmetries. This counting reproduces the closed-diagram coefficients without subtraction:
Instances of both cases are checked by enumeration in § 30 (Table S6 for m 2 = 0 m_2=0 m 2 = 0 , Figure S1 for m 1 = m 2 = 1 m_1=m_2=1 m 1 = m 2 = 1 ). By bilinearity and symmetry of K 2 \mathcal{K}_2 K 2 ,
K 2 [ W λ = 1 , W λ = 1 ] = K 2 [ v , v ] + 2 K 2 [ W λ = 1 − v , v ] + K 2 [ W λ = 1 − v , W λ = 1 − v ] , \mathcal{K}_2[W_{\lambda=1},W_{\lambda=1}]=\mathcal{K}_2[v,v]+2\mathcal{K}_2[W_{\lambda=1}-v,v]+\mathcal{K}_2[W_{\lambda=1}-v,W_{\lambda=1}-v], K 2 [ W λ = 1 , W λ = 1 ] = K 2 [ v , v ] + 2 K 2 [ W λ = 1 − v , v ] + K 2 [ W λ = 1 − v , W λ = 1 − v ] ,
and with Eq. (107) ,
K 2 [ W λ = 1 , W λ = 1 ] = e c l o s e d + K 2 [ W λ = 1 − v , W λ = 1 − v ] . (113) \mathcal{K}_2[W_{\lambda=1},W_{\lambda=1}]=e_{\mathrm{closed}}+\mathcal{K}_2[W_{\lambda=1}-v,W_{\lambda=1}-v]. \tag*{(113)} K 2 [ W λ = 1 , W λ = 1 ] = e closed + K 2 [ W λ = 1 − v , W λ = 1 − v ] . (113)
The chain family is the part of degree at most one in W λ = 1 − v W_{\lambda=1}-v W λ = 1 − v . Its constant − e 2 x -e_{2x} − e 2 x is compensated by the two-chain terms if K 2 [ W λ = 1 , W λ = 1 ] \mathcal{K}_2[W_{\lambda=1},W_{\lambda=1}] K 2 [ W λ = 1 , W λ = 1 ] vanishes at low density together with F F F . This vanishing is expected but is not established here. The spectral decomposition of the bilinear K 2 \mathcal{K}_2 K 2 is a double integral over the phase velocities of two Lorentzian lines, so the one-dimensional reductions of main-text § 2 do not apply directly.
Coupling averages and density scaling
With W λ ( r s ) = λ W λ = 1 ( λ r s ) W_\lambda(r_s)=\lambda W_{\lambda=1}(\lambda r_s) W λ ( r s ) = λ W λ = 1 ( λ r s ) [main-text § 2.3 , after Eq. (12) ] and the formal expansion (S47), the three energy expressions of main-text § 3.2 weight the m m m -bubble term as in table S7 (K 0 = e 2 x K_0=e_{2x} K 0 = e 2 x is counted once in each).
Table S7. Factors of K m K_m K m in the three energy expressions.
Direction of transfer. The asymptotic forms of F F F are derived directly at full coupling. At high density they follow from the Thomas–Fermi tail − h / t 2 -h/t^2 − h / t 2 of the full-coupling line (main-text table 1) and the window estimate (S29), in which ∣ c / ( 1 + c ) ∣ ≤ 1 |c/(1+c)|\le1 ∣ c / ( 1 + c ) ∣ ≤ 1 replaces ∣ ρ − 1 ∣ ≤ 1 |\rho-1|\le1 ∣ ρ − 1∣ ≤ 1 . At low density they follow from the bounds of § 29.2 , which hold at each fixed λ \lambda λ with explicit factors λ − 1 / 2 \lambda^{-1/2} λ − 1/2 , λ − 3 / 4 \lambda^{-3/4} λ − 3/4 and λ − 1 \lambda^{-1} λ − 1 and therefore at λ = 1 \lambda=1 λ = 1 . The moment rule of § 25.5 carries asymptotic terms only from the full-coupling energy to the coupling average, since an asymptotic remainder is not in general preserved by differentiation. A term of F F F with constant coefficient φ \varphi φ appears in e k i t e e_{\mathrm{kite}} e kite as
φ r s p ⟼ 2 ∫ 0 1 λ φ ( λ r s ) p d λ = 2 φ r s p p + 2 ( p > − 2 ) , \varphi r_s^p \longmapsto2\int_0^1\lambda\varphi(\lambda r_s)^p\,\mathrm{d}\lambda=\frac{2\varphi r_s^p}{p+2}\qquad(p>-2), φ r s p ⟼ 2 ∫ 0 1 λ φ ( λ r s ) p d λ = p + 2 2 φ r s p ( p > − 2 ) ,
and a term φ r s p log 2 r s \varphi r_s^p\log^2 r_s φ r s p log 2 r s as stated at the end of § 25.5 . The full-coupling tail is 3 2 \frac{3}{2} 2 3 times the averaged tail that determines c h d c_{\mathrm{hd}} c hd in (S32), and main-text Eq. (44) gives e c l o s e d e_{\mathrm{closed}} e closed ,
− h t 2 = 3 2 ( − 2 h 3 t 2 ) , e c l o s e d = 2 F − e 2 x . -\frac{h}{t^2}=\frac{3}{2}\left(-\frac{2h}{3t^2}\right),\qquad e_{\mathrm{closed}}=2F-e_{2x}. − t 2 h = 2 3 ( − 3 t 2 2 h ) , e closed = 2 F − e 2 x .
These relations give the F F F and e c l o s e d e_{\mathrm{closed}} e closed entries of main-text Eq. (50) and table 3.
Low density. At leading order F ( λ r s ) ≃ λ − 1 / 4 F ( r s ) F(\lambda r_s)\simeq\lambda^{-1/4}F(r_s) F ( λ r s ) ≃ λ − 1/4 F ( r s ) , and the first relation of main-text Eq. (45) gives
e k i t e = 2 ∫ 0 1 λ 3 / 4 d λ F = 8 7 F , F = 7 8 e k i t e . e_{\mathrm{kite}}=2\int_0^1\lambda^{3/4}\,\mathrm{d}\lambda\,F=\frac{8}{7}F,\qquad F=\frac{7}{8}e_{\mathrm{kite}}. e kite = 2 ∫ 0 1 λ 3/4 d λ F = 7 8 F , F = 8 7 e kite .
The value q ∗ − 1 = [ 3 π / ( 16 α L ) ] 1 / 4 r s − 1 / 4 = 1.0311348 … r s − 1 / 4 q_*^{-1}=[3\pi/(16\alpha_{\mathrm{L}})]^{1/4}r_s^{-1/4}=1.0311348\ldots r_s^{-1/4} q ∗ − 1 = [ 3 π / ( 16 α L ) ] 1/4 r s − 1/4 = 1.0311348 … r s − 1/4 expresses A 1 / q ∗ A_1/q_* A 1 / q ∗ in powers of r s r_s r s . At low density,
e k i t e ( r s ) = − 0.2213782293 … r s − 1 / 4 + o ( r s − 1 / 4 ) , F ( r s ) = − 0.1937059506 … r s − 1 / 4 + o ( r s − 1 / 4 ) , e_{\mathrm{kite}}(r_s)=-0.2213782293\ldots r_s^{-1/4}+o(r_s^{-1/4}),
\qquad
F(r_s)=-0.1937059506\ldots r_s^{-1/4}+o(r_s^{-1/4}), e kite ( r s ) = − 0.2213782293 … r s − 1/4 + o ( r s − 1/4 ) , F ( r s ) = − 0.1937059506 … r s − 1/4 + o ( r s − 1/4 ) ,
e c l o s e d ( r s ) = − e 2 x − 0.38741190129 … r s − 1 / 4 + o ( r s − 1 / 4 ) . (114) e_{\mathrm{closed}}(r_s)=-e_{2x}-0.38741190129\ldots r_s^{-1/4}+o(r_s^{-1/4}). \tag*{(114)} e closed ( r s ) = − e 2 x − 0.38741190129 … r s − 1/4 + o ( r s − 1/4 ) . (114)
Spectral representation and convergence
This section derives the spectral representation of main-text Eq. (14) and the absolute bounds that justify interchanging the spectral and energy integrations in main-text Eq. (15) .
Loss spectrum and Lorentzian lines
With i ν → ω + i 0 i\nu\to\omega+ i0 i ν → ω + i 0 for real frequency ω \omega ω , the imaginary-frequency Lorentzian line of main-text Eq. (14) continues analytically to the retarded line
b Λ ( q , ω ) = Λ 2 ( q Λ ) 2 − ( ω + i 0 ) 2 , ω p o l e s = ± q Λ − i 0. (115) b_{\Lambda}(q,\omega)=\frac{\Lambda^2}{(q\Lambda)^2-(\omega+i0)^2},
\qquad
\omega_{\mathrm{poles}}=\pm q\Lambda-i0. \tag*{(115)} b Λ ( q , ω ) = ( q Λ ) 2 − ( ω + i 0 ) 2 Λ 2 , ω poles = ± q Λ − i 0. (115)
The loss spectrum, defined for ω > 0 \omega>0 ω > 0 , gives the screened interaction as
B W , λ ( q , ω ) = − 1 π Im W λ R ( q , ω ) ≥ 0 , B_{W,\lambda}(q,\omega)=-\frac{1}{\pi}\operatorname{Im}W_{\lambda}^{R}(q,\omega)\geq0, B W , λ ( q , ω ) = − π 1 Im W λ R ( q , ω ) ≥ 0 ,
W λ ( q , i ν ) = λ v ( q ) − ∫ 0 ∞ 2 ω B W , λ ( q , ω ) ν 2 + ω 2 d ω . (116) W_{\lambda}(q,i\nu)=\lambda v(q)-\int_{0}^{\infty}\frac{2\omega B_{W,\lambda}(q,\omega)}{\nu^2+\omega^2}\,\mathrm{d}\omega. \tag*{(116)} W λ ( q , i ν ) = λ v ( q ) − ∫ 0 ∞ ν 2 + ω 2 2 ω B W , λ ( q , ω ) d ω . (116)
Here W λ R W_{\lambda}^{R} W λ R and Π 0 R \Pi_{0}^{R} Π 0 R are the retarded functions, obtained by i ν → ω + i 0 i\nu\to\omega+i0 i ν → ω + i 0 as in Eq. (115) . The second equation is the Kramers–Kronig relation at fixed q > 0 q>0 q > 0 [23 ] . It applies for two reasons. First, for a stable ground state W λ R − λ v W_{\lambda}^{R}-\lambda v W λ R − λ v is analytic in the upper half of the complex ω \omega ω plane and vanishes at large ∣ ω ∣ \lvert\omega\rvert ∣ ω ∣ there, since Π 0 R ( q , ω ) \Pi_{0}^{R}(q,\omega) Π 0 R ( q , ω ) decays as ∣ ω ∣ − 2 \lvert\omega\rvert^{-2} ∣ ω ∣ − 2 in the whole upper half plane (closed form of the Lindhard function). Second, W λ W_{\lambda} W λ is real in the time domain, hence
Im W λ R ( q , − ω ) = − Im W λ R ( q , ω ) . \operatorname{Im}W_{\lambda}^{R}(q,-\omega)=-\operatorname{Im}W_{\lambda}^{R}(q,\omega). Im W λ R ( q , − ω ) = − Im W λ R ( q , ω ) .
The inequality B W , λ ≥ 0 B_{W,\lambda}\geq0 B W , λ ≥ 0 is the passivity condition of main-text § 2.4 .
Phase velocity. With ω = Λ q \omega=\Lambda q ω = Λ q , d ω = q d Λ \mathrm{d}\omega=q\,\mathrm{d}\Lambda d ω = q d Λ and b Λ ( q , i ν ) = Λ 2 / ( q 2 Λ 2 + ν 2 ) b_{\Lambda}(q,i\nu)=\Lambda^2/(q^2\Lambda^2+\nu^2) b Λ ( q , i ν ) = Λ 2 / ( q 2 Λ 2 + ν 2 ) ,
2 ω B W , λ ( q , ω ) ν 2 + ω 2 d ω = 2 q 2 Λ B W , λ ( q , q Λ ) d Λ ν 2 + q 2 Λ 2 = 2 q 2 Λ B W , λ ( q , q Λ ) b Λ ( q , i ν ) d Λ , \frac{2\omega B_{W,\lambda}(q,\omega)}{\nu^2+\omega^2}\,\mathrm{d}\omega
=
\frac{2q^2\Lambda B_{W,\lambda}(q,q\Lambda)\,\mathrm{d}\Lambda}{\nu^2+q^2\Lambda^2}
=
\frac{2q^2}{\Lambda}B_{W,\lambda}(q,q\Lambda)b_{\Lambda}(q,i\nu)\,\mathrm{d}\Lambda, ν 2 + ω 2 2 ω B W , λ ( q , ω ) d ω = ν 2 + q 2 Λ 2 2 q 2 Λ B W , λ ( q , q Λ ) d Λ = Λ 2 q 2 B W , λ ( q , q Λ ) b Λ ( q , i ν ) d Λ ,
and therefore
W λ ( q , i ν ) = λ v ( q ) − ∫ 0 ∞ d Λ a λ ( q , Λ ) b Λ ( q , i ν ) , a λ ( q , Λ ) = 2 q 2 Λ B W , λ ( q , q Λ ) ≥ 0. (117) W_{\lambda}(q,i\nu)=\lambda v(q)-\int_{0}^{\infty}\mathrm{d}\Lambda\,a_{\lambda}(q,\Lambda)b_{\Lambda}(q,i\nu),
\qquad
a_{\lambda}(q,\Lambda)=\frac{2q^2}{\Lambda}B_{W,\lambda}(q,q\Lambda)\geq0. \tag*{(117)} W λ ( q , i ν ) = λ v ( q ) − ∫ 0 ∞ d Λ a λ ( q , Λ ) b Λ ( q , i ν ) , a λ ( q , Λ ) = Λ 2 q 2 B W , λ ( q , q Λ ) ≥ 0. (117)
Functionals. Linearity in the selected line gives, for T = Π v t x ( 1 ) , Π S E ( 1 ) , Π K T ( 1 ) T=\Pi_{\mathrm{vtx}}^{(1)},\Pi_{\mathrm{SE}}^{(1)},\Pi_{\mathrm{KT}}^{(1)} T = Π vtx ( 1 ) , Π SE ( 1 ) , Π KT ( 1 ) ,
K T [ W λ ] = λ K T [ v ] − ∫ 0 ∞ d Λ K T [ a λ ( q , Λ ) b Λ ( q , i ν ) ] . (118) \mathcal{K}_{T}[W_{\lambda}]
=
\lambda\mathcal{K}_{T}[v]
-
\int_{0}^{\infty}\mathrm{d}\Lambda\,\mathcal{K}_{T}[a_{\lambda}(q,\Lambda)b_{\Lambda}(q,i\nu)]. \tag*{(118)} K T [ W λ ] = λ K T [ v ] − ∫ 0 ∞ d Λ K T [ a λ ( q , Λ ) b Λ ( q , i ν )] . (118)
The functional acts on the product a λ ( q , Λ ) b Λ ( q , i ν ) a_{\lambda}(q,\Lambda)b_{\Lambda}(q,i\nu) a λ ( q , Λ ) b Λ ( q , i ν ) as a line at fixed Λ \Lambda Λ . Here K T [ f ] \mathcal{K}_{T}[f] K T [ f ] denotes main-text Eq. (52) with Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) replaced by T T T , hence
K Π v t x ( 1 ) = K ( Sec. S9.2 ) , K Π ( 1 ) = K Π v t x ( 1 ) + K Π S E ( 1 ) . \mathcal{K}_{\Pi_{\mathrm{vtx}}^{(1)}}=\mathcal{K}\quad(\text{Sec. S9.2}),
\qquad
\mathcal{K}_{\Pi^{(1)}}=\mathcal{K}_{\Pi_{\mathrm{vtx}}^{(1)}}+\mathcal{K}_{\Pi_{\mathrm{SE}}^{(1)}}. K Π vtx ( 1 ) = K ( Sec. S9.2 ) , K Π ( 1 ) = K Π vtx ( 1 ) + K Π SE ( 1 ) .
The spectral measure includes any plasmon point masses, and the momentum-dependent weight remains inside K T K_T K T (main-text § 2.4 ).
Interchange. The interchange of the spectral and energy integrations follows by the argument of § 24.4 from the bounds of § 27 (small momentum transfer) and § 24.3 and from a large-momentum estimate. This estimate uses
∣ δ p ∣ , ∣ δ k ∣ ≥ q 2 / 2 − q ( q > 2 ) , ∫ 0 ∞ d ν ν 2 + δ 2 = π 2 ∣ δ ∣ . |\delta_p|,|\delta_k| \ge q^2/2-q\quad(q>2),\qquad\int_0^\infty\frac{\mathrm{d}\nu}{\nu^2+\delta^2}=\frac{\pi}{2|\delta|}. ∣ δ p ∣ , ∣ δ k ∣ ≥ q 2 /2 − q ( q > 2 ) , ∫ 0 ∞ ν 2 + δ 2 d ν = 2∣ δ ∣ π .
the first from the end of § 29.1 , with the Coulomb integrals of the four regions from Sec. S17.1. For every q > 0 q>0 q > 0 ,
∣ ( i ν − δ p ) ( i ν − δ k ) ∣ − 1 ≤ 1 2 [ ( ν 2 + δ p 2 ) − 1 + ( ν 2 + δ k 2 ) − 1 ] , ∫ B d 3 k ∣ p − k ∣ 2 = 4 π f L ( p ) ≤ 4 π . \left|(i\nu-\delta_p)(i\nu-\delta_k)\right|^{-1}\le\frac{1}{2}\left[(\nu^2+\delta_p^2)^{-1}+(\nu^2+\delta_k^2)^{-1}\right],\qquad\int_{\mathbb{B}}\frac{\mathrm{d}^3k}{|\mathbf{p}-\mathbf{k}|^2}=4\pi f_L(p)\le4\pi. ∣ ( i ν − δ p ) ( i ν − δ k ) ∣ − 1 ≤ 2 1 [ ( ν 2 + δ p 2 ) − 1 + ( ν 2 + δ k 2 ) − 1 ] , ∫ B ∣ p − k ∣ 2 d 3 k = 4 π f L ( p ) ≤ 4 π .
The latter is applied to B \mathbb{B} B and to its translate B − q \mathbb{B}-\mathbf{q} B − q , which together contain the two transition regions of k \mathbf{k} k . By the free response of § 20.2 at ν = 0 \nu=0 ν = 0 , ∫ d 3 p ∣ Δ p ∣ / ∣ δ p ∣ = 4 π L ( q , 0 ) = 4 π f L ( q / 2 ) \int\mathrm{d}^3p\,|\Delta_p|/|\delta_p|=4\pi\mathcal{L}(q,0)=4\pi f_L(q/2) ∫ d 3 p ∣ Δ p ∣/∣ δ p ∣ = 4 π L ( q , 0 ) = 4 π f L ( q /2 ) . The two estimates, the symmetry p ↔ k \mathbf{p}\leftrightarrow\mathbf{k} p ↔ k and 4 π + 4 π = 8 π 4\pi+4\pi=8\pi 4 π + 4 π = 8 π for the k \mathbf{k} k integral give
∫ 0 ∞ ∣ X ∣ d ν ≤ π 2 ∫ d 3 p ∣ Δ p ∣ ∣ δ p ∣ ∫ B ∪ ( B − q ) d 3 k ∣ p − k ∣ 2 ≤ 4 π 2 ∫ d 3 p ∣ Δ p ∣ ∣ δ p ∣ = 4 π 2 ⋅ 4 π L ( q , 0 ) = 16 π 3 f L ( q / 2 ) ≤ 16 π 3 . \begin{aligned}
\int_0^\infty|X|\,\mathrm{d}\nu\le\frac{\pi}{2}\int\mathrm{d}^3p\,\frac{|\Delta_p|}{|\delta_p|}\int_{\mathbb{B}\cup(\mathbb{B}-\mathbf{q})}\frac{\mathrm{d}^3k}{|\mathbf{p}-\mathbf{k}|^2}\le4\pi^2\int\mathrm{d}^3p\,\frac{|\Delta_p|}{|\delta_p|} \\
&=4\pi^2\cdot4\pi\mathcal{L}(q,0)=16\pi^3f_L(q/2)\le16\pi^3.
\end{aligned} ∫ 0 ∞ ∣ X ∣ d ν ≤ 2 π ∫ d 3 p ∣ δ p ∣ ∣ Δ p ∣ ∫ B ∪ ( B − q ) ∣ p − k ∣ 2 d 3 k ≤ 4 π 2 ∫ d 3 p ∣ δ p ∣ ∣ Δ p ∣ = 4 π 2 ⋅ 4 π L ( q , 0 ) = 16 π 3 f L ( q /2 ) ≤ 16 π 3 .
At large q q q , f L ( q / 2 ) ≃ 4 / ( 3 q 2 ) f_L(q/2)\simeq4/(3q^2) f L ( q /2 ) ≃ 4/ ( 3 q 2 ) gives the entry O ( q − 2 ) O(q^{-2}) O ( q − 2 ) below. At small q q q , this bound is only O ( 1 ) O(1) O ( 1 ) . The entry O ( q log ( e / q ) ) O(q\log(e/q)) O ( q log ( e / q )) follows instead from the absolute frequency bound ∫ ∣ X ( q , q z ) ∣ d z = O ( log ( e / q ) ) \int|X(q,qz)|\,\mathrm{d}z=O(\log(e/q)) ∫ ∣ X ( q , q z ) ∣ d z = O ( log ( e / q )) of § 27 , with d ν = q d z \mathrm{d}\nu=q\,\mathrm{d}z d ν = q d z . The lower bound on ∫ 0 ∞ a λ b λ d Λ \int_0^\infty a_\lambda b_\lambda\,\mathrm{d}\Lambda ∫ 0 ∞ a λ b λ d Λ holds because a λ ≥ 0 a_\lambda\ge0 a λ ≥ 0 and b λ > 0 b_\lambda>0 b λ > 0 , and the upper bound because W λ = λ / ( q 2 + λ q T F 2 L ) ≥ 0 W_\lambda=\lambda/(q^2+\lambda q_{\mathrm{TF}}^2\mathcal{L})\ge0 W λ = λ / ( q 2 + λ q TF 2 L ) ≥ 0 . Below, log ( e / q ) = 1 + log ( 1 / q ) \log(e/q)=1+\log(1/q) log ( e / q ) = 1 + log ( 1/ q ) , and the last relation includes the angular integration of d 3 q \mathrm{d}^3q d 3 q . For every q > 0 q>0 q > 0 ,
0 ≤ ∫ 0 ∞ a λ b λ d Λ ≤ λ v , ∫ 0 ∞ ∣ X ( q , ν ) ∣ d ν = { O ( q log ( e / q ) ) , q → 0 , O ( q − 2 ) , q → ∞ , d 3 q = 4 π d q . (119) 0\le\int_0^\infty a_\lambda b_\lambda\,\mathrm{d}\Lambda\le\lambda v,\qquad\int_0^\infty|X(q,\nu)|\,\mathrm{d}\nu=
\begin{cases}
O(q\log(e/q)),&q\to0,\\
O(q^{-2}),&q\to\infty,
\end{cases}
\qquad\mathrm{d}^3q=4\pi\,\mathrm{d}q. \tag*{(119)} 0 ≤ ∫ 0 ∞ a λ b λ d Λ ≤ λ v , ∫ 0 ∞ ∣ X ( q , ν ) ∣ d ν = { O ( q log ( e / q )) , O ( q − 2 ) , q → 0 , q → ∞ , d 3 q = 4 π d q . (119)
Elementary frequency integrals
This subsection derives the two-factor frequency integral cited in main-text § 2.5 and its static limit, main-text Eq. (18) . Let a a a and b b b be the signed excitation energies and ω > 0 \omega>0 ω > 0 the Lorentzian width in ν \nu ν (ω = q Λ \omega=q\Lambda ω = q Λ for the line b Λ b_\Lambda b Λ ). For a , b ≠ 0 a,b\ne0 a , b = 0 ,
J ( ω ; a , b ) = 1 π ∫ 0 ∞ a b − ν 2 ( a 2 + ν 2 ) ( b 2 + ν 2 ) ( ν 2 + ω 2 ) d ν . (120) J(\omega;a,b)=\frac{1}{\pi}\int_0^\infty\frac{ab-\nu^2}{(a^2+\nu^2)(b^2+\nu^2)(\nu^2+\omega^2)}\,\mathrm{d}\nu. \tag*{(120)} J ( ω ; a , b ) = π 1 ∫ 0 ∞ ( a 2 + ν 2 ) ( b 2 + ν 2 ) ( ν 2 + ω 2 ) ab − ν 2 d ν . (120)
Partial fractions. With α = ∣ a ∣ \alpha=|a| α = ∣ a ∣ , β = ∣ b ∣ \beta=|b| β = ∣ b ∣ , distinct positive widths, and c c c running over the three widths,
1 ( α 2 + ν 2 ) ( β 2 + ν 2 ) ( ω 2 + ν 2 ) = ∑ c ∈ { α , β , ω } r c c 2 + ν 2 , (121) \frac{1}{(\alpha^2+\nu^2)(\beta^2+\nu^2)(\omega^2+\nu^2)}=\sum_{c\in\{\alpha,\beta,\omega\}}\frac{r_c}{c^2+\nu^2}, \tag*{(121)} ( α 2 + ν 2 ) ( β 2 + ν 2 ) ( ω 2 + ν 2 ) 1 = c ∈ { α , β , ω } ∑ c 2 + ν 2 r c , (121)
r c = ∏ c ′ ≠ c 1 c ′ 2 − c 2 , (122) r_c=\prod_{c'\ne c}\frac{1}{c'^2-c^2}, \tag*{(122)} r c = c ′ = c ∏ c ′2 − c 2 1 , (122)
and we denote the denominator on the left by
N ( ν ) : = ( α 2 + ν 2 ) ( β 2 + ν 2 ) ( ω 2 + ν 2 ) . (123) N(\nu):=(\alpha^2+\nu^2)(\beta^2+\nu^2)(\omega^2+\nu^2). \tag*{(123)} N ( ν ) := ( α 2 + ν 2 ) ( β 2 + ν 2 ) ( ω 2 + ν 2 ) . (123)
Moments. The coefficient of ν − 2 \nu^{-2} ν − 2 in (S59) at large ν \nu ν gives ∑ c r c = 0 \sum_c r_c=0 ∑ c r c = 0 , hence
ν 2 N ( ν ) = ∑ c r c ( 1 − c 2 c 2 + ν 2 ) = − ∑ c c 2 r c c 2 + ν 2 , ∫ 0 ∞ d ν c 2 + ν 2 = π 2 c . \frac{\nu^2}{N(\nu)}=\sum_c r_c\left(1-\frac{c^2}{c^2+\nu^2}\right)=-\sum_c\frac{c^2r_c}{c^2+\nu^2},\qquad\int_0^\infty\frac{\mathrm{d}\nu}{c^2+\nu^2}=\frac{\pi}{2c}. N ( ν ) ν 2 = c ∑ r c ( 1 − c 2 + ν 2 c 2 ) = − c ∑ c 2 + ν 2 c 2 r c , ∫ 0 ∞ c 2 + ν 2 d ν = 2 c π .
Term-by-term integration gives
1 π ∫ 0 ∞ d ν N ( ν ) = α + β + ω 2 α β ω ( α + β ) ( β + ω ) ( ω + α ) , (124) \frac{1}{\pi}\int_0^\infty\frac{\mathrm{d}\nu}{N(\nu)}=\frac{\alpha+\beta+\omega}{2\alpha\beta\omega(\alpha+\beta)(\beta+\omega)(\omega+\alpha)}, \tag*{(124)} π 1 ∫ 0 ∞ N ( ν ) d ν = 2 α β ω ( α + β ) ( β + ω ) ( ω + α ) α + β + ω , (124)
1 π ∫ 0 ∞ ν 2 d ν N ( ν ) = 1 2 ( α + β ) ( β + ω ) ( ω + α ) . (125) \frac{1}{\pi}\int_0^\infty\frac{\nu^2\mathrm{d}\nu}{N(\nu)}=\frac{1}{2(\alpha+\beta)(\beta+\omega)(\omega+\alpha)}. \tag*{(125)} π 1 ∫ 0 ∞ N ( ν ) ν 2 d ν = 2 ( α + β ) ( β + ω ) ( ω + α ) 1 . (125)
Equation (S58) equals a b ab ab times (S62) minus (S63). With a b = ± α β ab=\pm\alpha\beta ab = ± α β ,
J ( ω ; a , b ) = ± ( α + β + ω ) − ω 2 ω ( α + β ) ( β + ω ) ( ω + α ) , J(\omega;a,b)=\frac{\pm(\alpha+\beta+\omega)-\omega}{2\omega(\alpha+\beta)(\beta+\omega)(\omega+\alpha)}, J ( ω ; a , b ) = 2 ω ( α + β ) ( β + ω ) ( ω + α ) ± ( α + β + ω ) − ω ,
which is
J ( ω ; a , b ) = { 1 2 ω ( α + ω ) ( β + ω ) , a b > 0 , − α + β + 2 ω 2 ω ( α + β ) ( α + ω ) ( β + ω ) , a b < 0. (126) J(\omega;a,b)=
\begin{cases}
\frac{1}{2\omega(\alpha+\omega)(\beta+\omega)},&ab>0,\\[6pt]
-\frac{\alpha+\beta+2\omega}{2\omega(\alpha+\beta)(\alpha+\omega)(\beta+\omega)},&ab<0.
\end{cases} \tag*{(126)} J ( ω ; a , b ) = ⎩ ⎨ ⎧ 2 ω ( α + ω ) ( β + ω ) 1 , − 2 ω ( α + β ) ( α + ω ) ( β + ω ) α + β + 2 ω , ab > 0 , ab < 0. (126)
The closed form is continuous when positive widths coincide, although the partial-fraction coefficients diverge there.
Static limit. Since
Re 1 ( i ν − a ) ( i ν − b ) = a b − ν 2 ( a 2 + ν 2 ) ( b 2 + ν 2 ) , \operatorname{Re}\frac{1}{(i\nu-a)(i\nu-b)}=\frac{ab-\nu^2}{(a^2+\nu^2)(b^2+\nu^2)}, Re ( i ν − a ) ( i ν − b ) 1 = ( a 2 + ν 2 ) ( b 2 + ν 2 ) ab − ν 2 ,
the static integral of main-text Eq. (18) is the limit ω → ∞ \omega\to\infty ω → ∞ of ω 2 J ( ω ; a , b ) \omega^2J(\omega;a,b) ω 2 J ( ω ; a , b ) . By (S64),
lim ω → ∞ ω 2 J ( ω ; a , b ) = 0 ( a b > 0 ) , lim ω → ∞ ω 2 J ( ω ; a , b ) = − 1 α + β ( a b < 0 ) . \lim_{\omega\to\infty}\omega^2J(\omega;a,b)=0\quad(ab>0),\qquad\lim_{\omega\to\infty}\omega^2J(\omega;a,b)=-\frac{1}{\alpha+\beta}\quad(ab<0). ω → ∞ lim ω 2 J ( ω ; a , b ) = 0 ( ab > 0 ) , ω → ∞ lim ω 2 J ( ω ; a , b ) = − α + β 1 ( ab < 0 ) .
Absolute bounds for the self-energy pair
These bounds justify the spectral interchange for the self-energy pair and give the remainders of e S E e_{\mathrm{SE}} e SE used in Secs. S10.2 and S17.4 for main-text § 3.4 . For δ ≠ 0 \delta\ne0 δ = 0 ,
∫ 0 ∞ ∣ Re ( i ν − δ ) − 2 ∣ d ν = 1 ∣ δ ∣ . (127) \int_0^\infty\left|\operatorname{Re}(i\nu-\delta)^{-2}\right|\mathrm{d}\nu=\frac{1}{|\delta|}. \tag*{(127)} ∫ 0 ∞ Re ( i ν − δ ) − 2 d ν = ∣ δ ∣ 1 . (127)
This follows from
Re ( i ν − δ ) − 2 = δ 2 − ν 2 ( ν 2 + δ 2 ) 2 = ∂ ν ν ν 2 + δ 2 , ν ν 2 + δ 2 ∣ ν = ∣ δ ∣ = 1 2 ∣ δ ∣ , \operatorname{Re}(i\nu-\delta)^{-2}=\frac{\delta^2-\nu^2}{(\nu^2+\delta^2)^2}=\partial_\nu\frac{\nu}{\nu^2+\delta^2},\qquad\left.\frac{\nu}{\nu^2+\delta^2}\right|_{\nu=|\delta|}=\frac{1}{2|\delta|}, Re ( i ν − δ ) − 2 = ( ν 2 + δ 2 ) 2 δ 2 − ν 2 = ∂ ν ν 2 + δ 2 ν , ν 2 + δ 2 ν ν = ∣ δ ∣ = 2∣ δ ∣ 1 ,
since the integrand changes sign only at ν = ∣ δ ∣ \nu=|\delta| ν = ∣ δ ∣ . Inserting (S21), with 2 C = 2 π 2 q T F 2 2C=2\pi^2q_{\mathrm{TF}}^2 2 C = 2 π 2 q TF 2 , and (S65) gives
∫ 0 ∞ ∣ Π S E ( 1 ) ( q , i ν ) ∣ d ν ≤ 2 π 2 q T F 2 ∫ d 3 p ( 2 π ) 3 ∣ Δ p ∣ ∣ Σ x ( p + q ) − Σ x ( p ) ∣ ∣ δ p ∣ . (128) \int_0^\infty\left|\Pi_{\mathrm{SE}}^{(1)}(q,i\nu)\right|\mathrm{d}\nu\le2\pi^2q_{\mathrm{TF}}^2\int\frac{\mathrm{d}^3p}{(2\pi)^3}|\Delta_p|\frac{\left|\Sigma_x(p+q)-\Sigma_x(p)\right|}{|\delta_p|}. \tag*{(128)} ∫ 0 ∞ Π SE ( 1 ) ( q , i ν ) d ν ≤ 2 π 2 q TF 2 ∫ ( 2 π ) 3 d 3 p ∣ Δ p ∣ ∣ δ p ∣ ∣ Σ x ( p + q ) − Σ x ( p ) ∣ . (128)
Small momentum transfer. For 0 < q < 1 / 2 0<q<1/2 0 < q < 1/2 , we parameterize the Δ p = 1 \Delta_p=1 Δ p = 1 transition region by the distances of the two momenta from the Fermi surface,
r = ∣ p ∣ = 1 − ξ , r ′ = ∣ p + q ∣ = 1 + ζ , ξ , ζ ≥ 0 , s = ξ + ζ ≤ q , r=|\mathbf{p}|=1-\xi,\qquad r'=|\mathbf{p}+\mathbf{q}|=1+\zeta,\qquad\xi,\zeta\ge0,\qquad s=\xi+\zeta\le q, r = ∣ p ∣ = 1 − ξ , r ′ = ∣ p + q ∣ = 1 + ζ , ξ , ζ ≥ 0 , s = ξ + ζ ≤ q ,
with s s s a distance, not the Mellin variable. Then
d 3 p = 2 π q r r ′ d ξ d ζ , ∣ δ p ∣ = s ( r + r ′ ) 2 . (129) \mathrm{d}^{3}p=\frac{2\pi}{q}rr'\,\mathrm{d}\xi\,\mathrm{d}\zeta,\qquad|\delta_p|=\frac{s(r+r')}{2}. \tag*{(129)} d 3 p = q 2 π r r ′ d ξ d ζ , ∣ δ p ∣ = 2 s ( r + r ′ ) . (129)
In the bounds below, C b C_b C b is independent of q q q and r s r_s r s (Notation). Near r = 1 r=1 r = 1 , differentiating f L f_L f L in Eq. (81) gives ∣ Σ x ′ ( r ) ∣ ≤ C b q T F [ 1 + ∣ log ∣ r − 1 ∣ ∣ ] |\Sigma_x'(r)|\le C_bq_{\mathrm{TF}}[1+|\log|r-1||] ∣ Σ x ′ ( r ) ∣ ≤ C b q TF [ 1 + ∣ log ∣ r − 1∣∣ ] . Integrating from r r r to r ′ r' r ′ ,
∣ Σ x ( r ′ ) − Σ x ( r ) ∣ ≤ C b q T F 2 [ 2 s + s log ( 1 / s ) + ξ log ( s / ξ ) + ζ log ( s / ζ ) ] , r + r ′ ≥ 2 − q . |\Sigma_x(r')-\Sigma_x(r)|\le C_bq_{\mathrm{TF}}^2\left[2s+s\log(1/s)+\xi\log(s/\xi)+\zeta\log(s/\zeta)\right],\qquad r+r'\ge2-q. ∣ Σ x ( r ′ ) − Σ x ( r ) ∣ ≤ C b q TF 2 [ 2 s + s log ( 1/ s ) + ξ log ( s / ξ ) + ζ log ( s / ζ ) ] , r + r ′ ≥ 2 − q .
Combining this with ξ log ( s / ξ ) + ζ log ( s / ζ ) ≤ s log 2 \xi\log(s/\xi)+\zeta\log(s/\zeta)\le s\log2 ξ log ( s / ξ ) + ζ log ( s / ζ ) ≤ s log 2 gives
∣ Σ x ( r ′ ) − Σ x ( r ) ∣ ∣ δ p ∣ ≤ C b q T F 2 [ 1 + log ( 1 / s ) ] . (130) \frac{|\Sigma_x(r')-\Sigma_x(r)|}{|\delta_p|}\le C_bq_{\mathrm{TF}}^2[1+\log(1/s)]. \tag*{(130)} ∣ δ p ∣ ∣ Σ x ( r ′ ) − Σ x ( r ) ∣ ≤ C b q TF 2 [ 1 + log ( 1/ s )] . (130)
Including both transition regions and the response normalization gives
∫ 0 ∞ ∣ Π S E ( 1 ) ( q , i ν ) ∣ d ν ≤ C b q T F 4 q − 1 ∫ 0 q s [ 1 + log ( 1 / s ) ] d s = C b q T F 4 q 4 [ 3 + 2 log ( 1 / q ) ] . \int_0^\infty|\Pi_{\mathrm{SE}}^{(1)}(q,i\nu)|\,\mathrm{d}\nu\le C_bq_{\mathrm{TF}}^4q^{-1}\int_0^q s[1+\log(1/s)]\,\mathrm{d}s=C_bq_{\mathrm{TF}}^4\frac{q}{4}[3+2\log(1/q)]. ∫ 0 ∞ ∣ Π SE ( 1 ) ( q , i ν ) ∣ d ν ≤ C b q TF 4 q − 1 ∫ 0 q s [ 1 + log ( 1/ s )] d s = C b q TF 4 4 q [ 3 + 2 log ( 1/ q )] .
For q > 3 q>3 q > 3 , uniformly in q q q , every p \mathbf{p} p in the two transition regions satisfies
∣ δ p ∣ ≥ q 2 / 2 − q ≥ 3 / 2 , ∣ Σ x ∣ ≤ C b q T F 2 , |\delta_p|\ge q^2/2-q\ge3/2,\qquad|\Sigma_x|\le C_bq_{\mathrm{TF}}^2, ∣ δ p ∣ ≥ q 2 /2 − q ≥ 3/2 , ∣ Σ x ∣ ≤ C b q TF 2 ,
and the regions are already disjoint for q > 2 q>2 q > 2 . The small- and large-momentum estimates give
∫ 0 ∞ ∣ Π S E ( 1 ) ( q , i ν ) ∣ d ν = { O ( q T F 4 q log ( e / q ) ) , q → 0 , O ( q T F 4 q − 2 ) , q → ∞ . (131) \int_0^\infty|\Pi_{\mathrm{SE}}^{(1)}(q,i\nu)|\,\mathrm{d}\nu=
\begin{cases}
O(q_{\mathrm{TF}}^4q\log(e/q)), & q\to0,\\
O(q_{\mathrm{TF}}^4q^{-2}), & q\to\infty.
\end{cases} \tag*{(131)} ∫ 0 ∞ ∣ Π SE ( 1 ) ( q , i ν ) ∣ d ν = { O ( q TF 4 q log ( e / q )) , O ( q TF 4 q − 2 ) , q → 0 , q → ∞. (131)
On compact momentum intervals away from zero, the only remaining singularity is the integrable log ( 1 / s ) \log(1/s) log ( 1/ s ) at s = 0 s=0 s = 0 , including at q = 2 q=2 q = 2 . The complete bound is therefore integrable against d 3 q ν = 4 π d q \mathrm{d}^3q\,\nu=4\pi\,\mathrm{d}q d 3 q ν = 4 π d q . Combining it with the vertex bound gives the corresponding result for Π ( 1 ) \Pi^{(1)} Π ( 1 ) .
General spectral interchange and weighted remainders
The same argument applies to any functional linear in one selected line, provided its bare-line absolute integral is finite:
J [ f ] = ∫ f ( x ) f ( Q ( x ) ) d μ ( x ) , ∫ ∣ f ( x ) ∣ v ( q ( x ) ) d μ ( x ) < ∞ . (132) \mathcal{J}[f]=\int f(x)f(Q(x))\,\mathrm{d}\mu(x),\qquad\int|f(x)|v(q(x))\,\mathrm{d}\mu(x)<\infty. \tag*{(132)} J [ f ] = ∫ f ( x ) f ( Q ( x )) d μ ( x ) , ∫ ∣ f ( x ) ∣ v ( q ( x )) d μ ( x ) < ∞. (132)
Here x x x collects all integration variables, Q = ( q , i ν ) Q=(q,i\nu) Q = ( q , i ν ) is the momentum transfer carried by the selected line, and q ( x ) q(x) q ( x ) is its magnitude. The measure d μ ( x ) \mathrm{d}\mu(x) d μ ( x ) collects the integration domains and measures, and f f f includes the closed-diagram coefficient and the fixed unselected lines. The choice
x = ( q , ν ) , d μ = d 3 q d ν , f = Y D / π x=(q,\nu),\qquad\mathrm{d}\mu=\mathrm{d}^{3}q\,\mathrm{d}\nu,\qquad f=Y_D/\pi x = ( q , ν ) , d μ = d 3 q d ν , f = Y D / π
gives the condition stated after main-text Eq. (15) .
Majorant. Consider a line with a nonnegative weight a ( q , Λ ) a(q,\Lambda) a ( q , Λ ) , such as a λ / λ a_\lambda/\lambda a λ / λ of (S55), and assume
f = v − ∫ 0 ∞ a ( q , Λ ) b Λ d Λ , 0 ≤ ∫ 0 ∞ a b Λ d Λ ≤ v . f=v-\int_0^\infty a(q,\Lambda)b_\Lambda\,\mathrm{d}\Lambda,\qquad0\le\int_0^\infty ab_\Lambda\,\mathrm{d}\Lambda\le v. f = v − ∫ 0 ∞ a ( q , Λ ) b Λ d Λ , 0 ≤ ∫ 0 ∞ a b Λ d Λ ≤ v .
Tonelli’s theorem gives
∫ d Λ ∫ ∣ f ( x ) ∣ ∣ a ( q ( x ) , Λ ) b Λ ( Q ( x ) ) ∣ d μ ( x ) ≤ ∫ ∣ f ( x ) ∣ v ( q ( x ) ) d μ ( x ) . (133) \int\mathrm{d}\Lambda\int|f(x)|\,|a(q(x),\Lambda)b_{\Lambda}(Q(x))|\,\mathrm{d}\mu(x) \le\int|f(x)|\,v(q(x))\,\mathrm{d}\mu(x). \tag*{(133)} ∫ d Λ ∫ ∣ f ( x ) ∣ ∣ a ( q ( x ) , Λ ) b Λ ( Q ( x )) ∣ d μ ( x ) ≤ ∫ ∣ f ( x ) ∣ v ( q ( x )) d μ ( x ) . (133)
By (S71), the integrand f a b Λ f\,a\,b_{\Lambda} f a b Λ is absolutely integrable over ( x , Λ ) (x,\Lambda) ( x , Λ ) , and Fubini’s theorem gives
J [ f ] = J [ v ] − ∫ d Λ J [ a ( q , Λ ) b Λ ] . (134) \mathcal{J}[f]=\mathcal{J}[v]-\int\mathrm{d}\Lambda\,\mathcal{J}[a(q,\Lambda)b_{\Lambda}]. \tag*{(134)} J [ f ] = J [ v ] − ∫ d Λ J [ a ( q , Λ ) b Λ ] . (134)
Here J \mathcal{J} J acts at fixed Λ \Lambda Λ , as in (S56). The argument also covers positive spectral measures with point masses.
Weighted remainders. Inserting an asymptotic expansion into the resulting integrals also requires a bound on the integrated remainder. Write an integrand as
K η = K η a s y + R η , η → 0 , K_{\eta}=K_{\eta}^{\mathrm{asy}}+R_{\eta},\qquad\eta\to0, K η = K η asy + R η , η → 0 ,
with K η a s y K_{\eta}^{\mathrm{asy}} K η asy the retained terms of its asymptotic expansion and R η R_{\eta} R η the remainder, and let d ρ η \mathrm{d}\rho_{\eta} d ρ η include the momentum, spectral and coupling measures. With a nonnegative majorant M M M of the remainder, an error scale r ( η ) r(\eta) r ( η ) and a fixed η 0 > 0 \eta_{0}>0 η 0 > 0 , a sufficient condition for the remainder bound is
∣ R η ∣ ≤ r ( η ) M , sup 0 < η < η 0 ∫ M d ρ η < ∞ ⟹ ∣ ∫ R η d ρ η ∣ = O ( r ( η ) ) . (135) |R_{\eta}|\le r(\eta)M,\qquad\sup_{0<\eta<\eta_{0}}\int M\,\mathrm{d}\rho_{\eta}<\infty\quad\Longrightarrow\quad\left|\int R_{\eta}\,\mathrm{d}\rho_{\eta}\right|=O(r(\eta)). \tag*{(135)} ∣ R η ∣ ≤ r ( η ) M , 0 < η < η 0 sup ∫ M d ρ η < ∞ ⟹ ∫ R η d ρ η = O ( r ( η )) . (135)
This criterion can be applied separately to each momentum region, as in § 29.2 . The condition r ( η ) = o ( σ ( η ) ) r(\eta)=o(\sigma(\eta)) r ( η ) = o ( σ ( η )) makes the integrated remainder smaller than the retained scale σ ( η ) \sigma(\eta) σ ( η ) , so the expansion is controlled through that order.
Technical conditions and proofs for Sec. 2
This section collects the hypotheses and auxiliary steps behind main-text § 2.6 and § 2.7 .
Statement and proof of the endpoint–residue correspondence
The correspondence concerns the Mellin convolution of the main text and the Mellin transforms of its two factors,
I ( ε ) = ∫ 0 ∞ k ( q ) g ( q / ε ) d q , k ~ ( 1 − s ) = ∫ 0 ∞ q − s k ( q ) d q , g ~ ( s ) = ∫ 0 ∞ t s − 1 g ( t ) d t . I(\varepsilon)=\int_{0}^{\infty}k(q)\,g(q/\varepsilon)\,\mathrm{d}q,\qquad\widetilde{k}(1-s)=\int_{0}^{\infty}q^{-s}k(q)\,\mathrm{d}q,\qquad\widetilde{g}(s)=\int_{0}^{\infty}t^{s-1}g(t)\,\mathrm{d}t. I ( ε ) = ∫ 0 ∞ k ( q ) g ( q / ε ) d q , k ( 1 − s ) = ∫ 0 ∞ q − s k ( q ) d q , g ( s ) = ∫ 0 ∞ t s − 1 g ( t ) d t .
We follow López [24 ] . His method needs neither the Mellin–Parseval formula (23) nor the decay of the transforms along vertical lines. The contour shift of Refs. [25 , 26 ] rests on these two conditions. In the applications the decay h ( z ) ≃ 1 / ( 3 z 2 ) h(z)\simeq1/(3z^{2}) h ( z ) ≃ 1/ ( 3 z 2 ) at large z z z supplies the integrability required in condition (iii) below.
Proposition S3 (Endpoint–residue correspondence). Suppose the following. (i) The integral (22) converges absolutely, and k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) and g ~ ( s ) \widetilde{g}(s) g ( s ) converge absolutely in a nonempty common strip c − < Re s < c + c_{-}<\operatorname{Re}s<c_{+} c − < Re s < c + . The second part of (i) does not follow from the first. (ii) k k k admits finite endpoint expansions in terms of the form x a log m x x^{a}\log^{m}x x a log m x at 0 0 0 and at ∞ \infty ∞ , and g g g admits one in terms of the form x a x^{a} x a at the endpoint that generates the expansion (t → ∞ t\to\infty t → ∞ for ε → 0 \varepsilon\to0 ε → 0 and t → 0 t\to0 t → 0 for
ε → ∞ \varepsilon\to\infty ε → ∞ , with remainders of strictly lower order. For ε → 0 \varepsilon\to0 ε → 0 this means that for some σ > c + \sigma> c_{+} σ > c + , positive constants K K K and G G G , and p ≥ 0 p \ge0 p ≥ 0 ,
k ( q ) = ∑ 1 + α < σ ∑ m κ α , m q α log m ( 1 / q ) + k σ ( q ) , ∣ k σ ( q ) ∣ ≤ K q σ − 1 ( 1 + log ( 1 / q ) ) p ( 0 < q ≤ 1 ) , g ( t ) = ∑ β < σ γ β t − β + g σ ( t ) , ∣ g σ ( t ) ∣ ≤ G t − σ ( t ≥ 1 ) . (136) \begin{aligned}
k(q) &= \sum_{1+\alpha<\sigma}\sum_m \kappa_{\alpha,m}q^\alpha\log^m(1/q)+k_\sigma(q), \qquad|k_\sigma(q)|\le Kq^{\sigma-1}(1+\log(1/q))^p \qquad(0<q\le1),\\
g(t) &= \sum_{\beta<\sigma}\gamma_\beta t^{-\beta}+g_\sigma(t), \qquad|g_\sigma(t)|\le Gt^{-\sigma}\qquad(t\ge1). \tag*{(136)}
\end{aligned} k ( q ) g ( t ) = 1 + α < σ ∑ m ∑ κ α , m q α log m ( 1/ q ) + k σ ( q ) , ∣ k σ ( q ) ∣ ≤ K q σ − 1 ( 1 + log ( 1/ q ) ) p ( 0 < q ≤ 1 ) , = β < σ ∑ γ β t − β + g σ ( t ) , ∣ g σ ( t ) ∣ ≤ G t − σ ( t ≥ 1 ) . (136)
where the sums run over finitely many real exponents α \alpha α (with integers m ≥ 0 m\ge0 m ≥ 0 ) and β \beta β , κ α , m \kappa_{\alpha,m} κ α , m and γ β \gamma_\beta γ β are their coefficients, and k σ k_\sigma k σ , g σ g_\sigma g σ are the remainders. The substitutions q → 1 / q q\to1/q q → 1/ q and t → 1 / t t\to1/t t → 1/ t map ε → ∞ \varepsilon\to\infty ε → ∞ onto this case, with the poles to the left of the strip and the residues of opposite sign. (iii) When k k k and g g g depend on a parameter, such as z z z , that is integrated after the residues are taken, (i) and (ii) hold for almost every value of the parameter, with integrable coefficients and remainder constants. A small-q q q term q α log m ( 1 / q ) q^\alpha\log^m(1/q) q α log m ( 1/ q ) of k k k gives k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) a pole of order m + 1 m+1 m + 1 at s = 1 + α s=1+\alpha s = 1 + α , a large-q q q term q − α ′ q^{-\alpha'} q − α ′ a simple pole at s = 1 − α ′ s=1-\alpha' s = 1 − α ′ . A large-t t t term t − β t^{-\beta} t − β of g g g gives g ~ \widetilde{g} g a simple pole at s = β s=\beta s = β , a small-t t t term t β ′ t^{\beta'} t β ′ one at s = − β ′ s=-\beta' s = − β ′ . Then, as ε → 0 \varepsilon\to0 ε → 0 , I ( ε ) I(\varepsilon) I ( ε ) is minus the sum of the residues of k ~ ( 1 − s ) g ~ ( s ) ε s \widetilde{k}(1-s)\widetilde{g}(s)\varepsilon^s k ( 1 − s ) g ( s ) ε s at the poles to the right of the strip, in increasing order, and as ε → ∞ \varepsilon\to\infty ε → ∞ the sum of the residues at the poles to its left, in decreasing order. A pole of total order N N N at s 0 s_0 s 0 contributes ε s 0 \varepsilon^{s_0} ε s 0 times a polynomial of degree N − 1 N-1 N − 1 in log ε \log\varepsilon log ε . The remainder after the poles up to the line Re s = σ \operatorname{Re}s=\sigma Re s = σ is bounded by ε σ \varepsilon^\sigma ε σ times a power of 1 + ∣ log ε ∣ 1+|\log\varepsilon| 1 + ∣ log ε ∣ . It is bounded by a constant times ε σ \varepsilon^\sigma ε σ if the endpoint bounds extend beyond that line in the direction of the expansion with no further pole: to some σ ′ > σ \sigma' > \sigma σ ′ > σ with no pole in σ ≤ Re s < σ ′ \sigma\le\operatorname{Re}s<\sigma' σ ≤ Re s < σ ′ for ε → 0 \varepsilon\to0 ε → 0 , and to some σ ′ < σ \sigma'<\sigma σ ′ < σ with no pole in σ ′ < Re s ≤ σ \sigma'<\operatorname{Re}s\le\sigma σ ′ < Re s ≤ σ for ε → ∞ \varepsilon\to\infty ε → ∞ .
Proof. We prove the case ε → 0 \varepsilon\to0 ε → 0 in the form
I ( ε ) = − ∑ s 0 < σ Res s = s 0 [ k ~ ( 1 − s ) g ~ ( s ) ε s ] + R σ ( ε ) , I(\varepsilon)=-\sum_{s_0<\sigma}\operatorname{Res}_{s=s_0}\left[\widetilde{k}(1-s)\widetilde{g}(s)\varepsilon^s\right]+R_\sigma(\varepsilon), I ( ε ) = − s 0 < σ ∑ Res s = s 0 [ k ( 1 − s ) g ( s ) ε s ] + R σ ( ε ) ,
with the sum over the poles s 0 s_0 s 0 to the right of the strip and the remainder R σ R_\sigma R σ estimated at the end of the proof. Under (ii), k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) and g ~ ( s ) \widetilde{g}(s) g ( s ) continue meromorphically to c − < Re s < σ c_{-}<\operatorname{Re}s<\sigma c − < Re s < σ , with the stated poles Eqs. (9) and (10) of Ref. [24 ] . López subtracts the endpoint terms of the two factors one at a time, in the order of their poles. Each subtraction leaves a convergent integral equal to one term of the expansion Theorem 1 of Ref. [24 ] .
One factor. When only one factor has a term at a pole s 0 s_0 s 0 , that term contributes
− Res s = s 0 k ~ ( 1 − s ) g ~ ( s ) ε s , q α log m ( 1 / q ) = ( − ∂ α ) m q α . -\operatorname{Res}_{s=s_0}\widetilde{k}(1-s)\widetilde{g}(s)\varepsilon^s,\qquad q^\alpha\log^m(1/q)=(-\partial_\alpha)^m q^\alpha. − Res s = s 0 k ( 1 − s ) g ( s ) ε s , q α log m ( 1/ q ) = ( − ∂ α ) m q α .
The second relation reduces a term q α log m ( 1 / q ) q^\alpha\log^m(1/q) q α log m ( 1/ q ) to the pure-power case by m m m differentiations with respect to α \alpha α , which may be taken under the integral inside the strip.
Both factors. When both factors have terms at s 0 s_0 s 0 , López inserts a regulator q − z q^{-z} q − z and takes the limit z → 0 z\to0 z → 0 by dominated convergence. For two simple poles this limit gives the term − a k b j log x -a_kb_j\log x − a k b j log x of his (19) , with x = ε x=\varepsilon x = ε , and the same regulator applies to poles of higher order. The two regulated terms are minus the residues of F z ( s ) = k ~ ( 1 − s − z ) g ~ ( s ) ε s F_z(s)=\widetilde{k}(1-s-z)\widetilde{g}(s)\varepsilon^s F z ( s ) = k ( 1 − s − z ) g ( s ) ε s at s 0 − z s_0-z s 0 − z and at s 0 s_0 s 0 . With C C C a small circle about s 0 s_0 s 0 ,
− [ Res s = s 0 − z + Res s = s 0 ] F z ( s ) = − 1 2 π i ∮ C F z ( s ) d s → z → 0 − Res s = s 0 k ~ ( 1 − s ) g ~ ( s ) ε s . -\left[\operatorname{Res}_{s=s_0-z}+\operatorname{Res}_{s=s_0}\right]F_z(s)=-\frac{1}{2\pi i}\oint_C F_z(s)\,ds\xrightarrow[z\to0]{}-\operatorname{Res}_{s=s_0}\widetilde{k}(1-s)\widetilde{g}(s)\varepsilon^s. − [ Res s = s 0 − z + Res s = s 0 ] F z ( s ) = − 2 π i 1 ∮ C F z ( s ) d s z → 0 − Res s = s 0 k ( 1 − s ) g ( s ) ε s .
Remainder. After the term of the last pole is subtracted,
R σ ( ε ) = ∫ 0 ∞ k σ ( q ) g σ ( q / ε ) d q , R_\sigma(\varepsilon)=\int_0^\infty k_\sigma(q)\,g_\sigma(q/\varepsilon)\,dq, R σ ( ε ) = ∫ 0 ∞ k σ ( q ) g σ ( q / ε ) d q ,
where k σ k_\sigma k σ and g σ g_\sigma g σ extend to the whole half-line by the subtractions of (S74). On q < ε q < \varepsilon q < ε , where t = q / ε < 1 t = q/\varepsilon< 1 t = q / ε < 1 , and on q > 1 q > 1 q > 1 ,
∣ g σ ( t ) ∣ ≤ ∣ g ( t ) ∣ + ∑ β < σ ∣ γ β ∣ t − β ( t < 1 ) , ∣ k σ ( q ) ∣ ≤ ∣ k ( q ) ∣ + ∑ ∣ κ α , m ∣ q α log m q ( q > 1 ) . |g_\sigma(t)| \le|g(t)| + \sum_{\beta<\sigma} |\gamma_\beta|t^{-\beta} \quad(t<1), \qquad|k_\sigma(q)| \le|k(q)| + \sum| \kappa_{\alpha,m}|q^\alpha\log^m q \quad(q>1). ∣ g σ ( t ) ∣ ≤ ∣ g ( t ) ∣ + β < σ ∑ ∣ γ β ∣ t − β ( t < 1 ) , ∣ k σ ( q ) ∣ ≤ ∣ k ( q ) ∣ + ∑ ∣ κ α , m ∣ q α log m q ( q > 1 ) .
Condition (i), σ > c + \sigma> c_+ σ > c + and the strict inequalities β < σ \beta< \sigma β < σ and 1 + α < σ 1 + \alpha< \sigma 1 + α < σ make the contributions of these two ranges O ( ε σ [ 1 + log ( 1 / ε ) ] p ) O(\varepsilon^\sigma[1+\log(1/\varepsilon)]^p) O ( ε σ [ 1 + log ( 1/ ε ) ] p ) and O ( ε σ ) O(\varepsilon^\sigma) O ( ε σ ) , respectively. The range ε < q < 1 \varepsilon< q < 1 ε < q < 1 , where both bounds of (S74) hold, gives one more power of 1 + log ( 1 / ε ) 1+\log(1/\varepsilon) 1 + log ( 1/ ε ) ,
∫ ε 1 ∣ k σ ( q ) g σ ( q / ε ) ∣ d q ≤ K G ε σ ∫ ε 1 ( 1 + log ( 1 / q ) ) p d q q ≤ K G p + 1 ε σ [ 1 + log ( 1 / ε ) ] p + 1 . \int_\varepsilon^1 |k_\sigma(q)g_\sigma(q/\varepsilon)|\,dq \le KG\varepsilon^\sigma\int_\varepsilon^1 (1+\log(1/q))^p\frac{dq}{q} \le\frac{KG}{p+1}\varepsilon^\sigma[1+\log(1/\varepsilon)]^{p+1}. ∫ ε 1 ∣ k σ ( q ) g σ ( q / ε ) ∣ d q ≤ K G ε σ ∫ ε 1 ( 1 + log ( 1/ q ) ) p q d q ≤ p + 1 K G ε σ [ 1 + log ( 1/ ε ) ] p + 1 .
Splitting R σ ( ε ) R_\sigma(\varepsilon) R σ ( ε ) at q = ε q=\varepsilon q = ε and at q = 1 q=1 q = 1 therefore gives, with C b C_b C b independent of ε \varepsilon ε ,
∣ R σ ( ε ) ∣ ≤ C b ε σ [ 1 + log ( 1 / ε ) ] p + 1 ( 0 < ε ≤ 1 ) . |R_\sigma(\varepsilon)| \le C_b\varepsilon^\sigma[1+\log(1/\varepsilon)]^{p+1}\qquad(0<\varepsilon\le1). ∣ R σ ( ε ) ∣ ≤ C b ε σ [ 1 + log ( 1/ ε ) ] p + 1 ( 0 < ε ≤ 1 ) .
Sharper bound. If condition (ii) also holds for some σ ′ \sigma' σ ′ with
ε → 0 : σ ′ > σ , no pole in σ ≤ Re s < σ ′ , ε → ∞ : σ ′ < σ , no pole in σ ′ < Re s ≤ σ , \varepsilon\to0:\ \sigma'>\sigma,\ \text{no pole in }\sigma\le\operatorname{Re}s<\sigma', \qquad\varepsilon\to\infty:\ \sigma'<\sigma,\ \text{no pole in }\sigma'<\operatorname{Re}s\le\sigma, ε → 0 : σ ′ > σ , no pole in σ ≤ Re s < σ ′ , ε → ∞ : σ ′ < σ , no pole in σ ′ < Re s ≤ σ ,
the same estimate with σ ′ \sigma' σ ′ gives R σ = O ( ε σ ′ ) R_\sigma=O(\varepsilon^{\sigma'}) R σ = O ( ε σ ′ ) . The condition for ε → ∞ \varepsilon\to\infty ε → ∞ follows because the inversion reverses the direction of the expansion. These bounds are analogous to Theorem 2 and Eqs. (27)–(32) of Ref. [24 ] .
Proofs of Corollaries 1 and 2
Proof of Corollary 1. Both statements apply Proposition S3 with the scaling functions of main-text Sec. 2.6. Below, P j P_j P j denotes a polynomial of degree j j j .
(1) At high density ε = q T F → 0 \varepsilon=q_{\mathrm{TF}}\to0 ε = q TF → 0 , and the poles to the right of the strip contribute. The first pole of k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) there and the poles of g ~ ( s ) \widetilde{g}(s) g ( s ) from the Thomas–Fermi tail t − N ( d − 1 ) t^{-N(d-1)} t − N ( d − 1 ) , N ≥ 1 N\ge1 N ≥ 1 , of the scaling function are
k ~ ( 1 − s ) : s = 1 + α 0 , order m + 1 [Eq. (S77)] , g ~ ( s ) : s = N ( d − 1 ) , simple . \widetilde{k}(1-s):\ s=1+\alpha_0,\ \text{order }m+1\ \text{[Eq. (S77)]},\qquad\widetilde{g}(s):\ s=N(d-1),\ \text{simple}. k ( 1 − s ) : s = 1 + α 0 , order m + 1 [Eq. (S77)] , g ( s ) : s = N ( d − 1 ) , simple .
The leading term is the residue at the smaller of 1 + α 0 1+\alpha_0 1 + α 0 and d − 1 d-1 d − 1 and is of order q T F min ( 1 + α 0 , d − 1 ) q_{\mathrm{TF}}^{\min(1+\alpha_0,d-1)} q TF m i n ( 1 + α 0 , d − 1 ) , up to powers of log q T F \log q_{\mathrm{TF}} log q TF . For d = 3 d=3 d = 3 and α 0 = 1 \alpha_0=1 α 0 = 1 the two poles coincide,
1 + α 0 = d − 1 = 2 : total order m + 2 , residue = q T F 2 P m + 1 ( log q T F ) . 1+\alpha_0=d-1=2:\qquad\text{total order }m+2,\qquad\text{residue}=q_{\mathrm{TF}}^2P_{m+1}(\log q_{\mathrm{TF}}). 1 + α 0 = d − 1 = 2 : total order m + 2 , residue = q TF 2 P m + 1 ( log q TF ) .
(2) At low density ε = q ∗ → ∞ \varepsilon=q_*\to\infty ε = q ∗ → ∞ , and the poles to the left of the strip contribute. The large-q q q term q − α q^{-\alpha} q − α of the kernel and the large-width term of the scaling function, proportional to t 2 t^2 t 2 , give simple poles,
q − α ⟶ s = 1 − α , t 2 ⟶ s = − 2. q^{-\alpha}\longrightarrow s=1-\alpha,\qquad t^2\longrightarrow s=-2. q − α ⟶ s = 1 − α , t 2 ⟶ s = − 2.
The leading term is the residue at max ( 1 − α , − 2 ) \max(1-\alpha,-2) max ( 1 − α , − 2 ) and is of order q ∗ max ( 1 − α , − 2 ) = q ∗ − min ( α − 1 , 2 ) q_*^{\max(1-\alpha,-2)}=q_*^{-\min(\alpha-1,2)} q ∗ m a x ( 1 − α , − 2 ) = q ∗ − m i n ( α − 1 , 2 ) . For α = 3 \alpha=3 α = 3 the two poles coincide, and the double pole gives the factor log q ∗ \log q_* log q ∗ . For α > 3 \alpha>3 α > 3 the leading pole is at s = − 2 s=-2 s = − 2 , and Sec. S14.3 shows that its residue is main-text Eq. (25) under the conditions stated in the corollary. These conditions also justify the separation of the large-q q q kernel that the width requires at low density (main-text Sec. 2.7).
Proof of Corollary 2. The density enters through the prefactor of Y D = r s n − 2 Y ^ D Y_D = r_s^{n-2}\widehat{Y}_D Y D = r s n − 2 Y D and through q T F q_{\rm TF} q TF in the scaling function (main-text § 2.3 ). The kernel term q log m ( 1 / q ) q\log^m(1/q) q log m ( 1/ q ) gives k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) a pole of order m + 1 m+1 m + 1 at s = 2 s=2 s = 2 . With P j P_j P j as above, its residue is
2 ≠ N ( d − 1 ) for every N : q T F 2 P m ( log q T F ) , 2 = N ∗ ( d − 1 ) : q T F 2 P m + 1 ( log q T F ) . \begin{aligned}
2 \ne N(d-1)\ \text{for every }N:\quad&q_{\rm TF}^2 P_m(\log q_{\rm TF}),\\
2 = N_*(d-1):\quad&q_{\rm TF}^2 P_{m+1}(\log q_{\rm TF}).
\end{aligned} 2 = N ( d − 1 ) for every N : 2 = N ∗ ( d − 1 ) : q TF 2 P m ( log q TF ) , q TF 2 P m + 1 ( log q TF ) .
since the screening pole N = N ∗ N=N_* N = N ∗ adds one order in the second case. With q T F d − 1 = a T F r s q_{\rm TF}^{d-1}=a_{\rm TF}r_s q TF d − 1 = a TF r s ,
q T F 2 ∝ r s 2 / ( d − 1 ) , log q T F = log r s / ( d − 1 ) + O ( 1 ) , q_{\rm TF}^2 \propto r_s^{2/(d-1)},\qquad\log q_{\rm TF}=\log r_s/(d-1)+O(1), q TF 2 ∝ r s 2/ ( d − 1 ) , log q TF = log r s / ( d − 1 ) + O ( 1 ) ,
and with the prefactor these residues give the terms stated in the corollary. The poles s = N ( d − 1 ) < 2 s=N(d-1)<2 s = N ( d − 1 ) < 2 are simple, because the kernel transform is regular there, and give q T F N ( d − 1 ) ∝ r s N q_{\rm TF}^{N(d-1)}\propto r_s^N q TF N ( d − 1 ) ∝ r s N . For the density-independent kernel 4 π q Y ^ D ( q , q z ) 4\pi q\widehat{Y}_D(q,qz) 4 π q Y D ( q , q z ) , with κ D \kappa_D κ D and γ N ∗ \gamma_{N_*} γ N ∗ of Corollary 2, Eq. (140) gives
q T F 2 log m + 1 q T F ⋅ ( − 1 ) m + 1 κ D ( z ) γ N ∗ ( z ) m + 1 . q_{\rm TF}^2\log^{m+1}q_{\rm TF}\cdot\frac{(-1)^{m+1}\kappa_D(z)\gamma_{N_*}(z)}{m+1}. q TF 2 log m + 1 q TF ⋅ m + 1 ( − 1 ) m + 1 κ D ( z ) γ N ∗ ( z ) .
and the coefficient of the corollary is therefore proportional to ∫ 0 ∞ κ D γ N ∗ d z / π \int_0^\infty\kappa_D\gamma_{N_*}\,dz/\pi ∫ 0 ∞ κ D γ N ∗ d z / π . Replacing L ( q , q z ) \mathcal{L}(q,qz) L ( q , q z ) by h ( z ) h(z) h ( z ) in the scaling function changes c c c by
O ( q T F 2 q ( 1 + z 2 ) ) ( q < 1 , d = 3 ) [ x r e f ‘ E q . ( S 88 ) ‘ ] , O ( q T F q ) ( d = 2 ) [ x r e f ‘ E q . ( S 4 ) ‘ ] . O\left(\frac{q_{\rm TF}^2}{q(1+z^2)}\right)\quad(q<1,\ d=3)\ [{xref}`Eq. (S88)`],\qquad O(q_{\rm TF}q)\quad(d=2)\ [{xref}`Eq. (S4)`]. O ( q ( 1 + z 2 ) q TF 2 ) ( q < 1 , d = 3 ) [ x r e f ‘ E q . ( S 88 ) ‘ ] , O ( q TF q ) ( d = 2 ) [ x r e f ‘ E q . ( S 4 ) ‘ ] .
Since the scaling functions of main-text table 1 have bounded derivatives in c c c , this replacement changes the screening correction by O ( q T F 2 ) O(q_{\rm TF}^2) O ( q TF 2 ) , which lies within the remainder O ( q T F 2 ∣ log q T F ∣ m ) O(q_{\rm TF}^2|\log q_{\rm TF}|^m) O ( q TF 2 ∣ log q TF ∣ m ) , as in Sec. S14.6. For q > 1 q>1 q > 1 , where c = O ( q T F 2 / q 2 ) c=O(q_{\rm TF}^2/q^2) c = O ( q TF 2 / q 2 ) , both forms of g g g contribute O ( q T F 2 ) O(q_{\rm TF}^2) O ( q TF 2 ) . If the factors log ( 1 / q ) \log(1/q) log ( 1/ q ) of the kernels cancel in a sum of diagrams, the summed kernel has a smaller m m m , which lowers the order of its pole at s = 2 s=2 s = 2 and the power of log q T F \log q_{\rm TF} log q TF in the resonant term. □
Expansion of the radial kernel in 1 / Λ 1/\Lambda 1/Λ and the low-density pole
The second statement of Corollary 1 uses the expansion of the radial kernel (16) in 1 / Λ 1/\Lambda 1/Λ . If Y D Y_D Y D has a high-frequency tail A D ( q ) / ν 2 A_D(q)/\nu^2 A D ( q ) / ν 2 , then, formally at fixed internal momenta, the radial kernel of the Lorentzian line of width Λ \Lambda Λ differs from that of the bare line by
e D , Λ ( q ) − e D , ∞ ( q ) = − 2 π A D ( q ) q Λ + O ( ( q Λ ) − 2 ) , Λ → ∞ . (137) e_{D,\Lambda}(q)-e_{D,\infty}(q)=-\frac{2\pi A_D(q)}{q\Lambda}+O\left((q\Lambda)^{-2}\right),\qquad\Lambda\to\infty. \tag*{(137)} e D , Λ ( q ) − e D , ∞ ( q ) = − q Λ 2 π A D ( q ) + O ( ( q Λ ) − 2 ) , Λ → ∞. (137)
Derivation. By (16),
e D , Λ − e D , ∞ = − 4 π ∫ 0 ∞ d ν π ν 2 / q 2 Λ 2 + ν 2 / q 2 Y D ( q , ν ) . e_{D,\Lambda}-e_{D,\infty}=-4\pi\int_0^\infty\frac{d\nu}{\pi}\frac{\nu^2/q^2}{\Lambda^2+\nu^2/q^2}Y_D(q,\nu). e D , Λ − e D , ∞ = − 4 π ∫ 0 ∞ π d ν Λ 2 + ν 2 / q 2 ν 2 / q 2 Y D ( q , ν ) .
For Λ → ∞ \Lambda\to\infty Λ → ∞ the factor is small unless ν ≳ q Λ \nu\gtrsim q\Lambda ν ≳ q Λ , where Y D ≃ A D / ν 2 Y_D\simeq A_D/\nu^2 Y D ≃ A D / ν 2 . Inserting this tail, and bounding the rest by ν 2 / ( q 2 Λ 2 + ν 2 ) ≤ ν 2 / ( q Λ ) 2 \nu^2/(q^2\Lambda^2+\nu^2)\leq\nu^2/(q\Lambda)^2 ν 2 / ( q 2 Λ 2 + ν 2 ) ≤ ν 2 / ( q Λ ) 2 ,
tail = − 4 π A D q 2 ∫ 0 ∞ d ν π 1 Λ 2 + ν 2 / q 2 = − 2 π A D q Λ , \text{tail}=-\frac{4\pi A_D}{q^2}\int_0^\infty\frac{d\nu}{\pi}\frac{1}{\Lambda^2+\nu^2/q^2}=-\frac{2\pi A_D}{q\Lambda}, tail = − q 2 4 π A D ∫ 0 ∞ π d ν Λ 2 + ν 2 / q 2 1 = − q Λ 2 π A D ,
∣ remainder ∣ ≤ 4 ( q Λ ) 2 ∫ 0 ∞ d ν ν 2 ∣ Y D − A D ν 2 ∣ . |\text{remainder}|\leq\frac{4}{(q\Lambda)^2}\int_0^\infty d\nu\,\nu^2\left|Y_D-\frac{A_D}{\nu^2}\right|. ∣ remainder ∣ ≤ ( q Λ ) 2 4 ∫ 0 ∞ d ν ν 2 Y D − ν 2 A D .
This bound is finite when
∫ 0 ∞ ∣ Y D ∣ d ν < ∞ , Y D − A D ν 2 = O ( ν − 4 ) ( ν → ∞ ) . \int_0^\infty|Y_D|\,d\nu<\infty,\qquad Y_D-\frac{A_D}{\nu^2}=O(\nu^{-4})\quad(\nu\to\infty). ∫ 0 ∞ ∣ Y D ∣ d ν < ∞ , Y D − ν 2 A D = O ( ν − 4 ) ( ν → ∞ ) .
For the first-order exchange integrands the first condition follows from the absolute bounds of § 24.1 and § 24.3 , and the second holds because their frequency factors are even in ν \nu ν . For the kite, (41) at large ν \nu ν gives Y D = A 0 X ≃ − A 0 Z ( q ) / ν 2 Y_D=A_0X\simeq-A_0Z(q)/\nu^2 Y D = A 0 X ≃ − A 0 Z ( q ) / ν 2 with Z Z Z of (S36), and the self-energy pair has the opposite tail (Sec. S10.2).
Residue at s = − 2 s=-2 s = − 2 . With the single Lorentzian (21), and up to the contribution of ς λ \varsigma_\lambda ς λ bounded in the next paragraph,
J D [ W λ ] = λ ∫ 0 ∞ [ ( 1 − θ λ ) e D , ∞ + θ λ ( e D , ∞ − e D , Λ λ ) ] d q . \mathcal{J}_D[W_\lambda]=\lambda\int_0^\infty\left[(1-\theta_\lambda)e_{D,\infty}+\theta_\lambda(e_{D,\infty}-e_{D,\Lambda_\lambda})\right]\mathrm{d}q. J D [ W λ ] = λ ∫ 0 ∞ [ ( 1 − θ λ ) e D , ∞ + θ λ ( e D , ∞ − e D , Λ λ ) ] d q .
By the leading term of (S75), the second term in the brackets gives ∫ 0 ∞ k ( q ) g ( q / q ∗ ) d q \int_0^\infty k(q)\,g(q/q_*)\,\mathrm{d}q ∫ 0 ∞ k ( q ) g ( q / q ∗ ) d q with the kernel and the scaling function
k ( q ) = 2 π A D ( q ) q 2 , g ( t ) = θ λ q Λ λ = 2 λ t 2 ( t 4 + λ ) 3 / 2 ≃ 2 t 2 λ ( t → 0 ) . k(q)=\frac{2\pi A_D(q)}{q^2},\qquad g(t)=\frac{\theta_\lambda q}{\Lambda_\lambda}=\frac{2\lambda t^2}{(t^4+\lambda)^{3/2}}\simeq\frac{2t^2}{\sqrt{\lambda}}\quad(t\to0). k ( q ) = q 2 2 π A D ( q ) , g ( t ) = Λ λ θ λ q = ( t 4 + λ ) 3/2 2 λ t 2 ≃ λ 2 t 2 ( t → 0 ) .
The scaling function depends on q q q only through t = q / q ∗ t=q/q_* t = q / q ∗ . Its small-t t t term gives the simple pole, and k ~ ( 1 − s ) \widetilde{k}(1-s) k ( 1 − s ) is finite there by the first condition of the corollary,
g ~ ( s ) ≃ 2 / λ s + 2 , k ~ ( 1 − s ) ∣ s = − 2 = 2 π ∫ 0 ∞ A D d q . \widetilde{g}(s)\simeq\frac{2/\sqrt{\lambda}}{s+2},\qquad\left.\widetilde{k}(1-s)\right|_{s=-2}=2\pi\int_0^\infty A_D\,\mathrm{d}q. g ( s ) ≃ s + 2 2/ λ , k ( 1 − s ) s = − 2 = 2 π ∫ 0 ∞ A D d q .
By Proposition S3 with ϵ = q ∗ → ∞ \epsilon=q_*\to\infty ϵ = q ∗ → ∞ , this pole contributes, with the prefactor λ \lambda λ ,
λ Res s = − 2 [ k ~ ( 1 − s ) g ~ ( s ) q ∗ s ] = λ ⋅ 4 π λ 1 / 2 q ∗ 2 ∫ 0 ∞ A D d q = 4 π λ 1 / 2 q ∗ 2 ∫ 0 ∞ A D d q . \lambda\underset{s=-2}{\operatorname{Res}}\left[\widetilde{k}(1-s)\widetilde{g}(s)q_*^s\right]=\lambda\cdot\frac{4\pi}{\lambda^{1/2}q_*^2}\int_0^\infty A_D\,\mathrm{d}q=\frac{4\pi\lambda^{1/2}}{q_*^2}\int_0^\infty A_D\,\mathrm{d}q. λ s = − 2 Res [ k ( 1 − s ) g ( s ) q ∗ s ] = λ ⋅ λ 1/2 q ∗ 2 4 π ∫ 0 ∞ A D d q = q ∗ 2 4 π λ 1/2 ∫ 0 ∞ A D d q .
which at λ = 1 \lambda=1 λ = 1 is main-text Eq. (25) . The factor 1 − θ λ 1-\theta_\lambda 1 − θ λ and the remainder of (S75), which carries ( q / Λ λ ) 2 (q/\Lambda_\lambda)^2 ( q / Λ λ ) 2 , are of order t 4 t^4 t 4 at small t t t ,
1 − θ λ = t 4 t 4 + λ , ( q Λ λ ) 2 ≃ 4 t 4 λ , 1-\theta_\lambda=\frac{t^4}{t^4+\lambda},\qquad\left(\frac{q}{\Lambda_\lambda}\right)^2\simeq\frac{4t^4}{\lambda}, 1 − θ λ = t 4 + λ t 4 , ( Λ λ q ) 2 ≃ λ 4 t 4 ,
two powers beyond the pole term, and are controlled by the remainder condition of the corollary. The term with 1 − θ λ 1-\theta_\lambda 1 − θ λ is o ( q ∗ − 2 ) o(q_*^{-2}) o ( q ∗ − 2 ) also for q ≳ q ∗ q\gtrsim q_* q ≳ q ∗ when the bare-line kernel e D , ∞ e_{D,\infty} e D , ∞ decays faster than q − 3 q^{-3} q − 3 . This holds for the first-order exchange diagrams, since the bare-line radial kernel of the kite decays as q − 4 q^{-4} q − 4 (main-text § 2.7 and Sec. S17.1), and that of the self-energy pair vanishes for every q q q by Eq. (146) .
The remainder ς λ \varsigma_\lambda ς λ . The pole at s = − 2 s=-2 s = − 2 collects ∫ A D d q \int A_D\,\mathrm{d}q ∫ A D d q from all q ≪ q ∗ q\ll q_* q ≪ q ∗ , including q ≲ 1 q\lesssim1 q ≲ 1 , where the bound on ς λ \varsigma_\lambda ς λ in main-text (21) is not small. Fix q 0 ≥ 3 q_0\ge3 q 0 ≥ 3 . For q ≤ q 0 q\le q_0 q ≤ q 0 we compare the screened line with its single-Lorentzian form directly, using the representation of L \mathcal{L} L behind (167) , valid for every q q q . The splitting
1 δ p 2 + ν 2 = 1 ν 2 − δ p 2 ν 2 ( δ p 2 + ν 2 ) , \frac{1}{\delta_p^2+\nu^2}=\frac{1}{\nu^2}-\frac{\delta_p^2}{\nu^2(\delta_p^2+\nu^2)}, δ p 2 + ν 2 1 = ν 2 1 − ν 2 ( δ p 2 + ν 2 ) δ p 2 ,
the free f f f -sum identity of § 28.4 and ∣ δ p ∣ ≤ q + q 2 / 2 |\delta_p|\le q+q^2/2 ∣ δ p ∣ ≤ q + q 2 /2 give, with ω p 2 = q T F 2 / 3 \omega_p^2=q_{\mathrm{TF}}^2/3 ω p 2 = q TF 2 /3 of main-text Eq. (20) ,
L ( q , ν ) = q 2 3 ν 2 [ 1 + O ( q 2 ( 1 + q ) 2 ν 2 ) ] , c = ω p 2 ν 2 [ 1 + O ( q 2 ( 1 + q ) 2 ν 2 ) ] . \mathcal{L}(q,\nu)=\frac{q^2}{3\nu^2}\left[1+O\left(\frac{q^2(1+q)^2}{\nu^2}\right)\right],\qquad c=\frac{\omega_p^2}{\nu^2}\left[1+O\left(\frac{q^2(1+q)^2}{\nu^2}\right)\right]. L ( q , ν ) = 3 ν 2 q 2 [ 1 + O ( ν 2 q 2 ( 1 + q ) 2 ) ] , c = ν 2 ω p 2 [ 1 + O ( ν 2 q 2 ( 1 + q ) 2 ) ] .
The single-Lorentzian line of main-text (21) is
λ ν ( ν 2 + q 4 / 4 ) ν 2 + q 4 / 4 + λ ω p 2 = λ ν 2 ν 2 + λ ω p 2 + ν O ( q 4 ω p 2 ) . \frac{\lambda\nu(\nu^2+q^4/4)}{\nu^2+q^4/4+\lambda\omega_p^2}=\frac{\lambda\nu^2}{\nu^2+\lambda\omega_p^2}+\nu\,O\left(\frac{q^4}{\omega_p^2}\right). ν 2 + q 4 /4 + λ ω p 2 λ ν ( ν 2 + q 4 /4 ) = ν 2 + λ ω p 2 λ ν 2 + ν O ( ω p 2 q 4 ) .
and the correction lies within the bound below. For ν > q ( 1 + q / 2 ) \nu> q(1+q/2) ν > q ( 1 + q /2 ) one has δ p 2 < ν 2 \delta_p^2 < \nu^2 δ p 2 < ν 2 on the transition regions, and therefore c c c lies between ω p 2 / ( 2 ν 2 ) \omega_p^2/(2\nu^2) ω p 2 / ( 2 ν 2 ) and ω p 2 / ν 2 \omega_p^2/\nu^2 ω p 2 / ν 2 . Since λ c / ( 1 + λ c ) 2 ≤ min ( λ c , 1 / ( λ c ) ) \lambda c/(1+\lambda c)^2 \le\min(\lambda c,1/(\lambda c)) λ c / ( 1 + λ c ) 2 ≤ min ( λ c , 1/ ( λ c )) , the two lines then differ by v O ( q 2 ( 1 + q ) 2 / ω p 2 ) v O(q^2(1+q)^2/\omega_p^2) v O ( q 2 ( 1 + q ) 2 / ω p 2 ) . For ν ≤ q ( 1 + q / 2 ) \nu\le q(1+q/2) ν ≤ q ( 1 + q /2 ) both lines are v O ( q 2 ( 1 + q ) 2 / q T F 2 ) v O(q^2(1+q)^2/q_{\mathrm{TF}}^2) v O ( q 2 ( 1 + q ) 2 / q TF 2 ) , the exact line by Eq. (167) and the Lorentzian one because it is at most v ( ν 2 + q 4 / 4 ) / ω p 2 v(\nu^2+q^4/4)/\omega_p^2 v ( ν 2 + q 4 /4 ) / ω p 2 . If
∫ 0 q 0 q 2 ( 1 + q ) 2 ∫ 0 ∞ ∣ Y D ∣ d ν d q < ∞ \int_0^{q_0} q^2(1+q)^2 \int_0^\infty|Y_D|\,\mathrm{d}\nu\,\mathrm{d}q < \infty ∫ 0 q 0 q 2 ( 1 + q ) 2 ∫ 0 ∞ ∣ Y D ∣ d ν d q < ∞
(for the first-order exchange integrands by the bounds of § 24.1 and § 24.3 ), both differences, integrated against ∣ Y D ∣ |Y_D| ∣ Y D ∣ , contribute O ( q T F 2 ) = O ( q ∗ − 4 ) O(q_{\mathrm{TF}}^2)=O(q_*^{-4}) O ( q TF 2 ) = O ( q ∗ − 4 ) from q ≤ q 0 q\le q_0 q ≤ q 0 , with constants depending on q 0 q_0 q 0 . For q > q 0 q>q_0 q > q 0 we use a bound
∣ Y D ( q , ν ) ∣ ≤ a D ( q ) ν 2 + q 4 , ∫ q 0 ∞ a D q − 2 d q < ∞ . |Y_D(q,\nu)|\le\frac{a_D(q)}{\nu^2+q^4},\qquad\int_{q_0}^{\infty}a_Dq^{-2}\,\mathrm{d}q<\infty. ∣ Y D ( q , ν ) ∣ ≤ ν 2 + q 4 a D ( q ) , ∫ q 0 ∞ a D q − 2 d q < ∞.
For the first-order exchange integrands this bound holds with a D a_D a D bounded for q ≥ 3 q\ge3 q ≥ 3 , where q 2 / 2 − q ≥ q 2 / 6 q^2/2-q\ge q^2/6 q 2 /2 − q ≥ q 2 /6 , by Eq. (171) and the bound on Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) used in § 28.4 . With u = ν / q 2 u=\nu/q^2 u = ν / q 2 , t = q / q ∗ t=q/q_* t = q / q ∗ and c ^ \hat c c ^ of main-text Eq. (21) ,
∫ 0 ∞ d u 1 + u 2 λ c ^ ( 1 + λ c ^ ) 2 ≤ π 2 t 2 λ , q − 4 t 2 = q − 2 q ∗ − 2 . \int_0^\infty\frac{\mathrm{d}u}{1+u^2}\frac{\lambda\hat c}{(1+\lambda\hat c)^2}\le\frac{\pi}{2}\frac{t^2}{\sqrt{\lambda}},\qquad q^{-4}t^2=q^{-2}q_*^{-2}. ∫ 0 ∞ 1 + u 2 d u ( 1 + λ c ^ ) 2 λ c ^ ≤ 2 π λ t 2 , q − 4 t 2 = q − 2 q ∗ − 2 .
In this range the error bound therefore contributes at most a constant times q ∗ − 2 ∫ q 0 ∞ a D q − 2 d q q_*^{-2}\int_{q_0}^{\infty}a_Dq^{-2}\,\mathrm{d}q q ∗ − 2 ∫ q 0 ∞ a D q − 2 d q , and the integral vanishes as q 0 → ∞ q_0\to\infty q 0 → ∞ . Hence ς λ \varsigma_\lambda ς λ does not change the coefficient of q ∗ − 2 q_*^{-2} q ∗ − 2 , that is, the residue at s = − 2 s=-2 s = − 2 .
Conditions and the kite. The conditions of the corollary are that ∫ 0 ∞ A D d q \int_0^\infty A_D\,\mathrm{d}q ∫ 0 ∞ A D d q converges and that the remainder in (S75), including regions with large internal momentum transfers, is integrable uniformly in q ∗ q_* q ∗ after the internal integrations. For the kite the first condition fails. Here A D ( q ) = − A 0 Z ( q ) A_D(q)=-A_0Z(q) A D ( q ) = − A 0 Z ( q ) . Each of the two same-ball regions of Sec. S17.1 gives 4 π 2 4\pi^2 4 π 2 and the different-ball regions are O ( q − 2 ) O(q^{-2}) O ( q − 2 ) , hence, as q → ∞ q\to\infty q → ∞ ,
Z ( q ) = 2 ⋅ 4 π 2 + O ( q − 2 ) → 8 π 2 , A D → − 3 4 π 3 , − 2 π A D q Λ → 3 2 π 2 q Λ . Z(q)=2\cdot4\pi^2+O(q^{-2})\to8\pi^2,\qquad A_D\to-\frac{3}{4\pi^3},\qquad-\frac{2\pi A_D}{q\Lambda}\to\frac{3}{2\pi^2q\Lambda}. Z ( q ) = 2 ⋅ 4 π 2 + O ( q − 2 ) → 8 π 2 , A D → − 4 π 3 3 , − q Λ 2 π A D → 2 π 2 q Λ 3 .
This is the term of order 1 / Λ 1/\Lambda 1/Λ of the q − 2 q^{-2} q − 2 kernel (48) , and the leading low-density term of the kite is the kernel pole at s = − 1 s=-1 s = − 1 (α = 2 \alpha=2 α = 2 in the corollary).
Cancellation of the universal low-density residue
Lemma S1 (Cancellation of the universal low-density residue). Let S S S be a set of diagrams closed with the same screened line and the same coupling weight, and let the summed integrand Y S = ∑ D ∈ S Y D Y_S=\sum_{D\in S}Y_D Y S = ∑ D ∈ S Y D have the high-frequency tail Y S ≃ A S ( q ) / ν 2 Y_S\simeq A_S(q)/\nu^2 Y S ≃ A S ( q ) / ν 2 . If A S ( q ) = 0 A_S(q)=0 A S ( q ) = 0 for every q q q and the summed radial kernel satisfies the remainder condition of main-text Corollary 1 , the pole at s = − 2 s=-2 s = − 2 is absent from ∑ D ∈ S J D [ W ] \sum_{D\in S}\mathcal{J}_D[W] ∑ D ∈ S J D [ W ] . The condition A S = 0 A_S=0 A S = 0 holds for the first-order exchange corrections to the proper polarization at fixed density.
Proof. The residue (25) of the pole at s = − 2 s=-2 s = − 2 comes from the term − 2 π A / ( q Λ ) -2\pi A/(q\Lambda) − 2 π A / ( q Λ ) of the expansion (S75) in 1 / Λ 1/\Lambda 1/Λ and is linear in the tail A A A of the integrand. The summed integrand Y S = ∑ D ∈ S Y D Y_S=\sum_{D\in S}Y_D Y S = ∑ D ∈ S Y D of the lemma has the radial kernel e S , Λ e_{S,\Lambda} e S , Λ and the tail A S A_S A S , with
e S , Λ = ∑ D ∈ S e D , Λ , A S = ∑ D ∈ S A D , − 2 π A S q Λ = 0 when A S = 0. e_{S,\Lambda}=\sum_{D\in S}e_{D,\Lambda},\qquad A_S=\sum_{D\in S}A_D,\qquad-\frac{2\pi A_S}{q\Lambda}=0\quad\text{when }A_S=0. e S , Λ = D ∈ S ∑ e D , Λ , A S = D ∈ S ∑ A D , − q Λ 2 π A S = 0 when A S = 0.
The remainder condition of Corollary 1, imposed on the summed integrand, then leaves no contribution at s = − 2 s=-2 s = − 2 . For the first-order exchange corrections at fixed density, A s = 0 A_s=0 A s = 0 for every q q q by (S36) (Sec. S10.2).
Argument for complete sets of corrections of higher order. With ρ ^ q = ∑ j e − i q ⋅ r j \hat{\rho}_q=\sum_j e^{-i\mathbf{q}\cdot\mathbf{r}_j} ρ ^ q = ∑ j e − i q ⋅ r j the coefficient of ν − 2 \nu^{-2} ν − 2 in the exact density response is − ⟨ [ [ ρ ^ q , H ^ ] , ρ ^ − q ] ⟩ / ℏ 2 -\langle[[\hat{\rho}_q,\hat{H}],\hat{\rho}_{-q}]\rangle/\hbar^2 − ⟨[[ ρ ^ q , H ^ ] , ρ ^ − q ]⟩ / ℏ 2 per unit volume (the f f f -sum rule [27 Sec. 2.2] ). The interaction commutes with the density operator, and in physical units and in the present sign convention
[ [ ρ ^ q , T ^ ] , ρ ^ − q ] = N ℏ 2 q 2 m e , χ ( q , i ν ) = − n e q 2 m e ν 2 + ⋯ . [[\hat{\rho}_q,\hat{T}],\hat{\rho}_{-q}]=\frac{N\hbar^2q^2}{m_e},\qquad\chi(q,i\nu)=-\frac{n_e q^2}{m_e\nu^2}+\cdots. [[ ρ ^ q , T ^ ] , ρ ^ − q ] = m e N ℏ 2 q 2 , χ ( q , i ν ) = − m e ν 2 n e q 2 + ⋯ .
The coefficient therefore equals its noninteracting value for every N N N -electron state and every coupling. The same holds for the proper polarization, since
Π λ = χ λ 1 + λ v χ λ , λ v χ λ = O ( ν − 2 ) . \Pi_\lambda=\frac{\chi_\lambda}{1+\lambda v\chi_\lambda},\qquad\lambda v\chi_\lambda=O(\nu^{-2}). Π λ = 1 + λ v χ λ χ λ , λ v χ λ = O ( ν − 2 ) .
The argument needs two further assumptions. If the coupling expansion of the exact fixed-density response coincides, order by order, with the sum of the G 0 G_0 G 0 diagrams and their chemical-potential counterterms, and if the large-ν \nu ν limit may be taken term by term, the tails of the corrections of each order sum to zero. Neither assumption is proved here beyond first order. At second and higher order, individual diagrams with bare internal lines can be infrared divergent, and the statement therefore applies to the regularized or resummed sum, not diagram by diagram.
Moment rule for coupling averages
Let M ( j ) ( p ) = d j M / d p j M^{(j)}(p)=\mathrm{d}^jM/\mathrm{d}p^j M ( j ) ( p ) = d j M / d p j be the derivatives of the Mellin multiplier M ( p ) M(p) M ( p ) of main-text Sec. 2.7. A term φ r s p log m r s \varphi r_s^p\log^m r_s φ r s p log m r s of the full-coupling energy F D F_D F D with constant coefficient φ \varphi φ then appears in the average (12) as
φ r s p ∑ j = 0 m ( m j ) M ( j ) ( p ) log m − j r s . (138) \varphi r_s^p\sum_{j=0}^{m}\binom{m}{j}M^{(j)}(p)\log^{m-j}r_s. \tag*{(138)} φ r s p j = 0 ∑ m ( j m ) M ( j ) ( p ) log m − j r s . (138)
This follows by expanding log m ( λ r s ) \log^m(\lambda r_s) log m ( λ r s ) binomially in log r s \log r_s log r s and log λ \log\lambda log λ ,
∫ 0 1 w D ( λ ) φ ( λ r s ) p log m ( λ r s ) d λ = φ r s p ∑ j = 0 m ( m j ) log m − j r s ∫ 0 1 w D ( λ ) λ p log j λ d λ , \int_0^1 w_D(\lambda)\,\varphi(\lambda r_s)^p\log^m(\lambda r_s)\,\mathrm{d}\lambda
=\varphi r_s^p\sum_{j=0}^{m}\binom{m}{j}\log^{m-j}r_s\int_0^1w_D(\lambda)\lambda^p\log^j\lambda\,\mathrm{d}\lambda, ∫ 0 1 w D ( λ ) φ ( λ r s ) p log m ( λ r s ) d λ = φ r s p j = 0 ∑ m ( j m ) log m − j r s ∫ 0 1 w D ( λ ) λ p log j λ d λ ,
∫ 0 1 w D ( λ ) λ p log j λ d λ = M ( j ) ( p ) . \int_0^1w_D(\lambda)\lambda^p\log^j\lambda\,\mathrm{d}\lambda=M^{(j)}(p). ∫ 0 1 w D ( λ ) λ p log j λ d λ = M ( j ) ( p ) .
The rule requires that the moments converge and may be differentiated under the integral (for w D = 2 λ w_D=2\lambda w D = 2 λ , p > − 2 p>-2 p > − 2 ). For w D = 2 λ w_D=2\lambda w D = 2 λ and m = 2 m=2 m = 2 ,
M ( p ) = 2 p + 2 ; φ r s p [ M log 2 r s + 2 M ′ log r s + M ′ ′ ] = 2 φ r s p [ log 2 r s p + 2 − 2 log r s ( p + 2 ) 2 + 2 ( p + 2 ) 3 ] . M(p)=\frac{2}{p+2};\qquad\varphi r_s^p[M\log^2r_s+2M'\log r_s+M'']=2\varphi r_s^p\left[\frac{\log^2r_s}{p+2}-\frac{2\log r_s}{(p+2)^2}+\frac{2}{(p+2)^3}\right]. M ( p ) = p + 2 2 ; φ r s p [ M log 2 r s + 2 M ′ log r s + M ′′ ] = 2 φ r s p [ p + 2 log 2 r s − ( p + 2 ) 2 2 log r s + ( p + 2 ) 3 2 ] .
Residues at the high-density pole
This subsection gives the residue computations behind main-text Secs. 2.8, 2.9 and 3.3. Let the kernel and the scaling function behave as
k ( q ) ≃ κ q α 0 log m ( 1 / q ) ( q → 0 ) , g ( t ) ≃ γ t − β ( t → ∞ ) , 1 + α 0 = β , k(q)\simeq\kappa q^{\alpha_0}\log^m(1/q)\quad(q\to0),\qquad g(t)\simeq\gamma t^{-\beta}\quad(t\to\infty),\qquad1+\alpha_0=\beta, k ( q ) ≃ κ q α 0 log m ( 1/ q ) ( q → 0 ) , g ( t ) ≃ γ t − β ( t → ∞ ) , 1 + α 0 = β ,
where κ \kappa κ and β \beta β are generic (not κ ( Λ ) \kappa(\Lambda) κ ( Λ ) and β ( Λ ) \beta(\Lambda) β ( Λ ) of Eq. (88) ). Then, near s = β s=\beta s = β , the leading Laurent terms are
k ~ ( 1 − s ) ≃ m ! κ ( β − s ) m + 1 , g ~ ( s ) ≃ γ β − s , (139) \widetilde{k}(1-s)\simeq\frac{m!\,\kappa}{(\beta-s)^{m+1}},\qquad\widetilde{g}(s)\simeq\frac{\gamma}{\beta-s}, \tag*{(139)} k ( 1 − s ) ≃ ( β − s ) m + 1 m ! κ , g ( s ) ≃ β − s γ , (139)
from
∫ 0 1 q − s q α 0 log m ( 1 / q ) d q = m ! ( 1 + α 0 − s ) m + 1 , ∫ 1 ∞ t s − 1 − β d t = 1 β − s . \int_0^1 q^{-s}q^{\alpha_0}\log^m(1/q)\,dq=\frac{m!}{(1+\alpha_0-s)^{m+1}},\qquad\int_1^\infty t^{s-1-\beta}\,dt=\frac{1}{\beta-s}. ∫ 0 1 q − s q α 0 log m ( 1/ q ) d q = ( 1 + α 0 − s ) m + 1 m ! , ∫ 1 ∞ t s − 1 − β d t = β − s 1 .
Leading residue. The pole at s = β s=\beta s = β has total order m + 2 m+2 m + 2 . With the expansion of ε s \varepsilon^s ε s about s = β s=\beta s = β ,
k ~ ( 1 − s ) g ~ ( s ) ε s ≃ ( − 1 ) m m ! κ γ ε s ( s − β ) m + 2 , ε s = ε β ∑ n ≥ 0 ( s − β ) n log n ε n ! , \widetilde{k}(1-s)\widetilde{g}(s)\varepsilon^s\simeq\frac{(-1)^m m!\,\kappa\gamma\varepsilon^s}{(s-\beta)^{m+2}},\qquad\varepsilon^s=\varepsilon^\beta\sum_{n\geq0}\frac{(s-\beta)^n\log^n\varepsilon}{n!}, k ( 1 − s ) g ( s ) ε s ≃ ( s − β ) m + 2 ( − 1 ) m m ! κγ ε s , ε s = ε β n ≥ 0 ∑ n ! ( s − β ) n log n ε ,
Res s = β ε s ( s − β ) m + 2 = ε β log m + 1 ε ( m + 1 ) ! . \underset{s=\beta}{\operatorname{Res}}\frac{\varepsilon^s}{(s-\beta)^{m+2}}=\frac{\varepsilon^\beta\log^{m+1}\varepsilon}{(m+1)!}. s = β Res ( s − β ) m + 2 ε s = ( m + 1 )! ε β log m + 1 ε .
For the integral I ( ε ) I(\varepsilon) I ( ε ) of main-text Eq. (22) this gives, as ε → 0 \varepsilon\to0 ε → 0 ,
I ( ε ) = − Res s = β ( − 1 ) m m ! κ γ ε s ( s − β ) m + 2 + ⋯ = ( − 1 ) m + 1 κ γ m + 1 ε β log m + 1 ε + O ( ε β ∣ log ε ∣ m ) . (140) I(\varepsilon)=-\underset{s=\beta}{\operatorname{Res}}\frac{(-1)^m m!\,\kappa\gamma\varepsilon^s}{(s-\beta)^{m+2}}+\cdots=\frac{(-1)^{m+1}\kappa\gamma}{m+1}\varepsilon^\beta\log^{m+1}\varepsilon+O\left(\varepsilon^\beta\lvert\log\varepsilon\rvert^m\right). \tag*{(140)} I ( ε ) = − s = β Res ( s − β ) m + 2 ( − 1 ) m m ! κγ ε s + ⋯ = m + 1 ( − 1 ) m + 1 κγ ε β log m + 1 ε + O ( ε β ∣ log ε ∣ m ) . (140)
The same leading term follows from the window integral
∫ ε 1 κ q α 0 log m ( 1 / q ) γ ( ε / q ) β d q = κ γ ε β ∫ ε 1 log m ( 1 / q ) q d q = κ γ ε β log m + 1 ( 1 / ε ) m + 1 + ⋯ . \int_\varepsilon^1\kappa q^{\alpha_0}\log^m(1/q)\,\gamma(\varepsilon/q)^\beta\,dq=\kappa\gamma\varepsilon^\beta\int_\varepsilon^1\frac{\log^m(1/q)}{q}\,dq=\frac{\kappa\gamma\varepsilon^\beta\log^{m+1}(1/\varepsilon)}{m+1}+\cdots. ∫ ε 1 κ q α 0 log m ( 1/ q ) γ ( ε / q ) β d q = κγ ε β ∫ ε 1 q log m ( 1/ q ) d q = m + 1 κγ ε β log m + 1 ( 1/ ε ) + ⋯ .
Gell-Mann–Brueckner. The kernel and scaling function of main-text Eq. (26) for d = 3 d=3 d = 3 , whose prefactor is 3 8 π E 0 = 12 π 3 q T F 4 R y \frac{3}{8\pi}E_0=\frac{12}{\pi^3q_{\mathrm{TF}}^4}\,\mathrm{Ry} 8 π 3 E 0 = π 3 q TF 4 12 Ry , are at fixed z z z
k = q 3 ( κ , α 0 , m ) = ( 1 , 3 , 0 ) , k=q^3\qquad(\kappa,\alpha_0,m)=(1,3,0), k = q 3 ( κ , α 0 , m ) = ( 1 , 3 , 0 ) ,
g ( t ) = log ( 1 + h t 2 ) − h t 2 ≃ − h 2 2 t 4 ( t → ∞ ) , ( γ , β ) = ( − 1 2 h 2 , 4 ) . g(t)=\log\left(1+\frac{h}{t^2}\right)-\frac{h}{t^2}\simeq-\frac{h^2}{2t^4}\quad(t\to\infty),\qquad(\gamma,\beta)=\left(-\frac{1}{2}h^2,4\right). g ( t ) = log ( 1 + t 2 h ) − t 2 h ≃ − 2 t 4 h 2 ( t → ∞ ) , ( γ , β ) = ( − 2 1 h 2 , 4 ) .
Then (S78) gives − κ γ q T F 4 log q T F = 1 2 h 2 q T F 4 log q T F -\kappa\gamma q_{\mathrm{TF}}^4\log q_{\mathrm{TF}}=\frac{1}{2}h^2q_{\mathrm{TF}}^4\log q_{\mathrm{TF}} − κγ q TF 4 log q TF = 2 1 h 2 q TF 4 log q TF . Replacing L ( q , q z ) \mathcal{L}(q,qz) L ( q , q z ) by h ( z ) h(z) h ( z ) for q < 1 q<1 q < 1 changes c c c as in Eq. (150) , and for q > 1 q>1 q > 1 the z z z integral of c 2 c^2 c 2 follows from the large-q q q form of L \mathcal{L} L in Eq. (155) ,
q < 1 : Δ c = O ( q T F 2 q ( 1 + z 2 ) ) , ∣ ∂ c [ log ( 1 + c ) − c ] ∣ ≤ c , q<1:\quad\Delta c=O\left(\frac{q_{\mathrm{TF}}^2}{q(1+z^2)}\right),\qquad\left|\partial_c[\log(1+c)-c]\right|\leq c, q < 1 : Δ c = O ( q ( 1 + z 2 ) q TF 2 ) , ∣ ∂ c [ log ( 1 + c ) − c ] ∣ ≤ c ,
q > 1 : 0 ≤ c − log ( 1 + c ) ≤ c 2 2 , ∫ c 2 d z = O ( q T F 4 q − 7 ) . q>1:\quad0\leq c-\log(1+c)\leq\frac{c^2}{2},\qquad\int c^2\,dz=O\left(q_{\mathrm{TF}}^4q^{-7}\right). q > 1 : 0 ≤ c − log ( 1 + c ) ≤ 2 c 2 , ∫ c 2 d z = O ( q TF 4 q − 7 ) .
Hence the replacement for q < 1 q<1 q < 1 changes e R P A e_{\mathrm{RPA}} e RPA only at O ( 1 ) O(1) O ( 1 ) , and the region q > 1 q>1 q > 1 also contributes only at O ( 1 ) O(1) O ( 1 ) . With the prefactor 12 / ( π 3 q T F 4 ) 12/(\pi^3q_{\mathrm{TF}}^4) 12/ ( π 3 q TF 4 ) and log q T F = 1 2 log r s + O ( 1 ) \log q_{\mathrm{TF}}=\frac{1}{2}\log r_s+O(1) log q TF = 2 1 log r s + O ( 1 ) ,
e R P A = 3 π 3 ∫ − ∞ ∞ h 2 d z log r s + O ( 1 ) , ∫ − ∞ ∞ h 2 d z = 2 π 3 ( 1 − log 2 ) , (141) e_{\mathrm{RPA}}=\frac{3}{\pi^3}\int_{-\infty}^{\infty}h^2\,dz\log r_s+O(1),\qquad\int_{-\infty}^{\infty}h^2\,dz=\frac{2\pi}{3}(1-\log2), \tag*{(141)} e RPA = π 3 3 ∫ − ∞ ∞ h 2 d z log r s + O ( 1 ) , ∫ − ∞ ∞ h 2 d z = 3 2 π ( 1 − log 2 ) , (141)
which is main-text Eq. (29) .
Closed-diagram coefficients. Write c = c f + δ c c=c_{\mathrm f}+\delta c c = c f + δ c with c f = − v Π 0 c_{\mathrm f}=-v\Pi_0 c f = − v Π 0 and δ c = − v Π ( 1 ) \delta c=-v\Pi^{(1)} δ c = − v Π ( 1 ) . Then
log ( 1 + c ) − c = ∑ j ≥ 2 ( − 1 ) j + 1 j c j , c j = c f j + j c f j − 1 δ c + O ( δ c 2 ) , \log(1+c)-c=\sum_{j\geq2}\frac{(-1)^{j+1}}{j}c^j,\qquad c^j=c_{\mathrm f}^j+jc_{\mathrm f}^{j-1}\delta c+O(\delta c^2), log ( 1 + c ) − c = j ≥ 2 ∑ j ( − 1 ) j + 1 c j , c j = c f j + j c f j − 1 δ c + O ( δ c 2 ) ,
and the coefficient of c f j − 1 δ c c_{\mathrm f}^{j-1}\delta c c f j − 1 δ c is j j j times that of c f j c_{\mathrm f}^j c f j . A ring of j j j lines with one insertion therefore receives j j j times the coefficient 1 / ( 2 j ) 1/(2j) 1/ ( 2 j ) of the ring of j j j free bubbles, that is, 1 2 \frac{1}{2} 2 1 . A ring with one insertion Π v t x ( 1 ) \Pi_{\mathrm{vtx}}^{(1)} Π vtx ( 1 ) keeps only the reflection that reverses its direction, and therefore its closed-diagram coefficient, counted as in Sec. S12.1, is 1 2 \frac{1}{2} 2 1 . The two insertions of Π S E ( 1 ) \Pi_{\mathrm{SE}}^{(1)} Π SE ( 1 ) close to the same diagram, as for S D = 4 S_{\mathrm{D}}=4 S D = 4 in main-text § 3.4 . The first-order change of main-text Eq. (26) is therefore the closed-diagram series of Refs. [28 , 7 ] . A coupling average with δ Π λ = λ Π ( 1 ) \delta\Pi_{\lambda}=\lambda\Pi^{(1)} δ Π λ = λ Π ( 1 ) , main-text Eq. (54) , would weight the t − 2 t^{-2} t − 2 tail by
∫ 0 1 λ [ ( 1 + λ c ) − 2 − 1 ] d λ ≃ − 2 c ∫ 0 1 λ 2 d λ = − 2 3 c \int_{0}^{1}\lambda\left[(1+\lambda c)^{-2}-1\right]\,\mathrm{d}\lambda\simeq-2c\int_{0}^{1}\lambda^{2}\,\mathrm{d}\lambda=-\frac{2}{3}c ∫ 0 1 λ [ ( 1 + λ c ) − 2 − 1 ] d λ ≃ − 2 c ∫ 0 1 λ 2 d λ = − 3 2 c
in place of − c -c − c .
This weight multiplies the coefficient of r s log r s r_{\mathrm{s}}\log r_{\mathrm{s}} r s log r s by the Mellin multiplier 2 3 \frac{2}{3} 3 2 .
First-order exchange polarization in the ring chain. The first-order change of log ( 1 + c ) − c \log(1+c)-c log ( 1 + c ) − c (main-text § 2.9 ) and its scaling function are
− ν Π ( 1 ) [ 1 1 + c − 1 ] , 1 1 + c − 1 = − c 1 + c ≃ − h t 2 ( γ = − h , β = 2 ) . -\nu\Pi^{(1)}\left[\frac{1}{1+c}-1\right],\qquad\frac{1}{1+c}-1=-\frac{c}{1+c}\simeq-\frac{h}{t^{2}}\qquad(\gamma=-h,\ \beta=2). − ν Π ( 1 ) [ 1 + c 1 − 1 ] , 1 + c 1 − 1 = − 1 + c c ≃ − t 2 h ( γ = − h , β = 2 ) .
With Π ( 1 ) ( q , i q z ) → − ( q T F 4 / 16 ) R ( 1 ) ( z ) \Pi^{(1)}(q,iqz)\to-(q_{\mathrm{TF}}^{4}/16)R^{(1)}(z) Π ( 1 ) ( q , i q z ) → − ( q TF 4 /16 ) R ( 1 ) ( z ) [28 ] and [29 Eqs. (40) and (41)] , where R ( 1 ) R^{(1)} R ( 1 ) is given in main-text Eq. (35) , the kernel is q 3 ⋅ q T F 4 R ( 1 ) / ( 16 q 2 ) q^{3}\cdot q_{\mathrm{TF}}^{4}R^{(1)}/(16q^{2}) q 3 ⋅ q TF 4 R ( 1 ) / ( 16 q 2 ) ( α 0 = 1 , m = 0 ) (\alpha_{0}=1,\ m=0) ( α 0 = 1 , m = 0 ) . Then (S78) with the prefactor 12 / ( π 3 q T F 4 ) 12/(\pi^{3}q_{\mathrm{TF}}^{4}) 12/ ( π 3 q TF 4 ) gives
12 π 3 q T F 4 ∫ d z q T F 4 R ( 1 ) 16 h q T F 2 log q T F = 3 4 π 3 ∫ h R ( 1 ) d z q T F 2 log q T F , \frac{12}{\pi^{3}q_{\mathrm{TF}}^{4}}\int\mathrm{d}z\,\frac{q_{\mathrm{TF}}^{4}R^{(1)}}{16}h\,q_{\mathrm{TF}}^{2}\log q_{\mathrm{TF}}=\frac{3}{4\pi^{3}}\int hR^{(1)}\,\mathrm{d}z\,q_{\mathrm{TF}}^{2}\log q_{\mathrm{TF}}, π 3 q TF 4 12 ∫ d z 16 q TF 4 R ( 1 ) h q TF 2 log q TF = 4 π 3 3 ∫ h R ( 1 ) d z q TF 2 log q TF ,
q T F 2 log q T F = 2 α L π r s log r s + O ( r s ) , q_{\mathrm{TF}}^{2}\log q_{\mathrm{TF}}=\frac{2\alpha_{\mathrm{L}}}{\pi}r_{\mathrm{s}}\log r_{\mathrm{s}}+O(r_{\mathrm{s}}), q TF 2 log q TF = π 2 α L r s log r s + O ( r s ) ,
which is main-text Eq. (36) . The remainder O ( r s ) O(r_{\mathrm{s}}) O ( r s ) also contains the corrections to the limiting forms. Replacing L ( q , q z ) \mathcal{L}(q,qz) L ( q , q z ) by h ( z ) h(z) h ( z ) for q < 1 q<1 q < 1 , estimated as for the ring series above, and the region q > 1 q>1 q > 1 , bounded using ∣ c / ( 1 + c ) ∣ ≤ c \lvert c/(1+c)\rvert\leq c ∣ c / ( 1 + c )∣ ≤ c and the absolute frequency bounds of § 24.3 , both give corrections O ( q T F 2 ) = O ( r s ) O(q_{\mathrm{TF}}^{2})=O(r_{\mathrm{s}}) O ( q TF 2 ) = O ( r s ) . The correction to the limit − ( q T F 4 / 16 ) R ( 1 ) ( z ) -(q_{\mathrm{TF}}^{4}/16)R^{(1)}(z) − ( q TF 4 /16 ) R ( 1 ) ( z ) of Π ( 1 ) ( q , i q z ) \Pi^{(1)}(q,iqz) Π ( 1 ) ( q , i q z ) also enters at O ( r s ) O(r_{\mathrm{s}}) O ( r s ) if it vanishes as a positive power of q q q with a bound integrable in z z z . This condition is assumed, not proved. The order of the remainder agrees with the next term, − 0.021 r s -0.021r_{\mathrm{s}} − 0.021 r s Ry, of Ref. [7 ] and Ref. [30, Eq. (5)].
Lindhard correction in the ring series (main-text § 2.9 ). Expanding Eq. (149) at fixed z z z gives
L ( q , q z ) = h ( z ) + q 2 g 3 ( z ) + O ( q 4 ) , g 3 ( z ) = − 1 12 ( 1 + z 2 ) 2 . \mathcal{L}(q,qz)=h(z)+q^{2}g_{3}(z)+O(q^{4}),\qquad g_{3}(z)=-\frac{1}{12(1+z^{2})^{2}}. L ( q , q z ) = h ( z ) + q 2 g 3 ( z ) + O ( q 4 ) , g 3 ( z ) = − 12 ( 1 + z 2 ) 2 1 .
The resulting addition to the integrand of main-text Eq. (26) has the kernel parameters and the term of − c / ( 1 + c ) -c/(1+c) − c / ( 1 + c )
( κ , α 0 , m ) = ( q T F 2 g 3 , 3 , 0 ) , ( γ , β ) = ( h 2 , 4 ) , (\kappa,\alpha_{0},m)=(q_{\mathrm{TF}}^{2}g_{3},3,0),\qquad(\gamma,\beta)=(h^{2},4), ( κ , α 0 , m ) = ( q TF 2 g 3 , 3 , 0 ) , ( γ , β ) = ( h 2 , 4 ) ,
− κ γ q T F 4 log q T F = − h 2 g 3 q T F 6 log q T F , -\kappa\gamma q_{\mathrm{TF}}^{4}\log q_{\mathrm{TF}}=-h^{2}g_{3}q_{\mathrm{TF}}^{6}\log q_{\mathrm{TF}}, − κγ q TF 4 log q TF = − h 2 g 3 q TF 6 log q TF ,
the last by Eq. (140) . With the prefactor 12 / ( π 3 q T F 4 ) 12/(\pi^{3}q_{\mathrm{TF}}^{4}) 12/ ( π 3 q TF 4 ) , q T F 2 log q T F = ( 2 α L / π ) r s log r s + O ( r s ) q_{\mathrm{TF}}^{2}\log q_{\mathrm{TF}}=(2\alpha_{\mathrm{L}}/\pi)r_{\mathrm{s}}\log r_{\mathrm{s}}+O(r_{\mathrm{s}}) q TF 2 log q TF = ( 2 α L / π ) r s log r s + O ( r s ) and ∫ − ∞ ∞ h 2 g 3 d z = − π ( π 2 − 6 ) / 288 \int_{-\infty}^{\infty}h^{2}g_{3}\,\mathrm{d}z=-\pi(\pi^{2}-6)/288 ∫ − ∞ ∞ h 2 g 3 d z = − π ( π 2 − 6 ) /288 ,
− 12 π 3 ∫ − ∞ ∞ h 2 g 3 d z q T F 2 log q T F = 12 π 3 ⋅ π ( π 2 − 6 ) 288 ⋅ 2 α L π r s log r s + O ( r s ) = α L ( π 2 − 6 ) 12 π 3 r s log r s = 0.00542 r s log r s + O ( r s ) . \begin{aligned}
-\frac{12}{\pi^{3}}\int_{-\infty}^{\infty}h^{2}g_{3}\,\mathrm{d}z\,q_{\mathrm{TF}}^{2}\log q_{\mathrm{TF}}=\frac{12}{\pi^{3}}\cdot\frac{\pi(\pi^{2}-6)}{288}\cdot\frac{2\alpha_{\mathrm{L}}}{\pi}r_{\mathrm{s}}\log r_{\mathrm{s}}+O(r_{\mathrm{s}}) \\
&=\frac{\alpha_{\mathrm{L}}(\pi^{2}-6)}{12\pi^{3}}r_{\mathrm{s}}\log r_{\mathrm{s}}=0.00542r_{\mathrm{s}}\log r_{\mathrm{s}}+O(r_{\mathrm{s}}).
\end{aligned} − π 3 12 ∫ − ∞ ∞ h 2 g 3 d z q TF 2 log q TF = π 3 12 ⋅ 288 π ( π 2 − 6 ) ⋅ π 2 α L r s log r s + O ( r s ) = 12 π 3 α L ( π 2 − 6 ) r s log r s = 0.00542 r s log r s + O ( r s ) .
The next term of the expansion, of order q 4 q^{4} q 4 , has its kernel pole at s = 6 s=6 s = 6 , which meets the third screening pole. The term quadratic in q T F 2 g 3 q_{\mathrm{TF}}^{2}g_{3} q TF 2 g 3 carries a further factor q T F 2 q_{\mathrm{TF}}^{2} q TF 2 . Both therefore contribute only at r s 2 log r s r_{\mathrm{s}}^{2}\log r_{\mathrm{s}} r s 2 log r s and beyond. The region q > 1 q>1 q > 1 is bounded as for the ring series.
Kite. For e k i t e e_{\mathrm{kite}} e kite at fixed z z z , main-text Eq. (49) has
( κ , α 0 , m ) = ( a x ( z ) , 1 , 1 ) , ( γ , β ) = ( − 2 3 h ( z ) , 2 ) , (\kappa,\alpha_0,m)=(a_x(z),1,1),\qquad(\gamma,\beta)=\left(-\frac{2}{3}h(z),2\right), ( κ , α 0 , m ) = ( a x ( z ) , 1 , 1 ) , ( γ , β ) = ( − 3 2 h ( z ) , 2 ) ,
1 2 κ γ q T F 2 log 2 q T F = − 1 3 a x h q T F 2 log 2 q T F , \frac{1}{2}\kappa\gamma q_{\mathrm{TF}}^2\log^2 q_{\mathrm{TF}}=-\frac{1}{3}a_xh q_{\mathrm{TF}}^2\log^2 q_{\mathrm{TF}}, 2 1 κγ q TF 2 log 2 q TF = − 3 1 a x h q TF 2 log 2 q TF ,
the last by (S78). Integrating with 4 π A 0 ∫ 0 ∞ d z / π 4\pi A_0\int_0^\infty dz/\pi 4 π A 0 ∫ 0 ∞ d z / π and (S31), and using q T F 2 log 2 q T F = ( α L / π ) r s log 2 r s + O ( r s ∣ log r s ∣ ) q_{\mathrm{TF}}^2\log^2q_{\mathrm{TF}}=(\alpha_L/\pi)r_s\log^2r_s+O(r_s|\log r_s|) q TF 2 log 2 q TF = ( α L / π ) r s log 2 r s + O ( r s ∣ log r s ∣ ) , gives c h d c_{\mathrm{hd}} c hd of (S32). For the full-coupling energy F F F the tail is γ = − h \gamma=-h γ = − h , which multiplies the coefficient by 3 / 2 3/2 3/2 .
Radial integral of the Fock self-energy corrections
Frequency integration and radial variables
Frequency integral. With Re ( i ν − δ ) − 1 = − δ / ( ν 2 + δ 2 ) \operatorname{Re}(i\nu-\delta)^{-1}=-\delta/(\nu^2+\delta^2) Re ( i ν − δ ) − 1 = − δ / ( ν 2 + δ 2 ) and ∂ δ ( i ν − δ ) − 1 = ( i ν − δ ) − 2 \partial_\delta(i\nu-\delta)^{-1}=(i\nu-\delta)^{-2} ∂ δ ( i ν − δ ) − 1 = ( i ν − δ ) − 2 , the frequency identity (17) follows by differentiating the single-pole integral in δ \delta δ ,
∫ 0 ∞ d ν π Ω 2 Ω 2 + ν 2 Re ( i ν − δ ) − 1 = − Ω 2 δ π ∫ 0 ∞ d ν ( Ω 2 + ν 2 ) ( ν 2 + δ 2 ) = − sgn ( δ ) Ω 2 ( Ω + ∣ δ ∣ ) → ∂ δ Ω 2 ( Ω + ∣ δ ∣ ) 2 . \int_0^\infty\frac{d\nu}{\pi}\frac{\Omega^2}{\Omega^2+\nu^2}\operatorname{Re}(i\nu-\delta)^{-1}
=-\frac{\Omega^2\delta}{\pi}\int_0^\infty\frac{d\nu}{(\Omega^2+\nu^2)(\nu^2+\delta^2)}
=-\frac{\operatorname{sgn}(\delta)\Omega}{2(\Omega+|\delta|)}
\xrightarrow{\partial_\delta}\frac{\Omega}{2(\Omega+|\delta|)^2}. ∫ 0 ∞ π d ν Ω 2 + ν 2 Ω 2 Re ( i ν − δ ) − 1 = − π Ω 2 δ ∫ 0 ∞ ( Ω 2 + ν 2 ) ( ν 2 + δ 2 ) d ν = − 2 ( Ω + ∣ δ ∣ ) sgn ( δ ) Ω ∂ δ 2 ( Ω + ∣ δ ∣ ) 2 Ω .
Radial variables. On the transition region Δ p = 1 \Delta_p=1 Δ p = 1 , set r = ∣ p ∣ < 1 r=|\mathbf p|<1 r = ∣ p ∣ < 1 and s = ∣ p + q ∣ > 1 s=|\mathbf p+\mathbf q|>1 s = ∣ p + q ∣ > 1 (here s s s is a radius, not the Mellin variable). The map p ↦ − p − q \mathbf p\mapsto-\mathbf p-\mathbf q p ↦ − p − q takes the other transition region onto this one, with an equal contribution. With δ p \delta_p δ p of main-text Eq. (41) and Σ x \Sigma_x Σ x of (S19),
δ p = s 2 − r 2 2 , Σ x ( s ) − Σ x ( r ) = q T F 2 2 [ f L ( r ) − f L ( s ) ] , (142) \delta_p=\frac{s^2-r^2}{2},\qquad\Sigma_x(s)-\Sigma_x(r)=\frac{q_{\mathrm{TF}}^2}{2}[f_L(r)-f_L(s)], \tag*{(142)} δ p = 2 s 2 − r 2 , Σ x ( s ) − Σ x ( r ) = 2 q TF 2 [ f L ( r ) − f L ( s )] , (142)
and, with θ \theta θ the angle between p \mathbf p p and q \mathbf q q ,
s 2 = r 2 + q 2 + 2 r q cos θ , d 3 p = 2 π r 2 d r d cos θ = 2 π q r s d r d s , d 3 q = 4 π q 2 d q , s^2=r^2+q^2+2rq\cos\theta,\qquad d^3p=2\pi r^2\,dr\,d\cos\theta=\frac{2\pi}{q}rs\,dr\,ds,\qquad d^3q=4\pi q^2\,dq, s 2 = r 2 + q 2 + 2 r q cos θ , d 3 p = 2 π r 2 d r d cos θ = q 2 π r s d r d s , d 3 q = 4 π q 2 d q ,
give
d 3 q d 3 p ( 2 π ) 3 = r s q π d r d s d q , s − r < q < s + r . (143) \frac{d^3q\,d^3p}{(2\pi)^3}=\frac{rsq}{\pi}\,dr\,ds\,dq,\qquad s-r<q<s+r. \tag*{(143)} ( 2 π ) 3 d 3 q d 3 p = π r s q d r d s d q , s − r < q < s + r . (143)
Insertion. Insert f = b Λ = q − 2 Ω 2 / ( Ω 2 + ν 2 ) f=b_\Lambda=q^{-2}\Omega^2/(\Omega^2+\nu^2) f = b Λ = q − 2 Ω 2 / ( Ω 2 + ν 2 ) with Ω = Λ q \Omega=\Lambda q Ω = Λ q and (S21) into (52) . By (17) , with δ p > 0 \delta_p>0 δ p > 0 on Δ p = 1 \Delta_p=1 Δ p = 1 , a factor 2 2 2 for the two regions and 2 C = 2 π 2 q T F 2 2C=2\pi^2q_{\mathrm{TF}}^2 2 C = 2 π 2 q TF 2 ,
∫ 0 ∞ d ν π b Λ Re ( i ν − δ p ) − 2 = Λ 2 q ( Λ q + δ p ) 2 , \int_0^\infty\frac{d\nu}{\pi}\,b_\Lambda\operatorname{Re}(i\nu-\delta_p)^{-2}=\frac{\Lambda}{2q(\Lambda q+\delta_p)^2}, ∫ 0 ∞ π d ν b Λ Re ( i ν − δ p ) − 2 = 2 q ( Λ q + δ p ) 2 Λ ,
K S E ( 1 ) [ b Λ ] = − 3 π 3 q T F 4 ⋅ 2 ⋅ 2 π 2 q T F 2 ∫ Δ p = 1 d 3 q d 3 p ( 2 π ) 3 [ Σ x ( s ) − Σ x ( r ) ] Λ 2 q ( Λ q + δ p ) 2 . \mathcal{K}_{\mathrm{SE}}^{(1)}[b_\Lambda]=-\frac{3}{\pi^3q_{\mathrm{TF}}^4}\cdot2\cdot2\pi^2q_{\mathrm{TF}}^2\int_{\Delta_p=1}\frac{d^3q\,d^3p}{(2\pi)^3}[\Sigma_x(s)-\Sigma_x(r)]\frac{\Lambda}{2q(\Lambda q+\delta_p)^2}. K SE ( 1 ) [ b Λ ] = − π 3 q TF 4 3 ⋅ 2 ⋅ 2 π 2 q TF 2 ∫ Δ p = 1 ( 2 π ) 3 d 3 q d 3 p [ Σ x ( s ) − Σ x ( r )] 2 q ( Λ q + δ p ) 2 Λ .
In the radial variables the prefactor is
− 3 π 3 q T F 4 ⋅ 4 π 2 q T F 2 ⋅ q T F 2 2 ⋅ 1 π ⋅ Λ 2 = − 3 Λ π 2 . -\frac{3}{\pi^3q_{\mathrm{TF}}^4}\cdot4\pi^2q_{\mathrm{TF}}^2\cdot\frac{q_{\mathrm{TF}}^2}{2}\cdot\frac{1}{\pi}\cdot\frac{\Lambda}{2}=-\frac{3\Lambda}{\pi^2}. − π 3 q TF 4 3 ⋅ 4 π 2 q TF 2 ⋅ 2 q TF 2 ⋅ π 1 ⋅ 2 Λ = − π 2 3Λ .
The inner q q q integral is elementary, and the factorization Λ ( s ∓ r ) + 1 2 ( s 2 − r 2 ) = ( s ∓ r ) [ Λ + 1 2 ( s ± r ) ] \Lambda(s\mp r)+\frac{1}{2}(s^2-r^2)=(s\mp r)[\Lambda+\frac{1}{2}(s\pm r)] Λ ( s ∓ r ) + 2 1 ( s 2 − r 2 ) = ( s ∓ r ) [ Λ + 2 1 ( s ± r )] gives its closed form,
∫ s − r s + r d q [ Λ q + ( s 2 − r 2 ) / 2 ] 2 = 1 Λ [ 1 Λ ( s − r ) + ( s 2 − r 2 ) / 2 − 1 Λ ( s + r ) + ( s 2 − r 2 ) / 2 ] . \int_{s-r}^{s+r}\frac{dq}{[\Lambda q+(s^2-r^2)/2]^2}
=\frac{1}{\Lambda}\left[\frac{1}{\Lambda(s-r)+(s^2-r^2)/2}-\frac{1}{\Lambda(s+r)+(s^2-r^2)/2}\right]. ∫ s − r s + r [ Λ q + ( s 2 − r 2 ) /2 ] 2 d q = Λ 1 [ Λ ( s − r ) + ( s 2 − r 2 ) /2 1 − Λ ( s + r ) + ( s 2 − r 2 ) /2 1 ] .
These give the form below, whose last line is the inner q q q integral,
K Π S E ( 1 ) [ b Λ ] = − 3 Λ π 2 ∫ 0 1 r d r ∫ 1 ∞ s d s [ f L ( r ) − f L ( s ) ] ∫ s − r s + r d q [ Λ q + ( s 2 − r 2 ) / 2 ] 2 , (144) \mathcal{K}_{\Pi\mathrm{SE}}^{(1)}[b_{\Lambda}] = -\frac{3\Lambda}{\pi^2}\int_0^1 r\,\mathrm{d}r\int_1^\infty s\,\mathrm{d}s\,[f_L(r)-f_L(s)]
\int_{s-r}^{s+r}\frac{\mathrm{d}q}{[\Lambda q+(s^2-r^2)/2]^2}, \tag*{(144)} K Π SE ( 1 ) [ b Λ ] = − π 2 3Λ ∫ 0 1 r d r ∫ 1 ∞ s d s [ f L ( r ) − f L ( s )] ∫ s − r s + r [ Λ q + ( s 2 − r 2 ) /2 ] 2 d q , (144)
∫ s − r s + r d q [ Λ q + ( s 2 − r 2 ) / 2 ] 2 = 2 r ( s 2 − r 2 ) [ ( Λ + s / 2 ) 2 − r 2 / 4 ] , \int_{s-r}^{s+r}\frac{\mathrm{d}q}{[\Lambda q+(s^2-r^2)/2]^2}
=\frac{2r}{(s^2-r^2)[(\Lambda+s/2)^2-r^2/4]}, ∫ s − r s + r [ Λ q + ( s 2 − r 2 ) /2 ] 2 d q = ( s 2 − r 2 ) [( Λ + s /2 ) 2 − r 2 /4 ] 2 r ,
and hence
K Π S E ( 1 ) [ b Λ ] = − 6 Λ π 2 ∫ 0 1 r 2 d r ∫ 1 ∞ s d s f L ( r ) − f L ( s ) ( s 2 − r 2 ) [ ( Λ + s / 2 ) 2 − r 2 / 4 ] . (145) \mathcal{K}_{\Pi\mathrm{SE}}^{(1)}[b_{\Lambda}] = -\frac{6\Lambda}{\pi^2}\int_0^1 r^2\,\mathrm{d}r\int_1^\infty s\,\mathrm{d}s\,\frac{f_L(r)-f_L(s)}{(s^2-r^2)[(\Lambda+s/2)^2-r^2/4]}. \tag*{(145)} K Π SE ( 1 ) [ b Λ ] = − π 2 6Λ ∫ 0 1 r 2 d r ∫ 1 ∞ s d s ( s 2 − r 2 ) [( Λ + s /2 ) 2 − r 2 /4 ] f L ( r ) − f L ( s ) . (145)
Before the q q q integration, the integrand of the first line of (S82) is the radial kernel of the self-energy pair (Sec. S17.4). Section S19 evaluates (S83) numerically.
Sign. Apart from the prefactor − 6 Λ / π 2 -6\Lambda/\pi^2 − 6Λ/ π 2 , the integrand of (S83) is positive because f L f_L f L is decreasing,
f L ( r ) − f L ( s ) > 0 , ( Λ + s / 2 ) 2 > r 2 / 4 ( r < 1 < s ) ⟹ K Π S E ( 1 ) [ b Λ ] < 0 ( Λ > 0 ) . f_L(r)-f_L(s)>0,\qquad(\Lambda+s/2)^2>r^2/4\quad(r<1<s)
\qquad\Longrightarrow\qquad\mathcal{K}_{\Pi\mathrm{SE}}^{(1)}[b_{\Lambda}]<0\quad(\Lambda>0). f L ( r ) − f L ( s ) > 0 , ( Λ + s /2 ) 2 > r 2 /4 ( r < 1 < s ) ⟹ K Π SE ( 1 ) [ b Λ ] < 0 ( Λ > 0 ) .
The sign already holds at fixed momentum transfer q q q of the line. Apart from the prefactor − 3 Λ / π 2 -3\Lambda/\pi^2 − 3Λ/ π 2 , the r r r , s s s integrand in the first line of (S82) is positive at fixed q q q . This pointwise sign is needed because the weight a λ ( q , Λ ) a_{\lambda}(q,\Lambda) a λ ( q , Λ ) of main-text (14) depends on q q q . With K Π S E ( 1 ) [ v ] = 0 \mathcal{K}_{\Pi\mathrm{SE}}^{(1)}[v]=0 K Π SE ( 1 ) [ v ] = 0 [Eq. (S84)] and a λ ≥ 0 a_{\lambda}\geq0 a λ ≥ 0 , it gives
K Π S E ( 1 ) [ W λ ] = λ K Π S E ( 1 ) [ v ] − K Π S E ( 1 ) [ ∫ 0 ∞ a λ b Λ d Λ ] > 0 , \mathcal{K}_{\Pi\mathrm{SE}}^{(1)}[W_{\lambda}]
=\lambda\mathcal{K}_{\Pi\mathrm{SE}}^{(1)}[v]
-\mathcal{K}_{\Pi\mathrm{SE}}^{(1)}
\left[\int_0^\infty a_{\lambda}b_{\Lambda}\,\mathrm{d}\Lambda\right]>0, K Π SE ( 1 ) [ W λ ] = λ K Π SE ( 1 ) [ v ] − K Π SE ( 1 ) [ ∫ 0 ∞ a λ b Λ d Λ ] > 0 ,
consistent with e S E = − A 1 / q ∗ > 0 e_{\mathrm{SE}}=-A_1/q_{\ast}>0 e SE = − A 1 / q ∗ > 0 in (168) .
Bare limit
The bare-line energy of the self-energy pair vanishes because, for δ ≠ 0 \delta\ne0 δ = 0 ,
∫ 0 ∞ d ν π Re ( i ν − δ ) − 2 = 0. (146) \int_0^\infty\frac{\mathrm{d}\nu}{\pi}\operatorname{Re}(i\nu-\delta)^{-2}=0. \tag*{(146)} ∫ 0 ∞ π d ν Re ( i ν − δ ) − 2 = 0. (146)
This holds because (see after (127) )
Re ( i ν − δ ) − 2 = δ 2 − ν 2 ( ν 2 + δ 2 ) 2 = ∂ ν ν ν 2 + δ 2 , ν ν 2 + δ 2 ∣ ν = 0 = 0 = lim ν → ∞ ν ν 2 + δ 2 . \operatorname{Re}(i\nu-\delta)^{-2}
=\frac{\delta^2-\nu^2}{(\nu^2+\delta^2)^2}
=\partial_{\nu}\frac{\nu}{\nu^2+\delta^2},
\qquad
\left.\frac{\nu}{\nu^2+\delta^2}\right|_{\nu=0}=0
=\lim_{\nu\to\infty}\frac{\nu}{\nu^2+\delta^2}. Re ( i ν − δ ) − 2 = ( ν 2 + δ 2 ) 2 δ 2 − ν 2 = ∂ ν ν 2 + δ 2 ν , ν 2 + δ 2 ν ν = 0 = 0 = ν → ∞ lim ν 2 + δ 2 ν .
Integrating over frequency first, which the absolute bounds of Sec. S13.3 permit, gives K Π S E ( 1 ) [ v ] = 0 \mathcal{K}_{\Pi\mathrm{SE}}^{(1)}[v]=0 K Π SE ( 1 ) [ v ] = 0 . Since 0 ≤ b Λ ≤ v 0\leq b_{\Lambda}\leq v 0 ≤ b Λ ≤ v , dominated convergence gives K Π S E ( 1 ) [ b Λ ] → 0 \mathcal{K}_{\Pi\mathrm{SE}}^{(1)}[b_{\Lambda}]\to0 K Π SE ( 1 ) [ b Λ ] → 0 as Λ → ∞ \Lambda\to\infty Λ → ∞ , in agreement with (S83).
Frequency-integrated shell bounds
This section gives the bounds behind the frequency-integrated form of main-text (47) ,
∫ 0 ∞ d z π h ( z ) X ( q , q z ) = A log ( 1 / q ) + O ( 1 ) , \int_0^\infty\frac{\mathrm{d}z}{\pi}\,h(z)X(q,qz)=A\log(1/q)+O(1), ∫ 0 ∞ π d z h ( z ) X ( q , q z ) = A log ( 1/ q ) + O ( 1 ) ,
with constants independent of q q q and Λ \Lambda Λ , and the remainder O ( r s ∣ log r s ∣ ) O(r_s|\log r_s|) O ( r s ∣ log r s ∣ ) of main-text Eq. (50) . The frequency integral is taken first. It turns the electron–hole factors into a weight with bounded derivatives.
Weights. With a = δ p / q a=\delta_p/q a = δ p / q , b = δ k / q b=\delta_k/q b = δ k / q , α = ∣ a ∣ \alpha=|a| α = ∣ a ∣ , β = ∣ b ∣ \beta=|b| β = ∣ b ∣ ,
M Λ ( a , b ) : = a b Λ 2 J ( Λ ; a , b ) = { α β Λ 2 ( Λ + α ) ( Λ + β ) , a b > 0 , α β 2 ( α + β ) ( Λ Λ + α + Λ Λ + β ) , a b < 0. (147) M_\Lambda(a,b):=ab\Lambda^2J(\Lambda;a,b)=
\begin{cases}
\dfrac{\alpha\beta\Lambda}{2(\Lambda+\alpha)(\Lambda+\beta)}, & ab>0,\\
\dfrac{\alpha\beta}{2(\alpha+\beta)}\left(\dfrac{\Lambda}{\Lambda+\alpha}+\dfrac{\Lambda}{\Lambda+\beta}\right), & ab<0.
\end{cases} \tag*{(147)} M Λ ( a , b ) := ab Λ 2 J ( Λ ; a , b ) = ⎩ ⎨ ⎧ 2 ( Λ + α ) ( Λ + β ) α β Λ , 2 ( α + β ) α β ( Λ + α Λ + Λ + β Λ ) , ab > 0 , ab < 0. (147)
With R \mathcal{R} R of (S25) and h ( z ) = 1 − z arctan ( 1 / z ) = ∫ 0 1 Λ 2 d Λ Λ 2 + z 2 h(z)=1-z\arctan(1/z)=\int_0^1 \dfrac{\Lambda^2\,\mathrm{d}\Lambda}{\Lambda^2+z^2} h ( z ) = 1 − z arctan ( 1/ z ) = ∫ 0 1 Λ 2 + z 2 Λ 2 d Λ [main-text Eq. (9) ],
M Λ ( a , b ) = ∫ 0 ∞ d z π Λ 2 Λ 2 + z 2 a b R ( a , b ; z ) , M h ( a , b ) = ∫ 0 ∞ d z π h ( z ) a b R ( a , b ; z ) = ∫ 0 1 M Λ d Λ . M_\Lambda(a,b)=\int_0^\infty\frac{\mathrm{d}z}{\pi}\frac{\Lambda^2}{\Lambda^2+z^2}ab\mathcal{R}(a,b;z),\qquad M_h(a,b)=\int_0^\infty\frac{\mathrm{d}z}{\pi}h(z)ab\mathcal{R}(a,b;z)=\int_0^1M_\Lambda\,\mathrm{d}\Lambda. M Λ ( a , b ) = ∫ 0 ∞ π d z Λ 2 + z 2 Λ 2 ab R ( a , b ; z ) , M h ( a , b ) = ∫ 0 ∞ π d z h ( z ) ab R ( a , b ; z ) = ∫ 0 1 M Λ d Λ.
Extended by continuity, both weights vanish on the axes. From α β / ( α + β ) ≤ α \alpha\beta/(\alpha+\beta)\le\alpha α β / ( α + β ) ≤ α and α Λ / ( α + Λ ) 2 ≤ 1 / 4 \alpha\Lambda/(\alpha+\Lambda)^2\le1/4 α Λ/ ( α + Λ ) 2 ≤ 1/4 , uniformly in Λ > 0 \Lambda>0 Λ > 0 ,
∣ ∂ a M ∣ , ∣ ∂ b M ∣ ≤ 9 8 (one-sided) , ∣ M ( a , b ) ∣ ≤ 9 8 min ( ∣ a ∣ , ∣ b ∣ ) , |\partial_aM|,\ |\partial_bM|\le\frac{9}{8}\ \text{(one-sided)},\qquad|M(a,b)|\le\frac{9}{8}\min(|a|,|b|), ∣ ∂ a M ∣ , ∣ ∂ b M ∣ ≤ 8 9 (one-sided) , ∣ M ( a , b ) ∣ ≤ 8 9 min ( ∣ a ∣ , ∣ b ∣ ) ,
0 ≤ M ( c , c ) ≤ 1 8 ∣ c ∣ , M = M Λ , M h . 0\le M(c,c)\le\frac{1}{8}|c|,\qquad M=M_\Lambda,M_h. 0 ≤ M ( c , c ) ≤ 8 1 ∣ c ∣ , M = M Λ , M h .
Since ( i q z − δ p ) − 1 = q − 1 ( i z − a p ) − 1 (iqz-\delta_p)^{-1}=q^{-1}(iz-a_p)^{-1} ( i q z − δ p ) − 1 = q − 1 ( i z − a p ) − 1 , Re [ ( i z − a ) ( i z − b ) ] − 1 = R ( a , b ; z ) \operatorname{Re}[(iz-a)(iz-b)]^{-1}=\mathcal{R}(a,b;z) Re [( i z − a ) ( i z − b ) ] − 1 = R ( a , b ; z ) and Δ p = sgn a p \Delta_p=\operatorname{sgn}a_p Δ p = sgn a p on the transition regions, Eq. (147) and the representations of M Λ M_\Lambda M Λ , M h M_h M h above give, for either weight w = Λ 2 / ( Λ 2 + z 2 ) w=\Lambda^2/(\Lambda^2+z^2) w = Λ 2 / ( Λ 2 + z 2 ) or w = h w=h w = h ,
∫ 0 ∞ d z π w ( z ) X ( q , q z ) = 1 q 2 ∫ d 3 p d 3 k ∣ Δ p Δ k ∣ M w ( a p , a k ) ∣ a p ∣ ∣ a k ∣ ∣ p − k ∣ 2 , a p = δ p / q , a k = δ k / q . \int_0^\infty\frac{\mathrm{d}z}{\pi}w(z)X(q,qz)=\frac{1}{q^2}\int\mathrm{d}^3p\,\mathrm{d}^3k\,\frac{|\Delta_p\Delta_k|M_w(a_p,a_k)}{|a_p||a_k||p-k|^2},\qquad a_p=\delta_p/q,\ a_k=\delta_k/q. ∫ 0 ∞ π d z w ( z ) X ( q , q z ) = q 2 1 ∫ d 3 p d 3 k ∣ a p ∣∣ a k ∣∣ p − k ∣ 2 ∣ Δ p Δ k ∣ M w ( a p , a k ) , a p = δ p / q , a k = δ k / q .
End caps. At fixed q q q and a = p ⋅ q ^ + q / 2 a=p\cdot\hat q+q/2 a = p ⋅ q ^ + q /2 ,
Δ p = 1 : 1 − ( a + q / 2 ) 2 < p ⊥ 2 < 1 − ( a − q / 2 ) 2 , d 3 p = 2 π q m q ( a ) d a , \Delta_p=1:\quad1-(a+q/2)^2<p_\perp^2<1-(a-q/2)^2,\qquad\mathrm{d}^3p=2\pi q\,m_q(a)\,\mathrm{d}a, Δ p = 1 : 1 − ( a + q /2 ) 2 < p ⊥ 2 < 1 − ( a − q /2 ) 2 , d 3 p = 2 π q m q ( a ) d a ,
with m q m_q m q of (S87), and p ↦ − p − q p\mapsto-p-q p ↦ − p − q maps it onto Δ p = − 1 \Delta_p=-1 Δ p = − 1 . The thin-shell parameterization of Sec. S10.1 (a → c p = p ^ ⋅ q ^ a\to c_p=\hat p\cdot\hat q a → c p = p ^ ⋅ q ^ , density 2 π q c p 2\pi q c_p 2 π q c p on 0 < c p < 1 0<c_p<1 0 < c p < 1 ) agrees with 2 π q m q ( a ) 2\pi q m_q(a) 2 π q m q ( a ) for a ≤ 1 − q / 2 a\le1-q/2 a ≤ 1 − q /2 . They differ only in the end cap 1 − q / 2 < a ≤ 1 + q / 2 1-q/2<a\le1+q/2 1 − q /2 < a ≤ 1 + q /2 , the polar cap about ± q ^ \pm\hat q ± q ^ . There, for q ≤ 1 q\le1 q ≤ 1 and ∣ p ∣ ≤ 2 |p|\le2 ∣ p ∣ ≤ 2 ,
volume: 2 π q ∫ 1 − q / 2 1 + q / 2 m q d a = π q 2 [ 1 + O ( q ) ] , weight: ∣ M ∣ ∣ a p ∣ ∣ a k ∣ ≤ 9 8 ∣ a p ∣ ≤ 9 4 ( a p ≥ 1 2 ) , \text{volume:}\quad2\pi q\int_{1-q/2}^{1+q/2}m_q\,\mathrm{d}a=\pi q^2[1+O(q)],\qquad\text{weight:}\quad\frac{|M|}{|a_p||a_k|}\le\frac{9}{8|a_p|}\le\frac{9}{4}\quad(a_p\ge\tfrac12), volume: 2 π q ∫ 1 − q /2 1 + q /2 m q d a = π q 2 [ 1 + O ( q )] , weight: ∣ a p ∣∣ a k ∣ ∣ M ∣ ≤ 8∣ a p ∣ 9 ≤ 4 9 ( a p ≥ 2 1 ) ,
Coulomb: ∫ 1 − q 1 + q ρ 2 d ρ ∫ d Ω k ∣ p − ρ k ^ ∣ 2 = ∫ 1 − q 1 + q 2 π ρ ∣ p ∣ log ρ + ∣ p ∣ ρ − ∣ p ∣ d ρ = O ( q log ( e / q ) ) , \text{Coulomb:}\quad\int_{1-q}^{1+q}\rho^2\,\mathrm{d}\rho\int\frac{\mathrm{d}\Omega_k}{|p-\rho\hat k|^2}=\int_{1-q}^{1+q}\frac{2\pi\rho}{|p|}\log\frac{\rho+|p|}{\rho-|p|}\,\mathrm{d}\rho=O(q\log(e/q)), Coulomb: ∫ 1 − q 1 + q ρ 2 d ρ ∫ ∣ p − ρ k ^ ∣ 2 d Ω k = ∫ 1 − q 1 + q ∣ p ∣ 2 π ρ log ρ − ∣ p ∣ ρ + ∣ p ∣ d ρ = O ( q log ( e / q )) ,
end caps: 9 4 q − 2 ⋅ π q 2 [ 1 + O ( q ) ] ⋅ O ( q log ( e / q ) ) = O ( q log ( e / q ) ) , \text{end caps:}\quad\frac{9}{4}q^{-2}\cdot\pi q^2[1+O(q)]\cdot O(q\log(e/q))=O(q\log(e/q)), end caps: 4 9 q − 2 ⋅ π q 2 [ 1 + O ( q )] ⋅ O ( q log ( e / q )) = O ( q log ( e / q )) ,
for either sign of Δ p \Delta_p Δ p and either momentum. The thin-shell end cap has a volume of the same order. With the chord distance d = ∣ p ^ − k ^ ∣ d=|\hat p-\hat k| d = ∣ p ^ − k ^ ∣ , angular measure 2 π d d d 2\pi d\,\mathrm{d}d 2 π d d d and the other shell of radial width O ( q ) O(q) O ( q ) , the thin-shell Coulomb factors of (S24) obey the same bound,
d − 2 on d > ϵ : O ( q log ( 1 / ϵ ) ) , [ d 2 + q 2 ( x − y ) 2 ] − 1 on the disk : O ( q log ( e / q ) ) , d^{-2}\ \text{on }d>\epsilon:\quad O(q\log(1/\epsilon)),\qquad[d^2+q^2(x-y)^2]^{-1}\ \text{on the disk}:\quad O(q\log(e/q)), d − 2 on d > ϵ : O ( q log ( 1/ ϵ )) , [ d 2 + q 2 ( x − y ) 2 ] − 1 on the disk : O ( q log ( e / q )) ,
the second after the radial average of Sec. S10.1.
Diagonal. Subtracting the coincident-direction value leaves O ( d ) O(d) O ( d ) , and d Ω = d Ω p \mathrm{d}\Omega=\mathrm{d}\Omega_p d Ω = d Ω p . With C b C_b C b independent of q q q and Λ \Lambda Λ (Notation) and the radial average of § 21.1 [Eq. (86) ],
∫ 0 1 O ( d ) 2 π d d d d 2 ≤ C b ∫ 0 1 d d , radial average = log 1 q − log ∣ c ∣ + 3 2 ( c = c p ) . \int_0^1 O(d)\frac{2\pi d\,\mathrm{d}d}{d^2}\leq C_b\int_0^1\mathrm{d}d,\qquad\text{radial average}=\log\frac{1}{q}-\log|c|+\frac{3}{2}\quad(c=c_p). ∫ 0 1 O ( d ) d 2 2 π d d d ≤ C b ∫ 0 1 d d , radial average = log q 1 − log ∣ c ∣ + 2 3 ( c = c p ) .
The diagonal bound M ( c , c ) ≤ ∣ c ∣ / 8 M(c,c)\leq|c|/8 M ( c , c ) ≤ ∣ c ∣/8 controls the grazing directions, where the width of the transition regions vanishes.
Curvature correction. The first line of Eq. (86) uses d − 2 d^{-2} d − 2 , and its second line uses the flat form [ d 2 + q 2 ( x − y ) 2 ] − 1 \left[d^2+q^2(x-y)^2\right]^{-1} [ d 2 + q 2 ( x − y ) 2 ] − 1 on the disk. With p = ( 1 + q x ) p ^ \mathbf{p}=(1+qx)\hat{\mathbf{p}} p = ( 1 + q x ) p ^ , k = ( 1 + q y ) k ^ \mathbf{k}=(1+qy)\hat{\mathbf{k}} k = ( 1 + q y ) k ^ , ∣ x ∣ , ∣ y ∣ ≤ 9 8 |x|,|y|\leq\frac{9}{8} ∣ x ∣ , ∣ y ∣ ≤ 8 9 on the transition regions and q ≤ 1 4 q\leq\frac{1}{4} q ≤ 4 1 ,
∣ p − k ∣ 2 = q 2 ( x − y ) 2 + ( 1 + q x ) ( 1 + q y ) d 2 , |\mathbf{p}-\mathbf{k}|^2=q^2(x-y)^2+(1+qx)(1+qy)d^2, ∣ p − k ∣ 2 = q 2 ( x − y ) 2 + ( 1 + q x ) ( 1 + q y ) d 2 ,
∣ ( 1 + q x ) ( 1 + q y ) − 1 ∣ ≤ 3 q , ( 1 + q x ) ( 1 + q y ) ≥ 1 2 , ∣ p − k ∣ 2 ≥ 1 2 d 2 , |(1+qx)(1+qy)-1|\leq3q,\qquad(1+qx)(1+qy)\geq\frac{1}{2},\qquad|\mathbf{p}-\mathbf{k}|^2\geq\frac{1}{2}d^2, ∣ ( 1 + q x ) ( 1 + q y ) − 1∣ ≤ 3 q , ( 1 + q x ) ( 1 + q y ) ≥ 2 1 , ∣ p − k ∣ 2 ≥ 2 1 d 2 ,
and therefore
∣ ∣ p − k ∣ − 2 − d − 2 ∣ ≤ 2 q 2 ( x − y ) 2 d 4 + 6 q d 2 ≤ 24 q d 3 ( q ≤ d ≤ 2 ) , \left||\mathbf{p}-\mathbf{k}|^{-2}-d^{-2}\right|\leq\frac{2q^2(x-y)^2}{d^4}+\frac{6q}{d^2}\leq\frac{24q}{d^3}\qquad(q\leq d\leq2), ∣ p − k ∣ − 2 − d − 2 ≤ d 4 2 q 2 ( x − y ) 2 + d 2 6 q ≤ d 3 24 q ( q ≤ d ≤ 2 ) ,
∣ ∣ p − k ∣ − 2 − [ d 2 + q 2 ( x − y ) 2 ] − 1 ∣ ≤ 6 q d 2 . \left||\mathbf{p}-\mathbf{k}|^{-2}-\left[d^2+q^2(x-y)^2\right]^{-1}\right|\leq\frac{6q}{d^2}. ∣ p − k ∣ − 2 − [ d 2 + q 2 ( x − y ) 2 ] − 1 ≤ d 2 6 q .
After the radial averages the weight is at most 9 8 \frac{9}{8} 8 9 , and the correction is
ϵ < d ≤ 2 : C b q ∫ ϵ 2 d d d 2 = O ( q / ϵ ) (first bound) , \epsilon<d\leq2:\quad C_bq\int_{\epsilon}^{2}\frac{\mathrm{d}d}{d^2}=O(q/\epsilon)\quad\text{(first bound)}, ϵ < d ≤ 2 : C b q ∫ ϵ 2 d 2 d d = O ( q / ϵ ) (first bound) ,
q < d < ϵ : C b q ∫ q ϵ d d d = O ( q log ( ϵ / q ) ) (second bound, disk) , q<d<\epsilon:\quad C_bq\int_q^\epsilon\frac{\mathrm{d}d}{d}=O(q\log(\epsilon/q))\quad\text{(second bound, disk)}, q < d < ϵ : C b q ∫ q ϵ d d d = O ( q log ( ϵ / q )) (second bound, disk) ,
d < q : ∫ 0 q d d d ∬ d x d y [ q 2 ( x − y ) 2 + d 2 / 2 ] − 1 = O ( 1 ) . d<q:\quad\int_0^q d\,\mathrm{d}d\iint\mathrm{d}x\,\mathrm{d}y\,[q^2(x-y)^2+d^2/2]^{-1}=O(1). d < q : ∫ 0 q d d d ∬ d x d y [ q 2 ( x − y ) 2 + d 2 /2 ] − 1 = O ( 1 ) .
For d < q d<q d < q both the exact and the approximate Coulomb factors are controlled by the radial average of § 21.1 . Replacing a p a_p a p by c p c_p c p , with ∣ a p − c p ∣ ≤ q |a_p-c_p|\leq q ∣ a p − c p ∣ ≤ q ,
∣ δ M ∣ ≤ 9 8 ( ∣ a p − c p ∣ + ∣ a k − c k ∣ ) = O ( q ) , ∫ d x / ∣ a p ∣ ∫ d x / ∣ a p ∣ = O ( q / ∣ c p ∣ ) , |\delta M|\leq\frac{9}{8}(|a_p-c_p|+|a_k-c_k|)=O(q),\qquad\frac{\int\mathrm{d}x/|a_p|}{\int\mathrm{d}x/|a_p|}=O(q/|c_p|), ∣ δ M ∣ ≤ 8 9 ( ∣ a p − c p ∣ + ∣ a k − c k ∣ ) = O ( q ) , ∫ d x /∣ a p ∣ ∫ d x /∣ a p ∣ = O ( q /∣ c p ∣ ) ,
with the radial integral over the exact interval. Together these changes alter the weight by O ( q ) O(q) O ( q ) uniformly. For ∣ c p ∣ ≲ q |c_p|\lesssim q ∣ c p ∣ ≲ q the bound ∣ M ∣ ≤ 9 8 min ( ∣ a ∣ , ∣ b ∣ ) |M|\leq\frac{9}{8}\min(|a|,|b|) ∣ M ∣ ≤ 8 9 min ( ∣ a ∣ , ∣ b ∣ ) is used instead. These bounds give
∫ 0 ∞ d z π h ( z ) X ( q , q z ) = 2 π log ( 1 / q ) ∫ d Ω M h ( c , c ) + O ( 1 ) , \int_0^\infty\frac{\mathrm{d}z}{\pi}h(z)X(q,qz)=2\pi\log(1/q)\int\mathrm{d}\Omega\,M_h(c,c)+O(1), ∫ 0 ∞ π d z h ( z ) X ( q , q z ) = 2 π log ( 1/ q ) ∫ d Ω M h ( c , c ) + O ( 1 ) ,
2 π ∫ d Ω M h ( c , c ) = ∫ 0 ∞ d z π h a x = A , (148) 2\pi\int\mathrm{d}\Omega\,M_h(c,c)=\int_0^\infty\frac{\mathrm{d}z}{\pi}h\,a_x=A, \tag*{(148)} 2 π ∫ d Ω M h ( c , c ) = ∫ 0 ∞ π d z h a x = A , (148)
with a x a_x a x of (S30) and A A A of (S31).
Absolute frequency bound. The spectral interchange also needs, by Cauchy–Schwarz,
∫ 0 ∞ d z π ∣ a b R ( a , b ; z ) ∣ ≤ ∣ a b ∣ π ( ∫ 0 ∞ d z z 2 + a 2 ) 1 / 2 ( ∫ 0 ∞ d z z 2 + b 2 ) 1 / 2 = ∣ a b ∣ 2 . \int_0^\infty\frac{\mathrm{d}z}{\pi}|ab\mathcal{R}(a,b;z)|\leq\frac{|ab|}{\pi}\left(\int_0^\infty\frac{\mathrm{d}z}{z^2+a^2}\right)^{1/2}\left(\int_0^\infty\frac{\mathrm{d}z}{z^2+b^2}\right)^{1/2}=\frac{\sqrt{|ab|}}{2}. ∫ 0 ∞ π d z ∣ ab R ( a , b ; z ) ∣ ≤ π ∣ ab ∣ ( ∫ 0 ∞ z 2 + a 2 d z ) 1/2 ( ∫ 0 ∞ z 2 + b 2 d z ) 1/2 = 2 ∣ ab ∣ .
Each shell has radial width q ∣ a ∣ q|a| q ∣ a ∣ , so the shell integrand is a b R ab\mathcal{R} ab R , and the Coulomb integral between the two shells gives
∫ 0 ∞ ∣ X ( q , q z ) ∣ d z = O ( log ( e / q ) ) . \int_0^\infty|X(q,qz)|\,\mathrm{d}z=O(\log(e/q)). ∫ 0 ∞ ∣ X ( q , q z ) ∣ d z = O ( log ( e / q )) .
Lindhard function. For 0 < q < 1 0<q<1 0 < q < 1 , L ( q , q z ) = 1 4 π ∫ d 3 p Δ p δ p / ( δ p 2 + q 2 z 2 ) \mathcal{L}(q,qz)=\frac{1}{4\pi}\int\mathrm{d}^{3}p\,\Delta_p\delta_p/(\delta_p^2+q^2z^2) L ( q , q z ) = 4 π 1 ∫ d 3 p Δ p δ p / ( δ p 2 + q 2 z 2 ) (§ 20.2 ). The region Δ p = − 1 \Delta_p=-1 Δ p = − 1 contributes as much as the region Δ p = 1 \Delta_p=1 Δ p = 1 , by p ↦ − p − q \mathbf{p}\mapsto-\mathbf{p}-\mathbf{q} p ↦ − p − q , and d 3 p = 2 π q m q ( a ) d a \mathrm{d}^{3}p=2\pi q m_q(a)\,\mathrm{d}a d 3 p = 2 π q m q ( a ) d a there, so
L ( q , q z ) = ∫ 0 1 + q / 2 a m q ( a ) a 2 + z 2 d a , m q ( a ) = { a , 0 ≤ a ≤ 1 − q / 2 , [ 1 − ( a − q / 2 ) 2 ] / ( 2 q ) , 1 − q / 2 < a ≤ 1 + q / 2. (149) \mathcal{L}(q,qz)=\int_{0}^{1+q/2}\frac{a m_q(a)}{a^2+z^2}\,\mathrm{d}a,
\qquad
m_q(a)=
\begin{cases}
a, & 0\leq a\leq1-q/2,\\
[1-(a-q/2)^2]/(2q), & 1-q/2<a\leq1+q/2.
\end{cases} \tag*{(149)} L ( q , q z ) = ∫ 0 1 + q /2 a 2 + z 2 a m q ( a ) d a , m q ( a ) = { a , [ 1 − ( a − q /2 ) 2 ] / ( 2 q ) , 0 ≤ a ≤ 1 − q /2 , 1 − q /2 < a ≤ 1 + q /2. (149)
With h ( z ) = ∫ 0 1 a 2 d a / ( a 2 + z 2 ) h(z)=\int_{0}^{1}a^2\,\mathrm{d}a/(a^2+z^2) h ( z ) = ∫ 0 1 a 2 d a / ( a 2 + z 2 ) (main-text § 2.2 ) and 1 a < 1 1_{a<1} 1 a < 1 the indicator of a < 1 a<1 a < 1 ,
L ( q , q z ) − h ( z ) = ∫ 1 − q / 2 1 + q / 2 [ m q ( a ) − a 1 a < 1 ] a d a a 2 + z 2 , a a 2 + z 2 ≤ C b 1 + z 2 ( 1 2 ≤ a ≤ 3 2 ) , \mathcal{L}(q,qz)-h(z)=\int_{1-q/2}^{1+q/2}[m_q(a)-a1_{a<1}]\frac{a\,\mathrm{d}a}{a^2+z^2},
\qquad
\frac{a}{a^2+z^2}\leq\frac{C_b}{1+z^2}\quad\left(\frac{1}{2}\leq a\leq\frac{3}{2}\right), L ( q , q z ) − h ( z ) = ∫ 1 − q /2 1 + q /2 [ m q ( a ) − a 1 a < 1 ] a 2 + z 2 a d a , a 2 + z 2 a ≤ 1 + z 2 C b ( 2 1 ≤ a ≤ 2 3 ) ,
and therefore
∫ ∣ m q ( a ) − a 1 a < 1 ∣ d a = q / 4 , ∣ L ( q , q z ) − h ( z ) ∣ ≤ C b q 1 + z 2 . (150) \int\lvert m_q(a)-a1_{a<1}\rvert\,\mathrm{d}a=q/4,
\qquad
\lvert\mathcal{L}(q,qz)-h(z)\rvert\leq\frac{C_bq}{1+z^2}. \tag*{(150)} ∫ ∣ m q ( a ) − a 1 a < 1 ∣ d a = q /4 , ∣ L ( q , q z ) − h ( z )∣ ≤ 1 + z 2 C b q . (150)
With the absolute frequency bound, replacing h h h by L \mathcal{L} L changes the frequency-integrated shell expansion by
∣ ∫ 0 ∞ d z π [ L ( q , q z ) − h ( z ) ] X ( q , q z ) ∣ ≤ sup z ∣ L − h ∣ ∫ 0 ∞ d z π ∣ X ( q , q z ) ∣ = O ( q log ( e / q ) ) , \left\lvert\int_{0}^{\infty}\frac{\mathrm{d}z}{\pi}[\mathcal{L}(q,qz)-h(z)]X(q,qz)\right\rvert
\leq\sup_z\lvert\mathcal{L}-h\rvert\int_{0}^{\infty}\frac{\mathrm{d}z}{\pi}\lvert X(q,qz)\rvert
=O(q\log(\mathrm{e}/q)), ∫ 0 ∞ π d z [ L ( q , q z ) − h ( z )] X ( q , q z ) ≤ z sup ∣ L − h ∣ ∫ 0 ∞ π d z ∣ X ( q , q z )∣ = O ( q log ( e / q )) ,
which in the window integral ((S29)) costs q T F 2 ∫ q T F 1 ( d q / q ) O ( q log ( e / q ) ) = O ( q T F 2 ) = O ( r s ) q_{\mathrm{TF}}^2\int_{q_{\mathrm{TF}}}^{1}(\mathrm{d}q/q)\,O(q\log(\mathrm{e}/q))=O(q_{\mathrm{TF}}^2)=O(r_s) q TF 2 ∫ q TF 1 ( d q / q ) O ( q log ( e / q )) = O ( q TF 2 ) = O ( r s ) .
Two formulas used elsewhere. The signed occupation integral Z ( q ) Z(q) Z ( q ) of ((S36)) has the Fourier representation
Z ( q ) = 2 π 2 ( 2 π ) 3 ∫ d 3 t ∣ χ ^ B ( t ) ∣ 2 t 2 2 [ 1 − cos ( q t z ) ] , χ ^ B ( t ) = 4 π sin t − t cos t t 3 , (151) Z(q)=\frac{2\pi^2}{(2\pi)^3}\int\mathrm{d}^{3}t\,\frac{\lvert\widehat{\chi}_{\mathrm{B}}(t)\rvert^2}{t^2}2[1-\cos(qt_z)],
\qquad
\widehat{\chi}_{\mathrm{B}}(t)=4\pi\frac{\sin t-t\cos t}{t^3}, \tag*{(151)} Z ( q ) = ( 2 π ) 3 2 π 2 ∫ d 3 t t 2 ∣ χ B ( t ) ∣ 2 2 [ 1 − cos ( q t z )] , χ B ( t ) = 4 π t 3 sin t − t cos t , (151)
where t t t is conjugate to p \mathbf{p} p (not the scaled momentum t = q / ε t=q/\varepsilon t = q / ε ), t = ∣ t ∣ t=\lvert t\rvert t = ∣ t ∣ , t z = t ⋅ q ^ t_z=t\cdot\hat{\mathbf{q}} t z = t ⋅ q ^ and χ B \chi_{\mathrm{B}} χ B is the indicator of the unit Fermi ball. It follows from the convolution theorem, where → \to → denotes the Fourier transform,
Δ p = χ B ( p ) − χ B ( p + q ) → χ ^ B ( t ) ( 1 − e i t ⋅ q ) , ∣ 1 − e i t ⋅ q ∣ 2 = 2 [ 1 − cos ( q t z ) ] , \Delta_p=\chi_{\mathrm{B}}(p)-\chi_{\mathrm{B}}(p+q)\ \to\ \widehat{\chi}_{\mathrm{B}}(t)(1-\mathrm{e}^{\mathrm{i}t\cdot q}),
\qquad
\lvert1-\mathrm{e}^{\mathrm{i}t\cdot q}\rvert^2=2[1-\cos(qt_z)], Δ p = χ B ( p ) − χ B ( p + q ) → χ B ( t ) ( 1 − e i t ⋅ q ) , ∣ 1 − e i t ⋅ q ∣ 2 = 2 [ 1 − cos ( q t z )] ,
∫ d 3 x e − i t ⋅ x ∣ x ∣ − 2 = 2 π 2 t . \int\mathrm{d}^{3}x\,\mathrm{e}^{-\mathrm{i}t\cdot x}\lvert x\rvert^{-2}=\frac{2\pi^2}{t}. ∫ d 3 x e − i t ⋅ x ∣ x ∣ − 2 = t 2 π 2 .
The frequency integral behind κ ( Λ ) \kappa(\Lambda) κ ( Λ ) in ((S26)), with c c c the diagonal excitation energy c p c_p c p there, is
∫ 0 ∞ Λ 2 ( c 2 − z 2 ) d z ( Λ 2 + z 2 ) ( c 2 + z 2 ) 2 = π Λ 2 ( Λ + c ) 2 = π Λ 2 J ( Λ ; c , c ) , c > 0 , (152) \int_{0}^{\infty}\frac{\Lambda^2(c^2-z^2)\,\mathrm{d}z}{(\Lambda^2+z^2)(c^2+z^2)^2}
=\frac{\pi\Lambda}{2(\Lambda+c)^2}=\pi\Lambda^2J(\Lambda;c,c),
\qquad c>0, \tag*{(152)} ∫ 0 ∞ ( Λ 2 + z 2 ) ( c 2 + z 2 ) 2 Λ 2 ( c 2 − z 2 ) d z = 2 ( Λ + c ) 2 π Λ = π Λ 2 J ( Λ ; c , c ) , c > 0 , (152)
with J J J of ((S64)).
Low-density expansion with dynamical screening
This section derives the large-momentum kernel ((48) ), the screening decomposition ((21) ) and the coefficient ((51) ), and repeats the calculation for the self-energy pair,
e k i t e = A 1 q ∗ + O ( q ∗ − 3 / 2 ) , e k i t e + e S E = O ( q ∗ − 3 / 2 ) . e_{\mathrm{kite}}=\frac{A_1}{q_*}+O(q_*^{-3/2}),
\qquad
e_{\mathrm{kite}}+e_{\mathrm{SE}}=O(q_*^{-3/2}). e kite = q ∗ A 1 + O ( q ∗ − 3/2 ) , e kite + e SE = O ( q ∗ − 3/2 ) .
The radial kernel e Λ ( q ) e_{\Lambda}(q) e Λ ( q ) is defined in ((46) ). As in the main text, the relevant momentum transfers are q ∼ q ∗ ≫ 1 q\sim q_*\gg1 q ∼ q ∗ ≫ 1 .
Large-momentum expansion of the radial kernel e Λ e_\Lambda e Λ
Coulomb integrals. For q > 2 q>2 q > 2 each same-ball region of main-text § 3.2 uses ∫ B ∣ p − k ∣ − 2 d 3 k = 4 π f L ( p ) \int_{\mathbb{B}}|\mathbf{p}-\mathbf{k}|^{-2}d^3k=4\pi f_L(p) ∫ B ∣ p − k ∣ − 2 d 3 k = 4 π f L ( p ) and
∫ 0 1 r 2 f L ( r ) d r = 1 6 + 1 4 ∫ 0 1 r ( 1 − r 2 ) log 1 + r 1 − r d r = 1 6 + 1 12 = 1 4 , (153) \int_0^1 r^2 f_L(r)\,dr=\frac{1}{6}+\frac{1}{4}\int_0^1r(1-r^2)\log\frac{1+r}{1-r}\,dr=\frac{1}{6}+\frac{1}{12}=\frac{1}{4}, \tag*{(153)} ∫ 0 1 r 2 f L ( r ) d r = 6 1 + 4 1 ∫ 0 1 r ( 1 − r 2 ) log 1 − r 1 + r d r = 6 1 + 12 1 = 4 1 , (153)
hence
same ball: ∬ B 2 d 3 p d 3 k ∣ p − k ∣ 2 = 16 π 2 ⋅ 1 4 = 4 π 2 , different balls: ∣ p − k ∣ ≥ q − 2 , ∬ d 3 p d 3 k ∣ p − k ∣ 2 ≤ ( 4 π 3 ) 2 1 ( q − 2 ) 2 . \begin{aligned}
\text{same ball:}\quad&\iint_{\mathbb{B}^2}\frac{d^3p\,d^3k}{|\mathbf{p}-\mathbf{k}|^2}=16\pi^2\cdot\frac{1}{4}=4\pi^2,\\
\text{different balls:}\quad&|\mathbf{p}-\mathbf{k}|\ge q-2,\qquad\iint\frac{d^3p\,d^3k}{|\mathbf{p}-\mathbf{k}|^2}\le\left(\frac{4\pi}{3}\right)^2\frac{1}{(q-2)^2}.
\end{aligned} same ball: different balls: ∬ B 2 ∣ p − k ∣ 2 d 3 p d 3 k = 16 π 2 ⋅ 4 1 = 4 π 2 , ∣ p − k ∣ ≥ q − 2 , ∬ ∣ p − k ∣ 2 d 3 p d 3 k ≤ ( 3 4 π ) 2 ( q − 2 ) 2 1 .
Frequency integral. With ν = q 2 u \nu=q^2u ν = q 2 u , Λ = q w Λ \Lambda=qw_\Lambda Λ = q w Λ , δ = ± q 2 / 2 + O ( q ) \delta=\pm q^2/2+O(q) δ = ± q 2 /2 + O ( q ) and p l o c ( w ) = 2 π w / ( 1 + 2 w ) 2 p_{\mathrm{loc}}(w)=2\pi w/(1+2w)^2 p loc ( w ) = 2 π w / ( 1 + 2 w ) 2 , a same-ball region at δ p = δ k = ± q 2 / 2 \delta_p=\delta_k=\pm q^2/2 δ p = δ k = ± q 2 /2 has
q 4 Re 1 ( i ν − δ p ) ( i ν − δ k ) = 1 / 4 − u 2 ( u 2 + 1 / 4 ) 2 , Λ 2 Λ 2 + ν 2 / q 2 = w Λ 2 w Λ 2 + u 2 . q^4\operatorname{Re}\frac{1}{(i\nu-\delta_p)(i\nu-\delta_k)}=\frac{1/4-u^2}{(u^2+1/4)^2},\qquad\frac{\Lambda^2}{\Lambda^2+\nu^2/q^2}=\frac{w_\Lambda^2}{w_\Lambda^2+u^2}. q 4 Re ( i ν − δ p ) ( i ν − δ k ) 1 = ( u 2 + 1/4 ) 2 1/4 − u 2 , Λ 2 + ν 2 / q 2 Λ 2 = w Λ 2 + u 2 w Λ 2 .
The frequency integral follows by differentiating an elementary Lorentzian integral:
1 / 4 − u 2 ( u 2 + 1 / 4 ) 2 = − ∂ b b u 2 + b 2 ∣ b = 1 / 2 , ∫ 0 ∞ d u ( u 2 + w Λ 2 ) ( u 2 + b 2 ) = π 2 w Λ b ( w Λ + b ) , \frac{1/4-u^2}{(u^2+1/4)^2}=-\left.\partial_b\frac{b}{u^2+b^2}\right|_{b=1/2},\qquad\int_0^\infty\frac{du}{(u^2+w_\Lambda^2)(u^2+b^2)}=\frac{\pi}{2w_\Lambda b(w_\Lambda+b)}, ( u 2 + 1/4 ) 2 1/4 − u 2 = − ∂ b u 2 + b 2 b b = 1/2 , ∫ 0 ∞ ( u 2 + w Λ 2 ) ( u 2 + b 2 ) d u = 2 w Λ b ( w Λ + b ) π ,
∫ 0 ∞ w Λ 2 ( 1 / 4 − u 2 ) ( u 2 + w Λ 2 ) ( u 2 + 1 / 4 ) 2 d u = p l o c ( w Λ ) , 32 π 2 A 0 = 3 π 3 . (154) \int_0^\infty\frac{w_\Lambda^2(1/4-u^2)}{(u^2+w_\Lambda^2)(u^2+1/4)^2}\,du=p_{\mathrm{loc}}(w_\Lambda),\qquad32\pi^2A_0=\frac{3}{\pi^3}. \tag*{(154)} ∫ 0 ∞ ( u 2 + w Λ 2 ) ( u 2 + 1/4 ) 2 w Λ 2 ( 1/4 − u 2 ) d u = p loc ( w Λ ) , 32 π 2 A 0 = π 3 3 . (154)
The third identity is − w Λ 2 ∂ b [ π / ( 2 w Λ ( w Λ + b ) ) ] ∣ b = 1 / 2 = π w Λ / [ 2 ( w Λ + 1 / 2 ) 2 ] = p l o c ( w Λ ) -w_\Lambda^2\left.\partial_b[\pi/(2w_\Lambda(w_\Lambda+b))]\right|_{b=1/2}=\pi w_\Lambda/[2(w_\Lambda+1/2)^2]=p_{\mathrm{loc}}(w_\Lambda) − w Λ 2 ∂ b [ π / ( 2 w Λ ( w Λ + b ))] ∣ b = 1/2 = π w Λ / [ 2 ( w Λ + 1/2 ) 2 ] = p loc ( w Λ ) . The prefactor 3 / ( π 3 q 2 ) 3/(\pi^3q^2) 3/ ( π 3 q 2 ) of (48) collects the factor 4 π A 0 4\pi A_0 4 π A 0 of (46), the Coulomb integrals of the two same-ball regions and d ν = q 2 d u d\nu=q^2du d ν = q 2 d u ,
e Λ ( q ) ≃ 4 π A 0 ⋅ 2 ⋅ 4 π 2 ∫ 0 ∞ q 2 d u π w Λ 2 w Λ 2 + u 2 1 / 4 − u 2 q 4 ( u 2 + 1 / 4 ) 2 = 32 π 2 A 0 q 2 p l o c ( w Λ ) = 3 π 3 q 2 p l o c ( w Λ ) . e_\Lambda(q)\simeq4\pi A_0\cdot2\cdot4\pi^2\int_0^\infty\frac{q^2\,du}{\pi}\frac{w_\Lambda^2}{w_\Lambda^2+u^2}\frac{1/4-u^2}{q^4(u^2+1/4)^2}=\frac{32\pi^2A_0}{q^2}p_{\mathrm{loc}}(w_\Lambda)=\frac{3}{\pi^3q^2}p_{\mathrm{loc}}(w_\Lambda). e Λ ( q ) ≃ 4 π A 0 ⋅ 2 ⋅ 4 π 2 ∫ 0 ∞ π q 2 d u w Λ 2 + u 2 w Λ 2 q 4 ( u 2 + 1/4 ) 2 1/4 − u 2 = q 2 32 π 2 A 0 p loc ( w Λ ) = π 3 q 2 3 p loc ( w Λ ) .
Remainders. The O ( q ) O(q) O ( q ) correction to δ \delta δ gives O ( q − 3 / ( 1 + w Λ ) ) O(q^{-3}/(1+w_\Lambda)) O ( q − 3 / ( 1 + w Λ )) . The different-ball regions and the bare term e ∞ e_\infty e ∞ are O ( q − 4 ) O(q^{-4}) O ( q − 4 ) . These estimates give (48) uniformly for w Λ w_\Lambda w Λ bounded away from zero.
Large-momentum approximation of RPA screening
Lindhard function. Let q ∗ 4 = 4 q T F 2 / 3 q_*^4=4q_{\mathrm{TF}}^2/3 q ∗ 4 = 4 q TF 2 /3 as in main-text Eq. (20) , u = ν / q 2 u=\nu/q^2 u = ν / q 2 and c ^ = ( q ∗ / q ) 4 / ( 1 + 4 u 2 ) \hat{c}=(q_*/q)^4/(1+4u^2) c ^ = ( q ∗ / q ) 4 / ( 1 + 4 u 2 ) , the leading large-q q q screening ratio of main-text Eq. (21) . The Lindhard function has the large-momentum expansion
L ( q , q 2 u ) = 4 3 q 2 ( 1 + 4 u 2 ) [ 1 + g 4 ( u ) q 2 + O ( q − 4 ) ] , (155) \mathcal{L}(q,q^2u)=\frac{4}{3q^2(1+4u^2)}\left[1+\frac{g_4(u)}{q^2}+O(q^{-4})\right], \tag*{(155)} L ( q , q 2 u ) = 3 q 2 ( 1 + 4 u 2 ) 4 [ 1 + q 2 g 4 ( u ) + O ( q − 4 ) ] , (155)
with g 4 ( u ) = − 4 ( 12 u 2 − 1 ) / [ 5 ( 1 + 4 u 2 ) 2 ] g_4(u)=-4(12u^2-1)/[5(1+4u^2)^2] g 4 ( u ) = − 4 ( 12 u 2 − 1 ) / [ 5 ( 1 + 4 u 2 ) 2 ] . It is obtained by expanding the log function in powers of 1 / q 1/q 1/ q at fixed u u u and integrating over the radius k k k of the occupied momentum in the Lindhard function of Ref. [1 ] at ω = i ν \omega=i\nu ω = i ν ,
L ( q , q 2 u ) = 1 2 q ∫ 0 1 k log ϖ + k / q + k 2 / q 2 ϖ − k / q + k 2 / q 2 d k , ϖ = u 2 + 1 4 . (156) \mathcal{L}(q,q^2u)=\frac{1}{2q}\int_0^1 k\log\frac{\varpi+k/q+k^2/q^2}{\varpi-k/q+k^2/q^2}\,dk,\qquad\varpi=u^2+\frac{1}{4}. \tag*{(156)} L ( q , q 2 u ) = 2 q 1 ∫ 0 1 k log ϖ − k / q + k 2 / q 2 ϖ + k / q + k 2 / q 2 d k , ϖ = u 2 + 4 1 . (156)
At leading order
log ϖ ‾ + k / q + k 2 / q 2 ϖ ‾ − k / q + k 2 / q 2 = 2 k q ϖ ‾ + O ( q − 3 ) , \log\frac{\overline{\varpi}+k/q+k^2/q^2}{\overline{\varpi}-k/q+k^2/q^2}=\frac{2k}{q\overline{\varpi}}+O(q^{-3}), log ϖ − k / q + k 2 / q 2 ϖ + k / q + k 2 / q 2 = q ϖ 2 k + O ( q − 3 ) ,
L ( q , q 2 u ) = 1 3 q 2 ϖ ‾ + O ( q − 4 ) = 4 3 q 2 ( 1 + 4 u 2 ) + O ( q − 4 ) . \mathcal{L}(q,q^2u)=\frac{1}{3q^2\overline{\varpi}}+O(q^{-4})=\frac{4}{3q^2(1+4u^2)}+O(q^{-4}). L ( q , q 2 u ) = 3 q 2 ϖ 1 + O ( q − 4 ) = 3 q 2 ( 1 + 4 u 2 ) 4 + O ( q − 4 ) .
The expansion (S93) is uniform in u u u because ϖ ‾ ≥ 1 / 4 \overline{\varpi}\ge1/4 ϖ ≥ 1/4 .
Single Lorentzian. Since c = − v Π 0 = c ^ [ 1 + O ( q − 2 ) ] c=-v\Pi_0=\hat{c}[1+O(q^{-2})] c = − v Π 0 = c ^ [ 1 + O ( q − 2 )] ,
1 1 + λ c ^ = 1 + 4 u 2 1 + 4 u 2 + λ ( q ∗ / q ) 4 = 1 − λ q ∗ 4 / 4 ν 2 + ω λ ( q ) 2 , ω λ ( q ) 2 = q 4 + λ q ∗ 4 4 , (157) \frac{1}{1+\lambda\hat{c}}=\frac{1+4u^2}{1+4u^2+\lambda(q_*/q)^4}=1-\frac{\lambda q_*^4/4}{\nu^2+\omega_\lambda(q)^2},\qquad\omega_\lambda(q)^2=\frac{q^4+\lambda q_*^4}{4}, \tag*{(157)} 1 + λ c ^ 1 = 1 + 4 u 2 + λ ( q ∗ / q ) 4 1 + 4 u 2 = 1 − ν 2 + ω λ ( q ) 2 λ q ∗ 4 /4 , ω λ ( q ) 2 = 4 q 4 + λ q ∗ 4 , (157)
and, with the remainder ς λ = 1 / ( 1 + λ c ) − 1 / ( 1 + λ c ^ ) \varsigma_\lambda=1/(1+\lambda c)-1/(1+\lambda\hat{c}) ς λ = 1/ ( 1 + λ c ) − 1/ ( 1 + λ c ^ ) of the single-Lorentzian form, the inverse dielectric function is
1 ϵ λ ( q , i ν ) = 1 − θ λ ( q ) Λ λ ( q ) 2 Λ λ ( q ) 2 + z 2 + ς λ ( q , u ) , \frac{1}{\epsilon_\lambda(q,i\nu)}=1-\theta_\lambda(q)\frac{\Lambda_\lambda(q)^2}{\Lambda_\lambda(q)^2+z^2}+\varsigma_\lambda(q,u), ϵ λ ( q , i ν ) 1 = 1 − θ λ ( q ) Λ λ ( q ) 2 + z 2 Λ λ ( q ) 2 + ς λ ( q , u ) ,
ς λ = O ( 1 q 2 λ c ^ ( 1 + λ c ^ ) 2 ) , θ λ ( q ) = λ q ∗ 4 q 4 + λ q ∗ 4 , \varsigma_\lambda=O\left(\frac{1}{q^2}\frac{\lambda\hat{c}}{(1+\lambda\hat{c})^2}\right),\qquad\theta_\lambda(q)=\frac{\lambda q_*^4}{q^4+\lambda q_*^4}, ς λ = O ( q 2 1 ( 1 + λ c ^ ) 2 λ c ^ ) , θ λ ( q ) = q 4 + λ q ∗ 4 λ q ∗ 4 ,
Λ λ ( q ) = q 2 1 + λ ( q ∗ / q ) 4 . (158) \Lambda_\lambda(q)=\frac{q}{2}\sqrt{1+\lambda(q_*/q)^4}. \tag*{(158)} Λ λ ( q ) = 2 q 1 + λ ( q ∗ / q ) 4 . (158)
Multiplying by λ v \lambda v λ v gives (21) with
ς λ = − λ ( c − c ^ ) ( 1 + λ c ) ( 1 + λ c ^ ) , (159) \varsigma_\lambda=-\frac{\lambda(c-\hat{c})}{(1+\lambda c)(1+\lambda\hat{c})}, \tag*{(159)} ς λ = − ( 1 + λ c ) ( 1 + λ c ^ ) λ ( c − c ^ ) , (159)
and the single-Lorentzian form of the retarded line has poles at ω = ± ω λ ( q ) − i 0 \omega=\pm\omega_\lambda(q)-i0 ω = ± ω λ ( q ) − i 0 . The two Lorentzian forms agree,
λ q ∗ 4 / 4 ν 2 + ω λ 2 = θ λ Λ λ 2 Λ λ 2 + z 2 , ω λ = q Λ λ , z = ν / q . \frac{\lambda q_*^4/4}{\nu^2+\omega_\lambda^2}=\theta_\lambda\frac{\Lambda_\lambda^2}{\Lambda_\lambda^2+z^2},\qquad\omega_\lambda=q\Lambda_\lambda,\qquad z=\nu/q. ν 2 + ω λ 2 λ q ∗ 4 /4 = θ λ Λ λ 2 + z 2 Λ λ 2 , ω λ = q Λ λ , z = ν / q .
At fixed λ > 0 \lambda>0 λ > 0 ,
q ≫ q ∗ : ω λ ∼ q 2 / 2 , θ λ → 0 , 1 ≪ q ≪ q ∗ : ω λ ∼ λ q T F 2 / 3 , θ λ → 1. q\gg q_*:\quad\omega_\lambda\sim q^2/2,\quad\theta_\lambda\to0,\qquad1\ll q\ll q_*:\quad\omega_\lambda\sim\sqrt{\lambda q_{\mathrm{TF}}^2/3},\quad\theta_\lambda\to1. q ≫ q ∗ : ω λ ∼ q 2 /2 , θ λ → 0 , 1 ≪ q ≪ q ∗ : ω λ ∼ λ q TF 2 /3 , θ λ → 1.
Low-density coefficient and remainder estimates
Inserting main-text Eq. (21) into e k i t e = 2 ∫ 0 1 K [ W λ ] d λ e_{\mathrm{kite}}=2\int_0^1\mathcal{K}[W_\lambda]\,d\lambda e kite = 2 ∫ 0 1 K [ W λ ] d λ and collecting in the remainder the terms bounded in § 29.2 gives (main text, before Eq. (51) )
e k i t e ( r s ) = − 2 ∫ 0 1 λ d λ ∫ 0 ∞ d q θ λ ( q ) [ e Λ λ ( q ) − e ∞ ( q ) ] + O ( q ∗ − 2 ) = A 1 q ∗ + O ( q ∗ − 3 / 2 ) , (160) e_{\mathrm{kite}}(r_s)=-2\int_0^1\lambda\,d\lambda\int_0^\infty dq\,\theta_\lambda(q)[e_{\Lambda_\lambda}(q)-e_\infty(q)]+O(q_*^{-2})
=\frac{A_1}{q_*}+O(q_*^{-3/2}), \tag*{(160)} e kite ( r s ) = − 2 ∫ 0 1 λ d λ ∫ 0 ∞ d q θ λ ( q ) [ e Λ λ ( q ) − e ∞ ( q )] + O ( q ∗ − 2 ) = q ∗ A 1 + O ( q ∗ − 3/2 ) , (160)
with A 1 A_1 A 1 of (51). Its closed form follows from three substitutions.
First substitution. With x = q / q ∗ x=q/q_* x = q / q ∗ and w Λ , λ ( x ) w_{\Lambda,\lambda}(x) w Λ , λ ( x ) of main-text Eq. (51) ,
x = λ 1 / 4 y : w Λ , λ = 1 2 1 + y − 4 , λ x 4 + λ = 1 y 4 + 1 , d x x 2 = λ − 1 / 4 d y y 2 . x=\lambda^{1/4}y:\qquad w_{\Lambda,\lambda}=\frac{1}{2}\sqrt{1+y^{-4}},\qquad\frac{\lambda}{x^4+\lambda}=\frac{1}{y^4+1},\qquad\frac{dx}{x^2}=\lambda^{-1/4}\frac{dy}{y^2}. x = λ 1/4 y : w Λ , λ = 2 1 1 + y − 4 , x 4 + λ λ = y 4 + 1 1 , x 2 d x = λ − 1/4 y 2 d y .
The coupling integral separates as ∫ 0 1 λ 3 / 4 d λ = 4 / 7 \int_0^1 \lambda^{3/4}\,\mathrm{d}\lambda=4/7 ∫ 0 1 λ 3/4 d λ = 4/7 , and
A 1 = − 48 7 π 2 ∫ 0 ∞ d y 2 y 4 + 1 ( y 2 + y 4 + 1 ) 2 . (161) A_1=-\frac{48}{7\pi^2}\int_0^\infty\frac{\mathrm{d}y}{2\sqrt{y^4+1}\left(y^2+\sqrt{y^4+1}\right)^2}. \tag*{(161)} A 1 = − 7 π 2 48 ∫ 0 ∞ 2 y 4 + 1 ( y 2 + y 4 + 1 ) 2 d y . (161)
Second and third substitutions. The second is y 2 = τ / 1 − τ 2 y^2=\tau/\sqrt{1-\tau^2} y 2 = τ / 1 − τ 2 with 0 < τ < 1 0<\tau<1 0 < τ < 1 . The third, σ = 2 τ / ( 1 + τ ) \sigma=2\tau/(1+\tau) σ = 2 τ / ( 1 + τ ) , removes the factor ( 1 + τ ) (1+\tau) ( 1 + τ ) :
y 2 = τ 1 − τ 2 : y 4 + 1 = 1 1 − τ 2 , y 2 + y 4 + 1 = 1 + τ 1 − τ 2 , d y = d τ 2 τ 1 / 2 ( 1 − τ 2 ) 5 / 4 , y^2=\frac{\tau}{\sqrt{1-\tau^2}}:\quad\sqrt{y^4+1}=\frac{1}{\sqrt{1-\tau^2}},\quad y^2+\sqrt{y^4+1}=\frac{1+\tau}{\sqrt{1-\tau^2}},\quad\mathrm{d}y=\frac{\mathrm{d}\tau}{2\tau^{1/2}(1-\tau^2)^{5/4}}, y 2 = 1 − τ 2 τ : y 4 + 1 = 1 − τ 2 1 , y 2 + y 4 + 1 = 1 − τ 2 1 + τ , d y = 2 τ 1/2 ( 1 − τ 2 ) 5/4 d τ ,
σ = 2 τ 1 + τ : 1 + τ = 2 2 − σ , 1 − τ = 2 ( 1 − σ ) 2 − σ , d τ = 2 d σ ( 2 − σ ) 2 , \sigma=\frac{2\tau}{1+\tau}:\quad1+\tau=\frac{2}{2-\sigma},\quad1-\tau=\frac{2(1-\sigma)}{2-\sigma},\quad\mathrm{d}\tau=\frac{2\,\mathrm{d}\sigma}{(2-\sigma)^2}, σ = 1 + τ 2 τ : 1 + τ = 2 − σ 2 , 1 − τ = 2 − σ 2 ( 1 − σ ) , d τ = ( 2 − σ ) 2 2 d σ ,
hence
∫ 0 ∞ d y 2 y 4 + 1 ( y 2 + y 4 + 1 ) 2 = 1 4 ∫ 0 1 ( 1 − τ ) 1 / 4 d τ τ 1 / 2 ( 1 + τ ) 7 / 4 = 1 4 2 ∫ 0 1 σ − 1 / 2 ( 1 − σ ) 1 / 4 d σ = B ( 1 / 2 , 5 / 4 ) 4 2 , \int_0^\infty\frac{\mathrm{d}y}{2\sqrt{y^4+1}\left(y^2+\sqrt{y^4+1}\right)^2}
=\frac{1}{4}\int_0^1\frac{(1-\tau)^{1/4}\,\mathrm{d}\tau}{\tau^{1/2}(1+\tau)^{7/4}}
=\frac{1}{4\sqrt{2}}\int_0^1\sigma^{-1/2}(1-\sigma)^{1/4}\,\mathrm{d}\sigma
=\frac{B(1/2,5/4)}{4\sqrt{2}}, ∫ 0 ∞ 2 y 4 + 1 ( y 2 + y 4 + 1 ) 2 d y = 4 1 ∫ 0 1 τ 1/2 ( 1 + τ ) 7/4 ( 1 − τ ) 1/4 d τ = 4 2 1 ∫ 0 1 σ − 1/2 ( 1 − σ ) 1/4 d σ = 4 2 B ( 1/2 , 5/4 ) ,
where B B B is the Euler beta function. Since Γ ( 5 / 4 ) = 1 4 Γ ( 1 / 4 ) \Gamma(5/4)=\frac{1}{4}\Gamma(1/4) Γ ( 5/4 ) = 4 1 Γ ( 1/4 ) and Γ ( 7 / 4 ) = 3 4 Γ ( 3 / 4 ) \Gamma(7/4)=\frac{3}{4}\Gamma(3/4) Γ ( 7/4 ) = 4 3 Γ ( 3/4 ) , one has B ( 1 / 2 , 5 / 4 ) = π Γ ( 1 / 4 ) / [ 3 Γ ( 3 / 4 ) ] B(1/2,5/4)=\sqrt{\pi}\,\Gamma(1/4)/[3\Gamma(3/4)] B ( 1/2 , 5/4 ) = π Γ ( 1/4 ) / [ 3Γ ( 3/4 )] , and
A 1 = − 12 7 2 π 2 B ( 1 2 , 5 4 ) = − 4 7 2 π 3 / 2 Γ ( 1 / 4 ) Γ ( 3 / 4 ) = − 0.2146937609 … (162) A_1=-\frac{12}{7\sqrt{2}\pi^2}B\left(\frac{1}{2},\frac{5}{4}\right)=-\frac{4}{7\sqrt{2}\pi^{3/2}}\frac{\Gamma(1/4)}{\Gamma(3/4)}=-0.2146937609\ldots \tag*{(162)} A 1 = − 7 2 π 2 12 B ( 2 1 , 4 5 ) = − 7 2 π 3/2 4 Γ ( 3/4 ) Γ ( 1/4 ) = − 0.2146937609 … (162)
Static line. In a same-ball region, where a = b = q 2 / 2 a=b=q^2/2 a = b = q 2 /2 , the static frequency integral W ( a , b ) \mathcal{W}(a,b) W ( a , b ) of main-text Eq. (18) and the Lorentzian one of main-text Eq. (46) for one pair of excitation energies, with J J J from (S64), are
W ( a , b ) = 0 , ( Λ q ) 2 J ( Λ q ; a , b ) = p l o c ( ω Λ ) π q 2 > 0. (163) \mathcal{W}(a,b)=0,\qquad(\Lambda q)^2J(\Lambda q;a,b)=\frac{p_{\mathrm{loc}}(\omega_\Lambda)}{\pi q^2}>0. \tag*{(163)} W ( a , b ) = 0 , ( Λ q ) 2 J ( Λ q ; a , b ) = π q 2 p loc ( ω Λ ) > 0. (163)
The static kernel thus has no q − 2 q^{-2} q − 2 term, and the dynamical kernel has the q − 2 q^{-2} q − 2 term (q ∗ − 1 ∝ r s − 1 / 4 q_*^{-1}\propto r_s^{-1/4} q ∗ − 1 ∝ r s − 1/4 ). For the static Lindhard line v / [ 1 + c ( q , 0 ) ] v/[1+c(q,0)] v / [ 1 + c ( q , 0 )] , Eq. (155) at u = 0 u=0 u = 0 and main-text Eq. (20) give, for q ≫ 1 q\gg1 q ≫ 1 and t = q / q ∗ t=q/q_* t = q / q ∗ ,
c ( q , 0 ) = q T F 2 q 2 L ( q , 0 ) ≃ 4 q T F 2 3 q 4 = t − 4 , v 1 + c ( q , 0 ) ≃ v t 4 1 + t 4 . c(q,0)=\frac{q_{\mathrm{TF}}^2}{q^2}L(q,0)\simeq\frac{4q_{\mathrm{TF}}^2}{3q^4}=t^{-4},\qquad\frac{v}{1+c(q,0)}\simeq v\frac{t^4}{1+t^4}. c ( q , 0 ) = q 2 q TF 2 L ( q , 0 ) ≃ 3 q 4 4 q TF 2 = t − 4 , 1 + c ( q , 0 ) v ≃ v 1 + t 4 t 4 .
The region q ≤ 1 q\leq1 q ≤ 1 , where the line is bounded by 2 q 2 v / q T F 2 2q^2v/q_{\mathrm{TF}}^2 2 q 2 v / q TF 2 , and the relative corrections O ( q − 2 ) O(q^{-2}) O ( q − 2 ) to c ( q , 0 ) c(q,0) c ( q , 0 ) contribute O ( q ∗ − 4 ) O(q_*^{-4}) O ( q ∗ − 4 ) . With e ∞ = O ( q ) e_\infty=O(q) e ∞ = O ( q ) for q ≤ 1 q\leq1 q ≤ 1 (§ 21.1 ) and e ∞ = O ( q − 4 ) e_\infty=O(q^{-4}) e ∞ = O ( q − 4 ) for q ≥ 1 q\geq1 q ≥ 1 (Sec. S17.1), the kite energy of the leading-order line is
∫ 0 ∞ e ∞ ( q ) t 4 1 + t 4 d q : q < q ∗ : O ( q ∗ − 4 ∫ 0 q ∗ q 4 e ∞ d q ) = O ( q ∗ − 3 ) , q > q ∗ : O ( q ∗ − 3 ) . \int_0^\infty e_\infty(q)\frac{t^4}{1+t^4}\,\mathrm{d}q:\qquad q<q_*:\ O\left(q_*^{-4}\int_0^{q_*}q^4e_\infty\,\mathrm{d}q\right)=O(q_*^{-3}),\qquad q>q_*:\ O(q_*^{-3}). ∫ 0 ∞ e ∞ ( q ) 1 + t 4 t 4 d q : q < q ∗ : O ( q ∗ − 4 ∫ 0 q ∗ q 4 e ∞ d q ) = O ( q ∗ − 3 ) , q > q ∗ : O ( q ∗ − 3 ) .
The static line therefore gives at most O ( q ∗ − 3 ) ∝ r s − 3 / 4 O(q_*^{-3})\propto r_s^{-3/4} O ( q ∗ − 3 ) ∝ r s − 3/4 , smaller than the q ∗ − 1 ∝ r s − 1 / 4 q_*^{-1}\propto r_s^{-1/4} q ∗ − 1 ∝ r s − 1/4 of the dynamically screened line.
Self-energy pair and cancellation of the leading low-density term
This subsection derives main-text Eq. (60) and the large-momentum self-energy kernel stated in main-text § 3.4 . For e S E = 2 ∫ 0 1 K Π S E ( 1 ) [ W λ ] d λ e_{\mathrm{SE}}=2\int_{0}^{1}\mathcal{K}_{\Pi_{\mathrm{SE}}^{(1)}}[W_{\lambda}]\,\mathrm{d}\lambda e SE = 2 ∫ 0 1 K Π SE ( 1 ) [ W λ ] d λ , inserting (S96) multiplied by λ v \lambda v λ v and using K Π S E ( 1 ) [ v ] = 0 \mathcal{K}_{\Pi_{\mathrm{SE}}^{(1)}}[v]=0 K Π SE ( 1 ) [ v ] = 0 [Eq. (146) , § 26 ] gives
e S E = − 2 ∫ 0 1 λ d λ ∫ 0 ∞ θ λ ( q ) e Σ , Λ λ ( q ) d q + O ( q ∗ − 2 ) , (164) e_{\mathrm{SE}}=-2\int_{0}^{1}\lambda\,\mathrm{d}\lambda\int_{0}^{\infty}\theta_{\lambda}(q)e_{\Sigma,\Lambda_{\lambda}}(q)\,\mathrm{d}q+O(q_{*}^{-2}), \tag*{(164)} e SE = − 2 ∫ 0 1 λ d λ ∫ 0 ∞ θ λ ( q ) e Σ , Λ λ ( q ) d q + O ( q ∗ − 2 ) , (164)
where e Σ , Λ ( q ) e_{\Sigma,\Lambda}(q) e Σ , Λ ( q ) is the radial kernel of the self-energy pair, defined by K Π S E ( 1 ) [ b Λ ] = ∫ 0 ∞ e Σ , Λ ( q ) d q \mathcal{K}_{\Pi_{\mathrm{SE}}^{(1)}}[b_{\Lambda}]=\int_{0}^{\infty}e_{\Sigma,\Lambda}(q)\,\mathrm{d}q K Π SE ( 1 ) [ b Λ ] = ∫ 0 ∞ e Σ , Λ ( q ) d q as in main-text Eq. (16) . Before the q q q integration, Eq. (144) gives
e Σ , Λ ( q ) = − 3 Λ π 2 ∫ max ( 0 , 1 − q ) 1 r d r ∫ max ( 1 , q − r ) q + r s d s f L ( r ) − f L ( s ) [ Λ q + ( s 2 − r 2 ) / 2 ] 2 . e_{\Sigma,\Lambda}(q)=-\frac{3\Lambda}{\pi^{2}}\int_{\max(0,1-q)}^{1}r\,\mathrm{d}r\int_{\max(1,q-r)}^{q+r}s\,\mathrm{d}s\,\frac{f_{\mathrm{L}}(r)-f_{\mathrm{L}}(s)}{[\Lambda q+(s^{2}-r^{2})/2]^{2}}. e Σ , Λ ( q ) = − π 2 3Λ ∫ m a x ( 0 , 1 − q ) 1 r d r ∫ m a x ( 1 , q − r ) q + r s d s [ Λ q + ( s 2 − r 2 ) /2 ] 2 f L ( r ) − f L ( s ) .
Large momentum. For q > 2 q>2 q > 2 the lower limit of s s s is q − r q-r q − r , and
f L ( s ) = O ( q − 2 ) , s 2 − r 2 2 = q 2 2 + q ( s − q ) + ( s − q ) 2 − r 2 2 , ∫ q − r q + r s d s = 2 q r . f_{\mathrm{L}}(s)=O(q^{-2}),\qquad\frac{s^{2}-r^{2}}{2}=\frac{q^{2}}{2}+q(s-q)+\frac{(s-q)^{2}-r^{2}}{2},\qquad\int_{q-r}^{q+r}s\,\mathrm{d}s=2qr. f L ( s ) = O ( q − 2 ) , 2 s 2 − r 2 = 2 q 2 + q ( s − q ) + 2 ( s − q ) 2 − r 2 , ∫ q − r q + r s d s = 2 q r .
The O ( q ) O(q) O ( q ) term q ( s − q ) q(s-q) q ( s − q ) of the denominator is odd in s − q s-q s − q on the symmetric interval. At first order it therefore contributes only through the odd part s − q s-q s − q of the measure s d s = [ q + ( s − q ) ] d s s\,\mathrm{d}s=[q+(s-q)]\,\mathrm{d}s s d s = [ q + ( s − q )] d s , at relative order q − 2 q^{-2} q − 2 . With Λ = q w Λ \Lambda=qw_{\Lambda} Λ = q w Λ , Λ q + q 2 / 2 = q 2 ( w Λ + 1 / 2 ) \Lambda q+q^{2}/2=q^{2}(w_{\Lambda}+1/2) Λ q + q 2 /2 = q 2 ( w Λ + 1/2 ) and (S91), the leading term is
− 3 q w Λ π 2 ∫ 0 1 r f L ( r ) 2 q r d r q 4 ( w Λ + 1 / 2 ) 2 = − 6 w Λ π 2 ⋅ 1 4 ⋅ 4 q 2 ( 1 + 2 w Λ ) 2 = − 3 p l o c ( w Λ ) π 3 q 2 , -\frac{3qw_{\Lambda}}{\pi^{2}}\int_{0}^{1}\frac{r f_{\mathrm{L}}(r)\,2qr\,\mathrm{d}r}{q^{4}(w_{\Lambda}+1/2)^{2}}=-\frac{6w_{\Lambda}}{\pi^{2}}\cdot\frac{1}{4}\cdot\frac{4}{q^{2}(1+2w_{\Lambda})^{2}}=-\frac{3p_{\mathrm{loc}}(w_{\Lambda})}{\pi^{3}q^{2}}, − π 2 3 q w Λ ∫ 0 1 q 4 ( w Λ + 1/2 ) 2 r f L ( r ) 2 q r d r = − π 2 6 w Λ ⋅ 4 1 ⋅ q 2 ( 1 + 2 w Λ ) 2 4 = − π 3 q 2 3 p loc ( w Λ ) ,
and therefore (checked numerically in § 30 )
e Σ , Λ λ ( q ) = − 3 π 3 q 2 p l o c ( w Λ , λ ) + O ( w Λ , λ q 4 ( 1 + w Λ , λ ) 2 ) , w Λ , λ = Λ λ / q ≥ 1 2 . (165) e_{\Sigma,\Lambda_{\lambda}}(q)=-\frac{3}{\pi^{3}q^{2}}p_{\mathrm{loc}}(w_{\Lambda,\lambda})+O\left(\frac{w_{\Lambda,\lambda}}{q^{4}(1+w_{\Lambda,\lambda})^{2}}\right),\qquad w_{\Lambda,\lambda}=\Lambda_{\lambda}/q\geq\frac{1}{2}. \tag*{(165)} e Σ , Λ λ ( q ) = − π 3 q 2 3 p loc ( w Λ , λ ) + O ( q 4 ( 1 + w Λ , λ ) 2 w Λ , λ ) , w Λ , λ = Λ λ / q ≥ 2 1 . (165)
Low momentum. Bound Ω / ( Ω + δ ) 2 ≤ 1 / Ω \Omega/(\Omega+\delta)^{2}\leq1/\Omega Ω/ ( Ω + δ ) 2 ≤ 1/Ω in Eq. (144) , with Ω λ = Λ λ q = ω λ \Omega_{\lambda}=\Lambda_{\lambda}q=\omega_{\lambda} Ω λ = Λ λ q = ω λ the pole frequency of (S95). With 0 ≤ f L ( r ) − f L ( s ) ≤ f L ( r ) 0\leq f_{\mathrm{L}}(r)-f_{\mathrm{L}}(s)\leq f_{\mathrm{L}}(r) 0 ≤ f L ( r ) − f L ( s ) ≤ f L ( r ) ,
M Σ ( q ) = 2 π q ∫ max ( 0 , 1 − q ) 1 r d r ∫ max ( 1 , q − r ) q + r s [ f L ( r ) − f L ( s ) ] d s ≤ 4 π ∫ 0 1 r 2 f L ( r ) d r = π , M_{\Sigma}(q)=\frac{2\pi}{q}\int_{\max(0,1-q)}^{1}r\,\mathrm{d}r\int_{\max(1,q-r)}^{q+r}s\,[f_{\mathrm{L}}(r)-f_{\mathrm{L}}(s)]\,\mathrm{d}s\leq4\pi\int_{0}^{1}r^{2}f_{\mathrm{L}}(r)\,\mathrm{d}r=\pi, M Σ ( q ) = q 2 π ∫ m a x ( 0 , 1 − q ) 1 r d r ∫ m a x ( 1 , q − r ) q + r s [ f L ( r ) − f L ( s )] d s ≤ 4 π ∫ 0 1 r 2 f L ( r ) d r = π ,
and
∣ e Σ , Λ λ ( q ) ∣ ≤ 3 M Σ ( q ) 2 π 3 Ω λ , Ω λ ≥ 1 2 λ q ∗ 2 . (166) \left|e_{\Sigma,\Lambda_{\lambda}}(q)\right|\leq\frac{3M_{\Sigma}(q)}{2\pi^{3}\Omega_{\lambda}},\qquad\Omega_{\lambda}\geq\frac{1}{2}\sqrt{\lambda}\,q_{*}^{2}. \tag*{(166)} ∣ e Σ , Λ λ ( q ) ∣ ≤ 2 π 3 Ω λ 3 M Σ ( q ) , Ω λ ≥ 2 1 λ q ∗ 2 . (166)
and with q 1 = q ∗ 1 / 2 q_{1}=q_{*}^{1/2} q 1 = q ∗ 1/2 the contributions are
q < q 1 : O ( q 1 λ q ∗ 2 ) , error of the large- q kernel , q > q 1 : O ( q 1 − 3 ) , q<q_{1}:\ O\left(\frac{q_{1}}{\sqrt{\lambda}q_{*}^{2}}\right),\qquad\text{error of the large-}q\text{ kernel},\ q>q_{1}:\ O(q_{1}^{-3}), q < q 1 : O ( λ q ∗ 2 q 1 ) , error of the large- q kernel , q > q 1 : O ( q 1 − 3 ) ,
extending (S103) to q = 0 : O ( q 1 λ q ∗ 2 ) , \text{extending (S103) to }q=0:\ O\left(\frac{q_{1}}{\sqrt{\lambda}q_{*}^{2}}\right), extending (S103) to q = 0 : O ( λ q ∗ 2 q 1 ) ,
all O ( q ∗ − 3 / 2 ) O(q_{*}^{-3/2}) O ( q ∗ − 3/2 ) at fixed λ \lambda λ . The factor λ − 1 / 2 \lambda^{-1/2} λ − 1/2 in these bounds is integrable against 2 λ d λ 2\lambda\,\mathrm{d}\lambda 2 λ d λ .
Remainder and fixed interval. The remainder ς λ \varsigma_{\lambda} ς λ and the fixed momentum interval $0<q\leq q_{0}$ contribute O ( q ∗ − 2 ) O(q_{*}^{-2}) O ( q ∗ − 2 ) to (S102) when bounded as for the kite in § 29.2 , with the cutoff $q_{0}\gg1 chosen there. For the self-energy pair these bounds use the large-q q q estimate that follows from (S21),
∣ Π S E ( 1 ) ( q , i q 2 u ) ∣ q T F 4 = O ( q − 4 1 + u 2 ) , \frac{\left|\Pi_{\mathrm{SE}}^{(1)}(q,iq^{2}u)\right|}{q_{\mathrm{TF}}^{4}}=O\left(\frac{q^{-4}}{1+u^{2}}\right), q TF 4 Π SE ( 1 ) ( q , i q 2 u ) = O ( 1 + u 2 q − 4 ) ,
since
∣ Σ x ( p + q ) − Σ x ( p ) ∣ ≤ q T F 2 2 , ∣ i ν − δ p ∣ − 2 ≤ 1 ν 2 + ( q 2 / 2 − q ) 2 ( q > 2 ) . \left|\Sigma_x(\mathbf{p}+\mathbf{q})-\Sigma_x(\mathbf{p})\right|\leq\frac{q_{\mathrm{TF}}^{2}}{2},\qquad\left|i\nu-\delta_p\right|^{-2}\leq\frac{1}{\nu^{2}+(q^{2}/2-q)^{2}}\qquad(q>2). ∣ Σ x ( p + q ) − Σ x ( p ) ∣ ≤ 2 q TF 2 , ∣ i ν − δ p ∣ − 2 ≤ ν 2 + ( q 2 /2 − q ) 2 1 ( q > 2 ) .
They also use the lower bound on L \mathcal{L} L , derived below with Q 0 = ( 1 + q 0 / 2 ) 2 Q_0=(1+q_0/2)^2 Q 0 = ( 1 + q 0 /2 ) 2 ,
L ( q , ν ) ≥ q 2 3 ( ν 2 + Q 0 q 2 ) , W λ λ ν ≤ ν 2 + Q 0 q 2 ν 2 + Q 0 q 2 + λ q T F 2 / 3 ( 0 < q ≤ q 0 ) , (167) \mathcal{L}(q,\nu)\geq\frac{q^{2}}{3(\nu^{2}+Q_0q^{2})},\qquad\frac{W_\lambda}{\lambda\nu}\leq\frac{\nu^{2}+Q_0q^{2}}{\nu^{2}+Q_0q^{2}+\lambda q_{\mathrm{TF}}^{2}/3}\qquad(0<q\leq q_0), \tag*{(167)} L ( q , ν ) ≥ 3 ( ν 2 + Q 0 q 2 ) q 2 , λ ν W λ ≤ ν 2 + Q 0 q 2 + λ q TF 2 /3 ν 2 + Q 0 q 2 ( 0 < q ≤ q 0 ) , (167)
together with ∣ Re ( i ν − δ ) ∣ − 2 ≤ ( ν 2 + δ 2 ) − 1 \left|\operatorname{Re}(i\nu-\delta)\right|^{-2}\leq(\nu^{2}+\delta^{2})^{-1} ∣ Re ( i ν − δ ) ∣ − 2 ≤ ( ν 2 + δ 2 ) − 1 in (S21) and the bounds of § 24.3 .
*Lower bound on L \mathcal{L} L . (167) uses the free-response representation of § 27 , valid for every q > 0 q>0 q > 0 ,
L ( q , ν ) = ( 4 π ) − 1 ∫ d 3 p Δ p δ p δ p 2 + ν 2 . \mathcal{L}(q,\nu)=(4\pi)^{-1}\int\mathrm{d}^{3}p\,\frac{\Delta_p\delta_p}{\delta_p^{2}+\nu^{2}}. L ( q , ν ) = ( 4 π ) − 1 ∫ d 3 p δ p 2 + ν 2 Δ p δ p .
Its integrand is nonnegative because Δ p = sgn δ p \Delta_p=\operatorname{sgn}\delta_p Δ p = sgn δ p on the transition regions. There, for q ≤ q 0 q\leq q_0 q ≤ q 0 ,
∣ δ p ∣ ≤ q + q 2 / 2 ≤ q ( 1 + q 0 / 2 ) , δ p 2 + ν 2 ≤ ν 2 + Q 0 q 2 . |\delta_p|\leq q+q^{2}/2\leq q(1+q_0/2),\qquad\delta_p^{2}+\nu^{2}\leq\nu^{2}+Q_0q^{2}. ∣ δ p ∣ ≤ q + q 2 /2 ≤ q ( 1 + q 0 /2 ) , δ p 2 + ν 2 ≤ ν 2 + Q 0 q 2 .
Shifting p → p − q \mathbf{p}\to\mathbf{p}-\mathbf{q} p → p − q in the term with n p + q n_{\mathbf{p}+\mathbf{q}} n p + q gives the free f f f -sum identity, valid for every q q q ,
∫ d 3 p Δ p δ p = 1 2 ∫ d 3 p n p [ ( p + q ) 2 + ( p − q ) 2 − 2 p 2 ] = 4 π q 2 3 , \int\mathrm{d}^{3}p\,\Delta_p\delta_p=\frac{1}{2}\int\mathrm{d}^{3}p\,n_p\left[(p+q)^{2}+(p-q)^{2}-2p^{2}\right]=\frac{4\pi q^{2}}{3}, ∫ d 3 p Δ p δ p = 2 1 ∫ d 3 p n p [ ( p + q ) 2 + ( p − q ) 2 − 2 p 2 ] = 3 4 π q 2 ,
and hence the first inequality. Inserting it into W λ / ( λ ν ) = [ 1 + λ q T F 2 L / q 2 ] − 1 W_\lambda/(\lambda\nu)=[1+\lambda q_{\mathrm{TF}}^{2}\mathcal{L}/q^{2}]^{-1} W λ / ( λ ν ) = [ 1 + λ q TF 2 L / q 2 ] − 1 gives the second.
Cancellation. Inserting (S103) into (S102) gives (S98) with the large-momentum kernel (48) of opposite sign, and the substitutions of (51) then give
e S E = − A 1 q ∗ + O ( q ∗ − 3 / 2 ) , e k i t e + e S E = O ( q ∗ − 3 / 2 ) . (168) e_{\mathrm{SE}}=-\frac{A_1}{q_*}+O(q_*^{-3/2}),\qquad e_{\mathrm{kite}}+e_{\mathrm{SE}}=O(q_*^{-3/2}). \tag*{(168)} e SE = − q ∗ A 1 + O ( q ∗ − 3/2 ) , e kite + e SE = O ( q ∗ − 3/2 ) . (168)
The bounds above hold at each fixed λ \lambda λ , hence at full coupling λ = 1 \lambda=1 λ = 1 . Then
F Σ = K Π S E ( 1 ) [ W λ = 1 ] = − 7 A 1 8 q ∗ + O ( q ∗ − 3 / 2 ) , F_\Sigma=\mathcal{K}_{\Pi_{\mathrm{SE}}^{(1)}}[W_{\lambda=1}]=-\frac{7A_1}{8q_*}+O(q_*^{-3/2}), F Σ = K Π SE ( 1 ) [ W λ = 1 ] = − 8 q ∗ 7 A 1 + O ( q ∗ − 3/2 ) ,
since at λ = 1 \lambda=1 λ = 1 the factor 2 ∫ 0 1 λ 3 / 4 d λ = 8 / 7 2\int_0^1\lambda^{3/4}\mathrm{d}\lambda=8/7 2 ∫ 0 1 λ 3/4 d λ = 8/7 of A 1 A_1 A 1 is absent, as for F = 7 8 e k i t e F=\frac{7}{8}e_{\mathrm{kite}} F = 8 7 e kite in Sec. S12.4. Hence F + F Σ F+F_\Sigma F + F Σ also has no q ∗ − 1 q_*^{-1} q ∗ − 1 term. The first nonzero term after the cancellation is not determined here.
Remainder estimates for the low-density expansion
The bounds of this section prove the remainder O ( q ∗ − 3 / 2 ) O(q_*^{-3/2}) O ( q ∗ − 3/2 ) in main-text Eqs. (51) and (60) , including the limit λ → 0 \lambda\to0 λ → 0 of the coupling integral.
Fixed-frequency limit and the approach to the bare line
For every q > 0 q>0 q > 0 , with ∣ δ p ∣ , ∣ δ k ∣ ≤ q 2 / 2 + q |\delta_p|,|\delta_k|\leq q^{2}/2+q ∣ δ p ∣ , ∣ δ k ∣ ≤ q 2 /2 + q on the transition regions, the radial kernel e Λ ( q ) e_\Lambda(q) e Λ ( q ) of main-text Eq. (46) and its bare-line limit e ∞ ( q ) e_\infty(q) e ∞ ( q ) obey
e Λ ( q ) − e ∞ ( q ) = O ( Z ˉ ( q ) [ 1 Λ q + 1 + q 2 Λ 2 q 2 ] ) , e_\Lambda(q)-e_\infty(q)=O\left(\bar{Z}(q)\left[\frac{1}{\Lambda q}+\frac{1+q^{2}}{\Lambda^{2}q^{2}}\right]\right), e Λ ( q ) − e ∞ ( q ) = O ( Z ˉ ( q ) [ Λ q 1 + Λ 2 q 2 1 + q 2 ] ) ,
where
Z ˉ ( q ) = ∫ d 3 p d 3 k ∣ Δ p ( q ) Δ k ( q ) ∣ ∣ p − k ∣ 2 = O ( q 2 log ( e / q ) ) ( q → 0 ) , Z ˉ → 8 π 2 ( q → ∞ ) , (169) \bar{Z}(q)=\int\mathrm{d}^{3}p\,\mathrm{d}^{3}k\,\frac{|\Delta_p(q)\Delta_k(q)|}{|\mathbf{p}-\mathbf{k}|^{2}}=O\left(q^{2}\log(e/q)\right)\ (q\to0),\qquad\bar{Z}\to8\pi^{2}\ (q\to\infty), \tag*{(169)} Z ˉ ( q ) = ∫ d 3 p d 3 k ∣ p − k ∣ 2 ∣ Δ p ( q ) Δ k ( q ) ∣ = O ( q 2 log ( e / q ) ) ( q → 0 ) , Z ˉ → 8 π 2 ( q → ∞ ) , (169)
with an absolute constant in the O O O symbol.
Pair bound. For each pair of unscaled excitation energies a = δ p a = \delta_p a = δ p , b = δ k b = \delta_k b = δ k , the closed form (S64) and the static frequency integral W \mathcal{W} W of main-text Eq. (18) give
∣ Ω 2 J ( Ω ; a , b ) − W ( a , b ) ∣ ≤ 1 2 Ω + min ( ∣ a ∣ , ∣ b ∣ ) Ω 2 , Ω = Λ q . \left|\Omega^2 J(\Omega; a,b)-\mathcal{W}(a,b)\right| \le\frac{1}{2\Omega}+\frac{\min(|a|,|b|)}{\Omega^2}, \qquad\Omega=\Lambda q. Ω 2 J ( Ω ; a , b ) − W ( a , b ) ≤ 2Ω 1 + Ω 2 min ( ∣ a ∣ , ∣ b ∣ ) , Ω = Λ q .
Integrating this bound with the factor ∣ Δ p Δ k ∣ / ∣ p − k ∣ 2 |\Delta_p\Delta_k|/|p-k|^2 ∣ Δ p Δ k ∣/∣ p − k ∣ 2 gives the factor Z ˉ \bar{Z} Z ˉ ,
∣ e Λ ( q ) − e ∞ ( q ) ∣ ≤ 4 π A 0 Z ˉ ( q ) [ 1 2 Λ q + q 2 / 2 + q Λ 2 q 2 ] ≤ 4 π A 0 Z ˉ ( q ) [ 1 Λ q + 1 + q 2 Λ 2 q 2 ] . |e_{\Lambda}(q)-e_{\infty}(q)|\le4\pi A_0\bar{Z}(q)\left[\frac{1}{2\Lambda q}+\frac{q^2/2+q}{\Lambda^2q^2}\right]\le4\pi A_0\bar{Z}(q)\left[\frac{1}{\Lambda q}+\frac{1+q^2}{\Lambda^2q^2}\right]. ∣ e Λ ( q ) − e ∞ ( q ) ∣ ≤ 4 π A 0 Z ˉ ( q ) [ 2Λ q 1 + Λ 2 q 2 q 2 /2 + q ] ≤ 4 π A 0 Z ˉ ( q ) [ Λ q 1 + Λ 2 q 2 1 + q 2 ] .
Limits of Z ˉ \bar{Z} Z ˉ . The small-q q q form follows from the shell volume O ( q ) O(q) O ( q ) of either transition region and the shell Coulomb bound of § 27 . For q > 2 q>2 q > 2 the two same-ball regions contribute 2 ⋅ 4 π 2 2\cdot4\pi^2 2 ⋅ 4 π 2 (Sec. S17.1). Hence
Z ˉ ( q ) ≤ O ( q ) ⋅ O ( q log ( e / q ) ) ( q → 0 ) , \bar{Z}(q)\le O(q)\cdot O\left(q\log(e/q)\right)\qquad(q\to0), Z ˉ ( q ) ≤ O ( q ) ⋅ O ( q log ( e / q ) ) ( q → 0 ) ,
Z ˉ ( q ) = 2 ⋅ 4 π 2 (same ball) + O ( q − 2 ) (different balls) ( q > 2 ) . \bar{Z}(q)=2\cdot4\pi^2\ \text{(same ball)}+O(q^{-2})\ \text{(different balls)}\qquad(q>2). Z ˉ ( q ) = 2 ⋅ 4 π 2 (same ball) + O ( q − 2 ) (different balls) ( q > 2 ) .
Fixed frequency. For fixed Ω 0 > 0 \Omega_0>0 Ω 0 > 0 , the choice Λ = Ω 0 / q \Lambda=\Omega_0/q Λ = Ω 0 / q keeps the width in ν \nu ν fixed as q → 0 q\to0 q → 0 . Then Λ q = Ω 0 \Lambda q=\Omega_0 Λ q = Ω 0 , and (S107) gives
e Ω 0 / q ( q ) = e ∞ ( q ) + O Ω 0 ( q 2 log ( e / q ) ) , Ω 0 > 0 , q → 0. (170) e_{\Omega_0/q}(q)=e_{\infty}(q)+O_{\Omega_0}\left(q^2\log(e/q)\right),\qquad\Omega_0>0,\quad q\to0. \tag*{(170)} e Ω 0 / q ( q ) = e ∞ ( q ) + O Ω 0 ( q 2 log ( e / q ) ) , Ω 0 > 0 , q → 0. (170)
A spectral component of fixed frequency, such as the small-q q q plasmon with Λ ≃ ω p / q \Lambda\simeq\omega_p/q Λ ≃ ω p / q (main-text § 2.4 ), thus acts like the bare line at small q q q .
Large momentum. For q ≫ 1 q\gg1 q ≫ 1 , with the scaled frequency u = ν / q 2 u=\nu/q^2 u = ν / q 2 of Sec. S17.2, main-text Eq. (41) gives the decay bound
X ( q , q 2 u ) = O ( 1 q 4 ( 1 + u 2 ) ) , q ≫ 1 , u ≥ 0. (171) X(q,q^2u)=O\left(\frac{1}{q^4(1+u^2)}\right),\qquad q\gg1,\quad u\ge0. \tag*{(171)} X ( q , q 2 u ) = O ( q 4 ( 1 + u 2 ) 1 ) , q ≫ 1 , u ≥ 0. (171)
For q > 2 q>2 q > 2 every excitation energy on the transition regions obeys ∣ δ ∣ ≥ q 2 / 2 − q |\delta|\ge q^2/2-q ∣ δ ∣ ≥ q 2 /2 − q . The product of the two frequency factors therefore obeys
∣ 1 ( i ν − δ p ) ( i ν − δ k ) ∣ ≤ 1 ν 2 + ( q 2 / 2 − q ) 2 = 1 q 4 [ u 2 + ( 1 / 2 − 1 / q ) 2 ] , \left|\frac{1}{(i\nu-\delta_p)(i\nu-\delta_k)}\right|\le\frac{1}{\nu^2+(q^2/2-q)^2}=\frac{1}{q^4\left[u^2+(1/2-1/q)^2\right]}, ( i ν − δ p ) ( i ν − δ k ) 1 ≤ ν 2 + ( q 2 /2 − q ) 2 1 = q 4 [ u 2 + ( 1/2 − 1/ q ) 2 ] 1 ,
and the Coulomb integrals of the four regions are bounded as in Sec. S17.1.
Remainder bounds for the coupling-averaged kite
To justify (S98), we insert (S96) into (S16) and keep the error of the single-Lorentzian form as a separate term:
e k i t e = e 2 x − 2 ∫ 0 1 λ d λ ∫ 0 ∞ θ λ ( q ) e Λ λ ( q ) ( q ) d q + 2 ∫ 0 1 λ d λ ∫ 0 ∞ P λ ( q ) d q . (172) e_{\mathrm{kite}}=e_{2x}-2\int_0^1\lambda\,d\lambda\int_0^\infty\theta_{\lambda}(q)e_{\Lambda_{\lambda}(q)}(q)\,dq+2\int_0^1\lambda\,d\lambda\int_0^\infty\mathcal{P}_{\lambda}(q)\,dq. \tag*{(172)} e kite = e 2 x − 2 ∫ 0 1 λ d λ ∫ 0 ∞ θ λ ( q ) e Λ λ ( q ) ( q ) d q + 2 ∫ 0 1 λ d λ ∫ 0 ∞ P λ ( q ) d q . (172)
Here
P λ ( q ) = 4 π A 0 q 2 ∫ 0 ∞ ν ( q ) ς λ ( q , ν / q 2 ) X ( q , ν ) d ν π \mathcal{P}_{\lambda}(q)=4\pi A_0q^2\int_0^\infty\nu(q)\varsigma_{\lambda}(q,\nu/q^2)X(q,\nu)\frac{d\nu}{\pi} P λ ( q ) = 4 π A 0 q 2 ∫ 0 ∞ ν ( q ) ς λ ( q , ν / q 2 ) X ( q , ν ) π d ν
is the radial energy contribution of the error of the single-Lorentzian form, without the coupling factor λ \lambda λ . Three errors are bounded in turn: P λ \mathcal{P}_{\lambda} P λ , the bare remainder, and the replacement of e Λ λ − e ∞ e_{\Lambda_{\lambda}}-e_{\infty} e Λ λ − e ∞ by its large-momentum form.
Single-Lorentzian error, q ≫ 1 q \gg1 q ≫ 1 . We use u u u and c ^ \hat{c} c ^ of Sec. S17.2 and the local variable ρ = λ ( q ∗ / q ) 4 \rho= \lambda(q_*/q)^4 ρ = λ ( q ∗ / q ) 4 (not the function ρ ( c ) \rho(c) ρ ( c ) of Eq. (90) ). With ( 1 + 4 u 2 ) / ( 1 + u 2 ) ≤ 4 (1+4u^2)/(1+u^2) \le4 ( 1 + 4 u 2 ) / ( 1 + u 2 ) ≤ 4 ,
λ c ^ = ρ 1 + 4 u 2 , \lambda\hat{c} = \frac{\rho}{1+4u^2}, λ c ^ = 1 + 4 u 2 ρ ,
∫ 0 ∞ λ c ^ d u ( 1 + λ c ^ ) 2 ( 1 + u 2 ) ≤ ∫ 0 ∞ 4 ρ d u ( 1 + ρ + 4 u 2 ) 2 = π 2 ρ ( 1 + ρ ) 3 / 2 = π 2 θ λ 1 − θ λ . \int_0^\infty\frac{\lambda\hat{c}\,\mathrm{d}u}{(1+\lambda\hat{c})^2(1+u^2)} \le\int_0^\infty\frac{4\rho\,\mathrm{d}u}{(1+\rho+4u^2)^2} = \frac{\pi}{2}\frac{\rho}{(1+\rho)^{3/2}} = \frac{\pi}{2}\theta_\lambda\sqrt{1-\theta_\lambda}. ∫ 0 ∞ ( 1 + λ c ^ ) 2 ( 1 + u 2 ) λ c ^ d u ≤ ∫ 0 ∞ ( 1 + ρ + 4 u 2 ) 2 4 ρ d u = 2 π ( 1 + ρ ) 3/2 ρ = 2 π θ λ 1 − θ λ .
This estimate, (S109) and the bound on ς λ \varsigma_\lambda ς λ in (S96) give
P λ ( q ) = O ( q − 4 θ λ 1 − θ λ ) , q ≫ 1. (173) \mathcal{P}_\lambda(q)=O\left(q^{-4}\theta_\lambda\sqrt{1-\theta_\lambda}\right), \qquad q\gg1. \tag*{(173)} P λ ( q ) = O ( q − 4 θ λ 1 − θ λ ) , q ≫ 1. (173)
For a fixed cutoff q 0 ≫ 1 q_0\gg1 q 0 ≫ 1 ,
θ λ 1 − θ λ ≤ ρ − 1 / 2 = q 2 λ q ∗ 2 , ∫ q ∗ ∞ d q q 4 q 2 λ q ∗ 2 = 1 λ q ∗ q 0 2 , \theta_\lambda\sqrt{1-\theta_\lambda}\le\rho^{-1/2}=\frac{q^2}{\sqrt{\lambda}q_*^2}, \qquad\int_{q_*}^\infty\frac{\mathrm{d}q}{q^4}\frac{q^2}{\sqrt{\lambda}q_*^2}=\frac{1}{\sqrt{\lambda}q_*q_0^2}, θ λ 1 − θ λ ≤ ρ − 1/2 = λ q ∗ 2 q 2 , ∫ q ∗ ∞ q 4 d q λ q ∗ 2 q 2 = λ q ∗ q 0 2 1 ,
and integration over λ \lambda λ with the weight 2 λ d λ 2\lambda\,\mathrm{d}\lambda 2 λ d λ gives O ( q ∗ − 2 ) O(q_*^{-2}) O ( q ∗ − 2 ) .
Single-Lorentzian error, q ≤ q 0 q\le q_0 q ≤ q 0 . We use the positive lower bound (S105) on L \mathcal{L} L , the absolute shell bounds of § 27 , and the large-frequency bound from main-text Eq. (41) and (S107). With Ξ = λ q T F 2 / 3 \Xi=\lambda q_{\mathrm{TF}}^2/3 Ξ = λ q TF 2 /3 and the ratio W λ / ( λ ν ) W_\lambda/(\lambda\nu) W λ / ( λ ν ) of (S105),
∣ X ( q , ν ) ∣ ≤ Z ˉ ( q ) ν 2 , W λ λ ν ≤ ν 2 + Q 0 q 2 Ξ , W λ λ ν ≤ 1. |X(q,\nu)|\le\frac{\bar{Z}(q)}{\nu^2}, \qquad\frac{W_\lambda}{\lambda\nu}\le\frac{\nu^2+Q_0q^2}{\Xi}, \qquad\frac{W_\lambda}{\lambda\nu}\le1. ∣ X ( q , ν ) ∣ ≤ ν 2 Z ˉ ( q ) , λ ν W λ ≤ Ξ ν 2 + Q 0 q 2 , λ ν W λ ≤ 1.
The part Q 0 q 2 / Ξ Q_0q^2/\Xi Q 0 q 2 /Ξ multiplies the frequency integral of ∣ X ∣ |X| ∣ X ∣ , bounded by (S57). The part ν 2 / Ξ \nu^2/\Xi ν 2 /Ξ for ν < Ξ 1 / 2 \nu<\Xi^{1/2} ν < Ξ 1/2 and the ratio bound 1 1 1 for ν > Ξ 1 / 2 \nu>\Xi^{1/2} ν > Ξ 1/2 use the large-frequency bound:
Q 0 q 2 Ξ ∫ 0 ∞ ∣ X ∣ d ν = O ( Ξ − 1 ) , ∫ 0 Ξ 1 / 2 ν 2 Ξ Z ˉ ν 2 d ν = Z ˉ Ξ 1 / 2 , ∫ Ξ 1 / 2 ∞ Z ˉ ν 2 d ν = Z ˉ Ξ 1 / 2 . \frac{Q_0q^2}{\Xi}\int_0^\infty|X|\,\mathrm{d}\nu=O(\Xi^{-1}), \qquad\int_0^{\Xi^{1/2}}\frac{\nu^2}{\Xi}\frac{\bar{Z}}{\nu^2}\,\mathrm{d}\nu=\frac{\bar{Z}}{\Xi^{1/2}}, \qquad\int_{\Xi^{1/2}}^\infty\frac{\bar{Z}}{\nu^2}\,\mathrm{d}\nu=\frac{\bar{Z}}{\Xi^{1/2}}. Ξ Q 0 q 2 ∫ 0 ∞ ∣ X ∣ d ν = O ( Ξ − 1 ) , ∫ 0 Ξ 1/2 Ξ ν 2 ν 2 Z ˉ d ν = Ξ 1/2 Z ˉ , ∫ Ξ 1/2 ∞ ν 2 Z ˉ d ν = Ξ 1/2 Z ˉ .
Each of the last two parts is therefore O ( Ξ − 1 / 2 ) O(\Xi^{-1/2}) O ( Ξ − 1/2 ) . Integrated over 0 < q ≤ q 0 0<q\le q_0 0 < q ≤ q 0 , these estimates bound the contribution of the screened line, with its factor λ \lambda λ removed, by O ( ( λ q T F 2 ) − 1 / 2 + ( λ q T F 2 ) − 1 ) O((\lambda q_{\mathrm{TF}}^2)^{-1/2}+(\lambda q_{\mathrm{TF}}^2)^{-1}) O (( λ q TF 2 ) − 1/2 + ( λ q TF 2 ) − 1 ) , and the single-Lorentzian form obeys the same bound. Their difference therefore contributes O ( q ∗ − 2 ) O(q_*^{-2}) O ( q ∗ − 2 ) after integration with 2 λ d λ 2\lambda\,\mathrm{d}\lambda 2 λ d λ .
Bare remainder. Write e Λ λ = e ∞ + ( e Λ λ − e ∞ ) e_{\Lambda\lambda}=e_\infty+(e_{\Lambda\lambda}-e_\infty) e Λ λ = e ∞ + ( e Λ λ − e ∞ ) and combine the e ∞ e_\infty e ∞ term with e 2 x e_{2x} e 2 x . The remaining bare contribution is
2 ∫ 0 1 λ d λ ∫ 0 ∞ ( 1 − θ λ ) e ∞ d q = O ( q ∗ − 3 ) , e ∞ = O ( q − 4 ) . (174) 2\int_0^1\lambda\,\mathrm{d}\lambda\int_0^\infty(1-\theta_\lambda)e_\infty\,\mathrm{d}q=O(q_*^{-3}), \qquad e_\infty=O(q^{-4}). \tag*{(174)} 2 ∫ 0 1 λ d λ ∫ 0 ∞ ( 1 − θ λ ) e ∞ d q = O ( q ∗ − 3 ) , e ∞ = O ( q − 4 ) . (174)
Here e ∞ = O ( q − 4 ) e_\infty=O(q^{-4}) e ∞ = O ( q − 4 ) by Sec. S17.1 and ∫ 0 ∞ ∣ e ∞ ∣ d q < ∞ \int_0^\infty|e_\infty|\,\mathrm{d}q<\infty ∫ 0 ∞ ∣ e ∞ ∣ d q < ∞ by (S57). Splitting the q q q integral at λ 1 / 4 q ∗ \lambda^{1/4}q_* λ 1/4 q ∗ ,
1 − θ λ ≤ min { 1 , q 4 / ( λ q ∗ 4 ) } , 1-\theta_\lambda\le\min\{1,q^4/(\lambda q_*^4)\}, 1 − θ λ ≤ min { 1 , q 4 / ( λ q ∗ 4 )} ,
∫ 0 ∞ ( 1 − θ λ ) ∣ e ∞ ∣ d q ≤ ∫ 0 λ 1 / 4 q ∗ q 4 ∣ e ∞ ∣ λ q ∗ 4 d q + ∫ λ 1 / 4 q ∗ ∞ ∣ e ∞ ∣ d q = O ( λ − 3 / 4 q ∗ − 3 ) . \int_0^\infty(1-\theta_\lambda)|e_\infty|\,\mathrm{d}q\le\int_0^{\lambda^{1/4}q_*}\frac{q^4|e_\infty|}{\lambda q_*^4}\,\mathrm{d}q+\int_{\lambda^{1/4}q_*}^\infty|e_\infty|\,\mathrm{d}q=O(\lambda^{-3/4}q_*^{-3}). ∫ 0 ∞ ( 1 − θ λ ) ∣ e ∞ ∣ d q ≤ ∫ 0 λ 1/4 q ∗ λ q ∗ 4 q 4 ∣ e ∞ ∣ d q + ∫ λ 1/4 q ∗ ∞ ∣ e ∞ ∣ d q = O ( λ − 3/4 q ∗ − 3 ) .
Large-momentum kernel. Finally, replace the subtracted kernel e Λ λ − e ∞ e_{\Lambda\lambda}-e_\infty e Λ λ − e ∞ by its large-momentum limit 3 p l o c ( w Λ , λ ) / ( π 3 q 2 ) 3p_{\mathrm{loc}}(w_{\Lambda,\lambda})/(\pi^3q^2) 3 p loc ( w Λ , λ ) / ( π 3 q 2 ) of main-text Eq. (48) , and bound the resulting error separately below and above q 1 = q ∗ 4 q_1=q_*^4 q 1 = q ∗ 4 . For q < q 1 q<q_1 q < q 1 , take (S107) with Λ q = ω λ \Lambda q=\omega_\lambda Λ q = ω λ , the pole frequency of Eq. (157) , and p l o c ( w ) ≤ π / ( 2 w ) p_{\mathrm{loc}}(w)\le\pi/(2w) p loc ( w ) ≤ π / ( 2 w ) . Then
e Λ λ − e ∞ = O ( ω λ − 1 + ( 1 + q 2 ) ω λ − 2 ) , 3 p l o c ( w Λ , λ ) π 3 q 2 ≤ 3 2 π 2 ω λ , e_{\Lambda\lambda}-e_\infty=O\left(\omega_\lambda^{-1}+(1+q^2)\omega_\lambda^{-2}\right), \qquad\frac{3p_{\mathrm{loc}}(w_{\Lambda,\lambda})}{\pi^3q^2}\le\frac{3}{2\pi^2\omega_\lambda}, e Λ λ − e ∞ = O ( ω λ − 1 + ( 1 + q 2 ) ω λ − 2 ) , π 3 q 2 3 p loc ( w Λ , λ ) ≤ 2 π 2 ω λ 3 ,
and integration over this interval gives
O ( λ − 1 / 2 q 1 / q ∗ 2 + λ − 1 q 1 3 / q ∗ 4 ) , q 1 3 / q ∗ 4 = q ∗ − 5 / 2 < q 1 / q ∗ 2 . O\left(\lambda^{-1/2}q_1/q_*^2+\lambda^{-1}q_1^3/q_*^4\right),\qquad q_1^3/q_*^4=q_*^{-5/2}<q_1/q_*^2. O ( λ − 1/2 q 1 / q ∗ 2 + λ − 1 q 1 3 / q ∗ 4 ) , q 1 3 / q ∗ 4 = q ∗ − 5/2 < q 1 / q ∗ 2 .
For q > q 1 q>q_1 q > q 1 , integrate the remainder O ( q − 3 / ( 1 + w Λ , λ ) + q − 4 ) O(q^{-3}/(1+w_{\Lambda,\lambda})+q^{-4}) O ( q − 3 / ( 1 + w Λ , λ ) + q − 4 ) of main-text Eq. (48) with w Λ , λ = Λ λ / q w_{\Lambda,\lambda}=\Lambda_\lambda/q w Λ , λ = Λ λ / q of Eq. (165) . Since w Λ , λ ≃ 1 2 λ ( q ∗ / q ) 2 w_{\Lambda,\lambda}\simeq\frac{1}{2}\sqrt{\lambda}(q_*/q)^2 w Λ , λ ≃ 2 1 λ ( q ∗ / q ) 2 for q < λ 1 / 4 q ∗ q<\lambda^{1/4}q_* q < λ 1/4 q ∗ ,
∫ q 1 ∞ d q q 3 ( 1 + w Λ , λ ) = O ( λ − 1 / 2 q ∗ − 2 log q ∗ ) , ∫ q 1 ∞ d q q 4 = O ( q 1 − 3 ) . \int_{q_1}^{\infty}\frac{\mathrm{d}q}{q^3(1+w_{\Lambda,\lambda})}=O\left(\lambda^{-1/2}q_*^{-2}\log q_*\right),\qquad\int_{q_1}^{\infty}\frac{\mathrm{d}q}{q^4}=O(q_1^{-3}). ∫ q 1 ∞ q 3 ( 1 + w Λ , λ ) d q = O ( λ − 1/2 q ∗ − 2 log q ∗ ) , ∫ q 1 ∞ q 4 d q = O ( q 1 − 3 ) .
With the factors λ − 1 / 2 \lambda^{-1/2} λ − 1/2 and λ − 1 \lambda^{-1} λ − 1 suppressed, the bounds and integrated errors are as follows.
Table S113.
Coupling integral. The suppressed factors and the factor λ − 3 / 4 \lambda^{-3/4} λ − 3/4 below Eq. (174) are integrable against 2 λ d λ 2\lambda\,\mathrm{d}\lambda 2 λ d λ , and the estimates therefore survive the limit λ → 0 \lambda\to0 λ → 0 of the coupling integral. With q = q ∗ x q=q_*x q = q ∗ x and λ > 0 \lambda>0 λ > 0 ,
w Λ , λ ≥ λ 2 x 2 , p l o c ( w Λ , λ ) ≤ π x 2 λ , 1 q ∗ ∫ 0 x 1 d x x 2 p l o c ( w Λ , λ ) ≤ π x 1 λ q ∗ , x 1 = q ∗ − 1 / 2 . w_{\Lambda,\lambda}\ge\frac{\sqrt{\lambda}}{2x^2},\qquad p_{\mathrm{loc}}(w_{\Lambda,\lambda})\le\frac{\pi x^2}{\sqrt{\lambda}},\qquad\frac{1}{q_*}\int_0^{x_1}\frac{\mathrm{d}x}{x^2}p_{\mathrm{loc}}(w_{\Lambda,\lambda})\le\frac{\pi x_1}{\sqrt{\lambda}q_*},\qquad x_1=q_*^{-1/2}. w Λ , λ ≥ 2 x 2 λ , p loc ( w Λ , λ ) ≤ λ π x 2 , q ∗ 1 ∫ 0 x 1 x 2 d x p loc ( w Λ , λ ) ≤ λ q ∗ π x 1 , x 1 = q ∗ − 1/2 .
Integration with the weight 2 λ d λ 2\lambda\,\mathrm{d}\lambda 2 λ d λ shows that extending the x x x integral of main-text Eq. (51) from x 1 x_1 x 1 to zero adds O ( q ∗ − 3 / 2 ) O(q_*^{-3/2}) O ( q ∗ − 3/2 ) . The error terms of e k i t e e_{\mathrm{kite}} e kite are
single-Lorentzian error P λ \mathcal{P}_\lambda P λ , q ≥ q 0 q\ge q_0 q ≥ q 0 and q ≤ q 0 q\le q_0 q ≤ q 0 : O ( q ∗ − 2 ) O(q_*^{-2}) O ( q ∗ − 2 ) ,
bare remainder: O ( q ∗ − 3 ) O(q_*^{-3}) O ( q ∗ − 3 ) ,
large-momentum kernel, q < q 1 q<q_1 q < q 1 and q > q 1 q>q_1 q > q 1 : O ( q ∗ − 3 / 2 ) O(q_*^{-3/2}) O ( q ∗ − 3/2 ) ,
extension to 0 < x < x 1 0<x<x_1 0 < x < x 1 : O ( q ∗ − 3 / 2 ) O(q_*^{-3/2}) O ( q ∗ − 3/2 ) .
This completes the proof of the remainders in (S98). The exponent of the remainder bounds the omitted terms but does not identify the next nonzero power in the density expansion.
Numerical checks
The checks are independent calculations, mostly quadratures, in double precision unless a multiprecision working precision is stated. Each entry gives the largest deviation found.
Frequency integrals and normalization. The closed forms of J ( ω ; a , b ) J(\omega;a,b) J ( ω ; a , b ) of Sec. S13.2 (equal-sign and opposite-sign arguments, coincident limit) agree with direct positive-frequency quadrature to 2.2 × 10 − 16 2.2\times10^{-16} 2.2 × 1 0 − 16 . The closed form of the Lindhard function L \mathcal{L} L of [1 ] agrees with direct three-dimensional quadrature of its representation in Sec. S16, valid for every q > 0 q>0 q > 0 (Sec. S17.4), to 10 − 25 10^{-25} 1 0 − 25 at ( q , ν ) = ( 0.7 , 0.4 ) (q,\nu)=(0.7,0.4) ( q , ν ) = ( 0.7 , 0.4 ) and ( 3 , 2 ) (3,2) ( 3 , 2 ) , with a working precision of 30 digits. A symbolic evaluation of the kite prefactor and two further checks give
A 0 = 3 32 π 5 , ∫ 0 ∞ h 2 d z = 0.32133552109539704 vs. π ( 1 − log 2 ) 3 , A_0=\frac{3}{32\pi^5},\qquad\int_0^\infty h^2\,\mathrm{d}z=0.32133552109539704\ \text{vs.}\ \frac{\pi(1-\log2)}{3}, A 0 = 32 π 5 3 , ∫ 0 ∞ h 2 d z = 0.32133552109539704 vs. 3 π ( 1 − log 2 ) ,
c h d = − 0.000859445718977859 , c_{\mathrm{hd}}=-0.000859445718977859, c hd = − 0.000859445718977859 ,
the last from its analytic expression. Quadrature of main-text Eqs. (29) and (36) (main-text § 2.8 and § 2.9 ) at a working precision of 50 significant digits gives the values below, equal to the closed forms to 40 digits,
3 π 3 ∫ − ∞ ∞ h 2 d z = 0.0621813817 … = 2 ( 1 − log 2 ) π 2 , \frac{3}{\pi^{3}}\int_{-\infty}^{\infty} h^{2}\,\mathrm{d}z = 0.0621813817\ldots= \frac{2(1-\log2)}{\pi^{2}}, π 3 3 ∫ − ∞ ∞ h 2 d z = 0.0621813817 … = π 2 2 ( 1 − log 2 ) ,
residue integral = 0.0130393528 … = α L ( π 2 − 12 log 2 ) 2 π 3 . \text{residue integral} = 0.0130393528\ldots= \frac{\alpha_{L}(\pi^{2}-12\log2)}{2\pi^{3}}. residue integral = 0.0130393528 … = 2 π 3 α L ( π 2 − 12 log 2 ) .
For the direct evaluation of main-text Eq. (40) with f = v f=v f = v , the bare-line frequency integral, main-text Eq. (18) , leaves only transitions of opposite sign. The map k ↦ − k − q \boldsymbol{k}\mapsto-\boldsymbol{k}-\boldsymbol{q} k ↦ − k − q then gives, with C q \mathcal{C}_{q} C q of Eq. (100) ,
K [ v ] = 2 A 0 ∫ d 3 q q − 2 ∫ C q d 3 p ∫ C q d 3 k ∣ ℓ ∣ 2 q ⋅ ℓ − 1 , ℓ = p + k + q . \mathcal{K}[v]=2A_{0}\int\mathrm{d}^{3}q\,q^{-2}\int_{\mathcal{C}_{q}}\mathrm{d}^{3}p\int_{\mathcal{C}_{q}}\mathrm{d}^{3}k\,\lvert\boldsymbol{\ell}\rvert^{2}\,\boldsymbol{q}\cdot\boldsymbol{\ell}^{-1},\qquad\boldsymbol{\ell}=\boldsymbol{p}+\boldsymbol{k}+\boldsymbol{q}. K [ v ] = 2 A 0 ∫ d 3 q q − 2 ∫ C q d 3 p ∫ C q d 3 k ∣ ℓ ∣ 2 q ⋅ ℓ − 1 , ℓ = p + k + q .
The Monte Carlo integration fixes the axis of q \boldsymbol{q} q and draws p \boldsymbol{p} p uniformly in C q \mathcal{C}_{q} C q . It draws k \boldsymbol{k} k from an equal mixture of the uniform density on C q \mathcal{C}_{q} C q and a density proportional to ∣ ℓ ∣ − 2 \lvert\boldsymbol{\ell}\rvert^{-2} ∣ ℓ ∣ − 2 , which absorbs the Coulomb factor. Both densities enter the sample weights. Four independent runs of 10 8 10^{8} 1 0 8 samples give, with the standard error from the spread between runs,
K [ v ] = 0.04836 ( 2 ) vs. e 2 x = 0.0483583 [main-text xref‘Eq. (42)‘] . \mathcal{K}[v]=0.04836(2)\qquad\text{vs.}\qquad e_{2x}=0.0483583\quad\text{[main-text {xref}`Eq. (42)`]}. K [ v ] = 0.04836 ( 2 ) vs. e 2 x = 0.0483583 [main-text xref ‘Eq. (42)‘] .
This tests the prefactor A 0 A_{0} A 0 and the signs of Eqs. (75) –(80) independently of the BRM correspondence of Sec. S11.
Radial reduction. Two evaluations of the double radial integral (S83) for K Π S E ( 1 ) [ b Λ ] \mathcal{K}_{\Pi_{\mathrm{SE}}^{(1)}[b_{\Lambda}]} K Π SE ( 1 ) [ b Λ ] agree to relative differences below 1.5 × 10 − 14 1.5\times10^{-14} 1.5 × 1 0 − 14 at Λ = 0.5 \Lambda=0.5 Λ = 0.5 , 2 and 10. The first uses adaptive quadrature. The second reduces the integral independently to one dimension, with rational and log functions in the integrand, and evaluates it by quadrature at a working precision of 45 significant digits. A third evaluation in the original coordinates ( q , p , θ ) (q,p,\theta) ( q , p , θ ) with composite Gauss–Legendre quadrature converges to the same values as the order increases from 48 to 128 (differences 5.6 × 10 − 8 5.6\times10^{-8} 5.6 × 1 0 − 8 to 2.3 × 10 − 10 2.3\times10^{-10} 2.3 × 1 0 − 10 ). The large-momentum kernel (S103) of the self-energy pair was tested through the scaled error
q 4 [ e Σ , Λ + 3 p l o c π 3 q 2 ] , w Λ = 0.5 , 2 , 10 (fixed) , q = 8 , 32 , 128. q^{4}\left[e_{\Sigma,\Lambda}+\frac{3p_{\mathrm{loc}}}{\pi^{3}q^{2}}\right],\qquad w_{\Lambda}=0.5,\ 2,\ 10\ \text{(fixed)},\qquad q=8,\ 32,\ 128. q 4 [ e Σ , Λ + π 3 q 2 3 p loc ] , w Λ = 0.5 , 2 , 10 (fixed) , q = 8 , 32 , 128.
It approaches a finite value over these momenta, consistent with the q − 4 q^{-4} q − 4 remainder. The factor p l o c p_{\mathrm{loc}} p loc agrees with an independent frequency integral to 2.2 × 10 − 16 2.2\times10^{-16} 2.2 × 1 0 − 16 .
Low-density coefficient. The two-dimensional ( λ , x ) (\lambda,x) ( λ , x ) integral of main-text Eq. (51) , without the beta-function substitution, gives A 1 = − 0.21469376098909768 A_{1}=-0.21469376098909768 A 1 = − 0.21469376098909768 , the value of the gamma-function formula. An evaluation after the coupling substitution, at a working precision of 45 significant digits, agrees. The moment 1 / 4 1/4 1/4 and the self-energy coefficient − A 1 -A_{1} − A 1 are reproduced as
∫ 0 1 r 2 f L d r → 0.25000000000000006 , − A 1 → 0.21469376098909773. \int_{0}^{1}r^{2}f_{\mathrm{L}}\,\mathrm{d}r\to0.25000000000000006,\qquad-A_{1}\to0.21469376098909773. ∫ 0 1 r 2 f L d r → 0.25000000000000006 , − A 1 → 0.21469376098909773.
Hence the sum of the kite and self-energy coefficients is below 6 × 10 − 17 6\times10^{-17} 6 × 1 0 − 17 . This checks the relative normalization of the cancellation, not the next surviving power.
Diagram counts and coupling identities. A direct enumeration of Wick contractions at orders n = 2 n=2 n = 2 –6, and an independent brute-force enumeration at n = 2 n=2 n = 2 –4, give the coefficients of Table S6 and 1 / 4 1/4 1/4 for the two-chain graph with one bubble on each line at n = 4 n = 4 n = 4 (96 contractions). We checked the density-scaling identity, main-text Eq. (45) , and the moment rules for a term r s − p r_s^{-p} r s − p or r s − p log r s r_s^{-p}\log r_s r s − p log r s of F F F ,
2 ∫ 0 1 λ 1 − p d λ = 2 2 − p , 2 ∫ 0 1 λ 1 − p log λ d λ = − 2 ( 2 − p ) 2 , 2\int_{0}^{1}\lambda^{1-p}\,\mathrm{d}\lambda= \frac{2}{2-p}, \qquad2\int_{0}^{1}\lambda^{1-p}\log\lambda\,\mathrm{d}\lambda= -\frac{2}{(2-p)^2}, 2 ∫ 0 1 λ 1 − p d λ = 2 − p 2 , 2 ∫ 0 1 λ 1 − p log λ d λ = − ( 2 − p ) 2 2 ,
on a scalar test function at r s = 10 − 3 r_s = 10^{-3} r s = 1 0 − 3 –10 3 10^{3} 1 0 3 and p = 1 / 4 , 1 / 2 , 3 / 4 p = 1/4, 1/2, 3/4 p = 1/4 , 1/2 , 3/4 (largest discrepancy 8.5 × 10 − 12 8.5 \times10^{-12} 8.5 × 1 0 − 12 ).
Direct high-density values of the kite. None of the checks above evaluates the screened kite energy at finite density. Table S8 lists the direct values of e k i t e e_{\mathrm{kite}} e kite at r s ≤ 0.1 r_s \le0.1 r s ≤ 0.1 used in main-text Sec. 3.3, from the public repository [31 ] , with e 2 x = 48.35832 e_{2x} = 48.35832 e 2 x = 48.35832 mRy. Each value uses 64 replicates, and different r s r_s r s use disjoint seeds. The five values are therefore statistically independent. Weighted fits ( 1 / σ 2 ) (1/\sigma^2) ( 1/ σ 2 ) of
y = e k i t e − e 2 x to c f i t r s log 2 r s + c 1 , f i t r s log r s + c 2 r s y = e_{\mathrm{kite}} - e_{2x} \quad\text{to} \quad c_{\mathrm{fit}}r_s\log^2 r_s + c_{1,\mathrm{fit}}r_s\log r_s + c_2r_s y = e kite − e 2 x to c fit r s log 2 r s + c 1 , fit r s log r s + c 2 r s
give c f i t = − 0.00091 ( 5 ) c_{\mathrm{fit}} = -0.00091(5) c fit = − 0.00091 ( 5 ) Ry for all five points and − 0.00092 ( 12 ) -0.00092(12) − 0.00092 ( 12 ) Ry for the four points with r s ≤ 0.03 r_s \le0.03 r s ≤ 0.03 . The weighted sum of squared residuals for the five points is 0.02, and 380 without the r s log 2 r s r_s\log^2 r_s r s log 2 r s term. With c f i t = c h d c_{\mathrm{fit}} = c_{\mathrm{hd}} c fit = c hd fixed,
c 1 , f i t = 0.01125 ( 6 ) , 0.01134 ( 12 ) , 0.01143 ( 25 ) for r s ≤ 0.1 , 0.03 , 0.01 , c_{1,\mathrm{fit}} = 0.01125(6),\ 0.01134(12),\ 0.01143(25) \quad\text{for } r_s \le0.1,\ 0.03,\ 0.01, c 1 , fit = 0.01125 ( 6 ) , 0.01134 ( 12 ) , 0.01143 ( 25 ) for r s ≤ 0.1 , 0.03 , 0.01 ,
against c 1 = 0.01150 c_1 = 0.01150 c 1 = 0.01150 [9 ] .
Table S8. Direct values of the coupling-averaged kite energy at high density, in mRy, with the standard error σ \sigma σ over replicates [31 ] .
References