betaFeedback

WorksPhysics and Astronomy

A Systematic Convergent Sequence of Approximations (of Integral Equation Form) to the Solutions of the Hedin Equations

Abstract

In many ways the solution to the Hedin equations represents an exact solution to the many body problem. However, for most systems of practical interest, the solution to the Hedin equations is rendered nearly numerically intractable because the Hedin equations are of functional derivative form. Integral equations, on the other hand, are much more numerically tractable, than functional derivative equations, as they can often be solved iteratively. In this work we present a systematic set of integral equations (with no functional derivatives) - Hedin approximations I, II, III, IV etc. - whose solutions converge to the solutions of the exact Hedin equations. The Hedin approximations are well suited to iterative numerical solutions (which we also describe). Furthermore Hedin approximation I is just the GW approximation (as such this work may be viewed as a systematic improvement of the GW approximation). We present a systematic study of the Hedin equations for zero dimensional field theory (which, in particular, is a method to enumerate Feynman diagrams for field theories in arbitrary dimensions) and show better and better convergence to the solutions of the Hedin equations for higher and higher Hedin approximations, with Hedin approximations I, II and III being explicitly studied. We, in particular, show that the higher Hedin approximations capture more and and more Feynman diagrams for the self energy. We also show that already Hedin approximation II captures more diagrams than the state of the art diagrammatic vertex corrections approach. Furthermore Hedin approximation III is a near perfect match to the exact solutions of the Hedin equations, at least in the zero dimensional case, and enumerates a large number of Feynman diagrams.

Introduction

It is known that Many Body Perturbation Theory (MBPT) [1, 2, 3] is well suited for the study of complex quantum systems. MBPT, along with its competitors - Quantum Monte Carlo (QMC) [4, 5] and Coupled Cluster (C.C.) calculations [1] - to name a few, represents some of our best understanding of correlated electron systems. One of the main advantages of MBPT is the direct relation of the computed quantities, in MBPT, to physical observables. In particular, single particle Green’s functions are directly accessible through MBPT - these can be used to compute density of states, as well as the excitation spectra and, with the use of the Galitskii-Migdal formula, the ground state energy of an interacting many-body system [3, 6, 7]. Furthermore two particle Green’s functions, also accessible through MBPT, can be used to compute linear response functions [3, 8].

One of the key methods within MBPT are the Hedin equations [3, 7, 9, 10, 11]. The Hedin equations are a relationship between the single particle propagator GG, the proper self energy Σ\Sigma (which does not contain Hartree insertions), the effective potential WW, the proper polarization PP, and the dressed vertex Γ\Gamma. The Hedin equations are capable of solving most many body problems in that they compute the single particle Green’s functions and, through the associated Bethe-Salpeter equation, the two particle Green’s functions [2]. Furthermore the Hedin equations are widely applicable, in particular they apply when the many body Hamiltonian is of the form:

H=∑ih(xi,pi)+∑i<jv(xi,xj),(1)H = \sum_i h(\mathbf{x}_i,\mathbf{p}_i) + \sum_{i<j} v(\mathbf{x}_i,\mathbf{x}_j), \tag*{(1)}

spin orbit coupling and spin-spin interactions is also possible [12, 13, 14] as well as phonons [2] and transverse photons [15]. Here xi\mathbf{x}_i and pi\mathbf{p}_i are the position and momentum. Here hh is the single particle part while vv is the two particle part of the Hamiltonian, see however also Refs. Yukalov [16, 17]. The Hedin equations for the Hamiltonian in Eq. (1) are given by [3, 7, 9]:

Γ(1,2,3)=δ(1−2)δ(2−3)+∫d(4,5,6,7)δΣ(1,2)δG(4,5)G(4,6)Γ(6,7,3)G(5,7)\Gamma(1,2,3)=\delta(1-2)\delta(2-3)+\int d(4,5,6,7)\frac{\delta\Sigma(1,2)}{\delta G(4,5)}G(4,6)\Gamma(6,7,3)G(5,7)
P(1,2)=−i∫d(3,4)G(1,3)Γ(3,4,2)G(4,1)P(1,2)=-i\int d(3,4)G(1,3)\Gamma(3,4,2)G(4,1)
W(1,2)=v(1,2)+∫d(3,4)v(1,3)P(3,4)W(4,2)W(1,2)=v(1,2)+\int d(3,4)v(1,3)P(3,4)W(4,2)
Σ(1,2)=i∫d(3,4)G(1,3)W(4,1)Γ(3,2,4)\Sigma(1,2)=i\int d(3,4)G(1,3)W(4,1)\Gamma(3,2,4)
G(1,2)=g(1,2)+∫d(3,4)g(1,3)Σ(3,4)G(4,2)(2)G(1,2)=g(1,2)+\int d(3,4)g(1,3)\Sigma(3,4)G(4,2) \tag*{(2)}

Here i=(xi,σi,ti)i=(x_i,\sigma_i,t_i) represents the position, spin and time (both real and imaginary time is possible). Keldysh indices may also be added [2]. Furthermore, g(i,j)g(i,j) is the bare single particle Green’s function with the inclusion of Hartree insertions.

We explicitly see the functional derivative term δΣ(1,2)δG(4,5)\frac{\delta\Sigma(1,2)}{\delta G(4,5)} in Eq. (2). This term greatly increases the complexity of the numerical solution to the Hedin equations [3, 7]. In order to obtain equations without functional derivatives and reduce the Hedin equations to integral equations (which are much easier to solve then functional derivative equations) we propose to introduce new independent variables, Γ\Gamma, PP, WW, Σ\Sigma, GG, δδGΓ\frac{\delta}{\delta G}\Gamma, δδGP\frac{\delta}{\delta G}P, δδGW\frac{\delta}{\delta G}W, δδGΣ\frac{\delta}{\delta G}\Sigma, δ2δG2Γ\frac{\delta^2}{\delta G^2}\Gamma, δ2δG2P\frac{\delta^2}{\delta G^2}P, δ2δG2W\frac{\delta^2}{\delta G^2}W, δ2δG2Σ\frac{\delta^2}{\delta G^2}\Sigma, etc. (where, from now on, we have dropped the labels 1, 2, 3 etc.) That is, we wish to have the functional derivatives become independent variables. We then write the exact equations in terms of these variables (that is use product rule for derivatives and differentiate under the integral sign, repeatedly, on the r.h.s of Eq. (2)) and expand the total number of equations to include those for the functional derivatives - now promoted to independent variables - not functional derivative relations. To make the calculations tractable we truncate the equations at some order of derivative - that is drop all terms of a higher derivative. Hedin approximation I (the GW approximation [3, 7, 9, 10, 11, 18, 19, 20]) is when we truncated at zeroth order derivatives kept, Hedin approximation II is when we truncate at first order derivatives kept, Hedin approximation III is when we truncate at second order derivatives kept etc. Solutions of these equations produces a sequence of better and better approximations to the solutions of the Hedin equations with solutions to Hedin approximation n→∞n\to\infty converging to the exact results. Since Hedin approximation I is just the GW approximation this work may be viewed as a systematic improvement of the GW approximation [3, 7, 9, 10, 11, 19, 21]. We further note that we do not explicitly compute any Feynman diagrams in our approach (which is another method to improve GW approximation [22, 23, 24, 25]), this greatly simplifies the calculations.

We test this approach using zero dimensional field theory [26, 27, 28]. It is known that the Hedin Equations in zero dimensions enumerate Feynman diagrams for field theories in arbitrary dimensions [26, 27, 28]. We show that Hedin approximations I, II, and III correspond to progressively better and better enumerations with more and more diagrams kept. In particular, already Hedin approximation II enumerates more Feynman diagrams then the leading state of the art diagrammatic vertex corrections approach currently used, as far as the author is aware [22, 23, 24, 25]. Furthermore already Hedin approximation III enumerates most low order Feynman diagrams and nearly matches the exact solutions of the Hedin equations.

Main Idea and Examples

The key idea, presented in this work, behind improving the GW approximation to more accurately solve the Hedin equations, without the use of functional derivatives or Feynman diagrams, is to introduce more independent variables, that is promote δδGΓ\frac{\delta}{\delta G}\Gamma, δδGP\frac{\delta}{\delta G}P, δδGW\frac{\delta}{\delta G}W, δδGΣ\frac{\delta}{\delta G}\Sigma, δ2δG2Γ\frac{\delta^2}{\delta G^2}\Gamma, δ2δG2P\frac{\delta^2}{\delta G^2}P, δ2δG2W\frac{\delta^2}{\delta G^2}W, δ2δG2Σ\frac{\delta^2}{\delta G^2}\Sigma, etc. to independent variables (not functional derivative relations) and then write the exact equations in terms of these variables (that is increase the number of integral equations). This is done by repeated application of the product rule for derivatives and differentiating under the integral sign of the r.h.s. of Eq. (2). One then truncates the equations at some order of derivative with respect to GG. If one truncates at zero order derivatives one obtains Hedin approximation I (GW approximation) if one truncates at first order derivatives one obtains Hedin approximation II, if one truncates at second order derivatives one obtains Hedin approximation III etc. We now present formulas for Hedin approximations I, II, III explicitly and then we present the general formula for Hedin approximation n.

Hedin approximation I (GW Approximation)

The Hedin equations for Hedin approximation I (GW approximation) reduce to:

Γ=1P=GGΓW=v+vPWΣ=GWΓG=g+gΣG(3)\begin{aligned} \Gamma&= 1\\ P &= GG\Gamma\\ W &= v+vPW\\ \Sigma&= GW\Gamma\\ G &= g+g\Sigma G \tag*{(3)} \end{aligned}

Where, on top of dropping the labels i,j...., we drop, from now on, the integral signs and replace δ(1−2)δ(2−3)→1\delta(1-2)\delta(2-3)\to1. These equations are well suited for iterative solutions as they from a closed set of equations. Indeed one chooses some initial conditions Γ0\Gamma_0, P0P_0, W0W_0, Σ0\Sigma_0, G0G_0 and writes the loop:

…→Γn→Pn→Wn→Σn→Gn→Γn+1→…(4)\ldots\to\Gamma_n\to P_n\to W_n\to\Sigma_n\to G_n\to\Gamma_{n+1}\to\ldots \tag*{(4)}

Where one uses left to right assignments at each step e.g.

Γn=1Pn=Gn−1Gn−1ΓnWn=v+vPnWn−1Σn=Gn−1WnΓnGn=g+gΣnGn−1(5)\begin{aligned} \Gamma_n &= 1\\ P_n &= G_{n-1}G_{n-1}\Gamma_n\\ W_n &= v+vP_nW_{n-1}\\ \Sigma_n &= G_{n-1}W_n\Gamma_n\\ G_n &= g+g\Sigma_nG_{n-1} \tag*{(5)} \end{aligned}

We note that we have dropped the δΣδGGGΓ\frac{\delta\Sigma}{\delta G}GG\Gamma

Γ=1+(δΣδGGGΓ)(6)\Gamma=1+\left(\frac{\delta\Sigma}{\delta G}GG\Gamma\right) \tag*{(6)}

That is truncated the equations at zero functional derivatives and therefore had to add no new additional equations.

Hedin approximation II

We now write the equations for Hedin approximation II:

Γ=1+δΣδGGGΓP=GGΓW=v+vPWΣ=GWΓG=g+gΣG\begin{aligned} \Gamma&= 1+\frac{\delta\Sigma}{\delta G}GG\Gamma\\ P &= GG\Gamma\\ W &= v+vPW\\ \Sigma&= GW\Gamma\\ G &= g+g\Sigma G \end{aligned}
δΓδG=δδG(1+δΣδGGGΓ)−δ2ΣδG2GGΓ=δΣδGGGδΓδG+δΣδG2GΓδPδG=δδG(GGΓ)=2GΓ+GGδΓδGδWδG=WδPδGWδΣδG=δδG(GWΓ)=WΓ+GδWδGΓ+GWδΓδG(7)\begin{aligned} \frac{\delta\Gamma}{\delta G} &=\frac{\delta}{\delta G}\left(1+\frac{\delta\Sigma}{\delta G}GG\Gamma\right)-\frac{\delta^2\Sigma}{\delta G^2}GG\Gamma\\ &=\frac{\delta\Sigma}{\delta G}GG\frac{\delta\Gamma}{\delta G}+\frac{\delta\Sigma}{\delta G}2G\Gamma\\ \frac{\delta P}{\delta G} &=\frac{\delta}{\delta G}(GG\Gamma)=2G\Gamma+GG\frac{\delta\Gamma}{\delta G}\\ \frac{\delta W}{\delta G} &=W\frac{\delta P}{\delta G}W\\ \frac{\delta\Sigma}{\delta G} &=\frac{\delta}{\delta G}(GW\Gamma)=W\Gamma+G\frac{\delta W}{\delta G}\Gamma+GW\frac{\delta\Gamma}{\delta G} \tag*{(7)} \end{aligned}

We note that we used the exact result that δWδG=WδPδGW\frac{\delta W}{\delta G}=W\frac{\delta P}{\delta G}W [2] to simplify one of the equations. In Equation (7) we have dropped the term δ2ΣδG2GGΓ\frac{\delta^2\Sigma}{\delta G^2}GG\Gamma from:

δΓδG=δδG(1+δΣδGGGΓ)=δΣδGGGδΓδG+δΣδG2GΓ+(δ2ΣδG2GGΓ)(8)\begin{aligned} \frac{\delta\Gamma}{\delta G} &=\frac{\delta}{\delta G}\left(1+\frac{\delta\Sigma}{\delta G}GG\Gamma\right)\\ &=\frac{\delta\Sigma}{\delta G}GG\frac{\delta\Gamma}{\delta G}+\frac{\delta\Sigma}{\delta G}2G\Gamma+\left(\frac{\delta^2\Sigma}{\delta G^2}GG\Gamma\right) \tag*{(8)} \end{aligned}

(which is obtained by applying the product rule on the r.h.s. of Eq. (2) and differentiating under the integral sign). That is we truncated at one functional derivative kept. We note that all the functional derivatives are to be viewed as independent variables in Eq. (7) and not as functional derivative relations. As such we see there are now many new integral equations added to the original Hedin equations - which are obtained by differentiating the original Hedin equations and truncating at first derivative order. With this, we now have the same number of equations as unknowns and these equations are well suited for iterative solutions. Indeed one writes:

…→Γn→Pn→Wn→Σn→Gn→(δΓδG)n→(δPδG)n→(δWδG)n→(δΣδG)n→Γn+1→…(9)\ldots\to\Gamma_n\to P_n\to W_n\to\Sigma_n\to G_n\to\left(\frac{\delta\Gamma}{\delta G}\right)_n\to\left(\frac{\delta P}{\delta G}\right)_n\to\left(\frac{\delta W}{\delta G}\right)_n\to\left(\frac{\delta\Sigma}{\delta G}\right)_n\to\Gamma_{n+1}\to\ldots \tag*{(9)}

Where one uses right to left assignments at each step e.g. (δΓδG)n=(δΓδG)n−1GnGn+(δΣδG)n−12GnΓn\left(\frac{\delta\Gamma}{\delta G}\right)_n=\left(\frac{\delta\Gamma}{\delta G}\right)_{n-1}G_nG_n+\left(\frac{\delta\Sigma}{\delta G}\right)_{n-1}2G_n\Gamma_n.

Hedin approximation III

For Hedin approximation III we write the equations:

Γ=1+δΣδGGGΓP=GGΓW=v+vPWΣ=GWΓG=g+gΣGδΓδG=δΣδGGGδΓδG+δΣδG2GΓ+δ2ΣδG2GGΓδPδG=2GΓ+GGδΓδGδWδG=WδPδGWδΣδG=WΓ+GδWδGΓ+GWδΓδGδ2ΓδG2=δ2ΣδG2GGδΓδG+δΣδG2GδΓδG+δΣδGGGδ2ΓδG2+δ2ΣδG22GΓ+δΣδG2Γ+δΣδG2GδΓδG+δ2ΣδG22GΓ+δ2ΣδG2GGδΓδGδ2PδG2=2Γ+2GδΓδG+2GδΓδG+G2δ2ΓδG2δ2WδG2=WδPδGδPδGW+Wδ2PδG2W+WδPδGWδPδGWδ2ΣδG2=δWδGΓ+WδΓδG+δWδGΓ+Gδ2WδG2Γ+GδWδGδΓδG+WδΓδG+GδWδGδΓδG+GWδ2ΓδG2(10)\begin{aligned} \Gamma&= 1+\frac{\delta\Sigma}{\delta G}GG\Gamma\\ P &= GG\Gamma\\ W &= v+vPW \\ \Sigma&= GW\Gamma\\ G &= g+g\Sigma G \\ \frac{\delta\Gamma}{\delta G} &= \frac{\delta\Sigma}{\delta G}GG\frac{\delta\Gamma}{\delta G}+\frac{\delta\Sigma}{\delta G}2G\Gamma+\frac{\delta^{2}\Sigma}{\delta G^{2}}GG\Gamma\\ \frac{\delta P}{\delta G} &= 2G\Gamma+GG\frac{\delta\Gamma}{\delta G} \\ \frac{\delta W}{\delta G} &= W\frac{\delta P}{\delta G}W \\ \frac{\delta\Sigma}{\delta G} &= W\Gamma+G\frac{\delta W}{\delta G}\Gamma+GW\frac{\delta\Gamma}{\delta G} \\ \frac{\delta^{2}\Gamma}{\delta G^{2}} &= \frac{\delta^{2}\Sigma}{\delta G^{2}}GG\frac{\delta\Gamma}{\delta G}+\frac{\delta\Sigma}{\delta G}2G\frac{\delta\Gamma}{\delta G}+\frac{\delta\Sigma}{\delta G}GG\frac{\delta^{2}\Gamma}{\delta G^{2}} \\ &\quad+\frac{\delta^{2}\Sigma}{\delta G^{2}}2G\Gamma+\frac{\delta\Sigma}{\delta G}2\Gamma+\frac{\delta\Sigma}{\delta G}2G\frac{\delta\Gamma}{\delta G}+\frac{\delta^{2}\Sigma}{\delta G^{2}}2G\Gamma+\frac{\delta^{2}\Sigma}{\delta G^{2}}GG\frac{\delta\Gamma}{\delta G} \\ \frac{\delta^{2}P}{\delta G^{2}} &= 2\Gamma+2G\frac{\delta\Gamma}{\delta G}+2G\frac{\delta\Gamma}{\delta G}+G^{2}\frac{\delta^{2}\Gamma}{\delta G^{2}} \\ \frac{\delta^{2}W}{\delta G^{2}} &= W\frac{\delta P}{\delta G}\frac{\delta P}{\delta G}W+W\frac{\delta^{2}P}{\delta G^{2}}W+W\frac{\delta P}{\delta G}W\frac{\delta P}{\delta G}W \\ \frac{\delta^{2}\Sigma}{\delta G^{2}} &= \frac{\delta W}{\delta G}\Gamma+W\frac{\delta\Gamma}{\delta G}+\frac{\delta W}{\delta G}\Gamma+G\frac{\delta^{2}W}{\delta G^{2}}\Gamma\\ &\quad+G\frac{\delta W}{\delta G}\frac{\delta\Gamma}{\delta G}+W\frac{\delta\Gamma}{\delta G}+G\frac{\delta W}{\delta G}\frac{\delta\Gamma}{\delta G}+GW\frac{\delta^{2}\Gamma}{\delta G^{2}} \tag*{(10)} \end{aligned}

Where we have dropped δ3ΣδG3GGΓ\frac{\delta^{3}\Sigma}{\delta G^{3}}GG\Gamma from:

δ2ΓδG2=δ2δG2(1+δΣδGGGΓ)=δ2ΣδG2GGδΓδG+δΣδG2GδΓδG+δΣδGGGδ2ΓδG2+δ2ΣδG22GΓ+δΣδG2Γ+δΣδG2GδΓδG+δ2ΣδG22GΓ+δ2ΣδG2GGδΓδG+(δ3ΣδG3GGΓ)(11)\begin{aligned} \frac{\delta^{2}\Gamma}{\delta G^{2}} &=\frac{\delta^{2}}{\delta G^{2}}\left(1+\frac{\delta\Sigma}{\delta G}GG\Gamma\right) \\ &=\frac{\delta^{2}\Sigma}{\delta G^{2}}GG\frac{\delta\Gamma}{\delta G}+\frac{\delta\Sigma}{\delta G}2G\frac{\delta\Gamma}{\delta G}+\frac{\delta\Sigma}{\delta G}GG\frac{\delta^{2}\Gamma}{\delta G^{2}} \\ &\quad+\frac{\delta^{2}\Sigma}{\delta G^{2}}2G\Gamma+\frac{\delta\Sigma}{\delta G}2\Gamma+\frac{\delta\Sigma}{\delta G}2G\frac{\delta\Gamma}{\delta G}+\frac{\delta^{2}\Sigma}{\delta G^{2}}2G\Gamma\\ &\quad+\frac{\delta^{2}\Sigma}{\delta G^{2}}GG\frac{\delta\Gamma}{\delta G}+\left(\frac{\delta^{3}\Sigma}{\delta G^{3}}GG\Gamma\right) \tag*{(11)} \end{aligned}

which would be obtained by using product rule on the r.h.s. of Eq. (2) and differentiating under the integral sign (that is truncated to two derivatives kept). These equations are well suited for iterative solutions as they form a closed set of equations as we have added many additional equations beyond the Hedin ones - by differentiating the Hedin equations and truncating at second derivative order. To numerically solve these equations one writes:

…→Γn→Pn→Wn→Σn→Gn→(δΓδG)n→(δPδG)n→(δWδG)n→(δΣδG)n→(δ2ΓδG2)n→(δ2PδG2)n→(δ2WδG2)n→(δ2ΣδG2)n→Γn+1→…(12)\ldots\to\Gamma_n \to P_n \to W_n \to\Sigma_n \to G_n \to\left(\frac{\delta\Gamma}{\delta G}\right)_n \to\left(\frac{\delta P}{\delta G}\right)_n \to\left(\frac{\delta W}{\delta G}\right)_n \to\left(\frac{\delta\Sigma}{\delta G}\right)_n \to\left(\frac{\delta^{2}\Gamma}{\delta G^{2}}\right)_n \to\left(\frac{\delta^{2}P}{\delta G^{2}}\right)_n \to\left(\frac{\delta^{2}W}{\delta G^{2}}\right)_n \to\left(\frac{\delta^{2}\Sigma}{\delta G^{2}}\right)_n \to\Gamma_{n+1}\to\ldots \tag*{(12)}

Where one uses right to left assignments at each step. Hedin IV and higher are similar to Hedin I, II, III but more tedious. We will not need their explicit forms below, but we given the general form of Hedin n.

Hedin approximation n

For Hedin approximation n we write the equations:

Γ=1+δΣδGGGΓP=GGΓW=v+vPWΣ=GWΓG=g+gΣG⋯=⋯⋯=⋯⋯=⋯⋯=⋯δn−1ΓδGn−1=δn−1δGn−1(δΣδGGGΓ)−δnΣδGnGGΓδn−1PδGn−1=δn−1δGn−1(GGΓ)δn−1WδGn−1=δn−2δGn−2(WδPδGW)δn−1ΣδGn−1=δn−1δGn−1(GWΓ)(13)\begin{aligned} \Gamma&= 1+\frac{\delta\Sigma}{\delta G}GG\Gamma\\ P &= GG\Gamma\\ W &= v+vPW \\ \Sigma&= GW\Gamma\\ G &= g+g\Sigma G \\ \cdots&= \cdots\\ \cdots&= \cdots\\ \cdots&= \cdots\\ \cdots&= \cdots\\ \frac{\delta^{n-1}\Gamma}{\delta G^{n-1}} &= \frac{\delta^{n-1}}{\delta G^{n-1}}\left(\frac{\delta\Sigma}{\delta G}GG\Gamma\right)-\frac{\delta^{n}\Sigma}{\delta G^{n}}GG\Gamma\\ \frac{\delta^{n-1}P}{\delta G^{n-1}} &= \frac{\delta^{n-1}}{\delta G^{n-1}}(GG\Gamma) \\ \frac{\delta^{n-1}W}{\delta G^{n-1}} &= \frac{\delta^{n-2}}{\delta G^{n-2}}\left(W\frac{\delta P}{\delta G}W\right) \\ \frac{\delta^{n-1}\Sigma}{\delta G^{n-1}} &= \frac{\delta^{n-1}}{\delta G^{n-1}}(GW\Gamma) \tag*{(13)} \end{aligned}

We notice that we have truncated the equations at n−1n-1 functional derivatives by dropping the term δnΣδGnGGΓ\frac{\delta^{n}\Sigma}{\delta G^{n}}GG\Gamma. All the variables in Eq. (13) are viewed as independent variables and not as functional derivative relations. This makes these equations well suited for iterative solutions as they are of integral equation form. We note that for n=In=I we explicitly have the GW equations while for n→∞n\to\infty the dropped term δnΣδGnGGΓ\frac{\delta^{n}\Sigma}{\delta G^{n}}GG\Gamma has negligible effect on Γ\Gamma, PP, WW, Σ\Sigma and GG so that the solutions of Hedin approximation n→∞n\to\infty reduce to the solutions of the Hedin equations. We note that for finite nn, when we drop the term δnΣδGnGGΓ\frac{\delta^{n}\Sigma}{\delta G^{n}}GG\Gamma - we drop a certain class of Feynman diagrams. As such, we do not enumerate all Feynman diagrams that enter the MBPT Feynman diagram series for any finite nn - which we emphasize is fully captured by the exact Hedin Equations.

Zero Dimensional Field Theory

Here we do some numerical simulations that confirm the efficiency of our approach. We focus on zero dimensional field theory [26, 27, 28] - which is an efficient way to enumerate Feynman diagrams for the Schrödinger field theory in arbitrary dimensions [26, 27, 28, 29]. We explicitly calculate the number of Feynman diagrams that enter the self energy at every order of perturbation theory upto fifth order beyond Hartree-Fock (we focus on the self energy for simplicity and concreteness; because of its physical significance [2, 3] and to follow Ref. [26]). We find that the exact field theory captures more diagrams then Hedin approximation III, which captures more diagrams the Hedin Approximation II which captures more diagrams then the state of the art diagrammatic vertex corrections approach which captures more diagrams then the GW approximation (Hedin approximation I). We note that Hedin approximation III is already a nearly perfect match to the exact solutions of the Hedin equations. We note that the Hedin equations in zero dimensions are given by [26]:

Γ=1+G2∂Σ∂GΓP=lG2ΓW=v+PWΣ=GWΓG=g+gΣG(14)\begin{aligned} \Gamma&= 1 + G^2 \frac{\partial\Sigma}{\partial G}\Gamma\\ P &= lG^2\Gamma\\ W &= v + PW \\ \Sigma&= GW\Gamma\\ G &= g + g\Sigma G \tag*{(14)} \end{aligned}

Where we have introduced the fermion loop counting parameter ll following Ref. [26].

Diagrammatic vertex corrections to GW

We introduce the dimensionless interaction strength x=g2vx = g^2v following Ref. [26, 28]. The Hedin equations for the state of the art diagrammatic vertex correction approach [22, 23, 24] can be written as:

ΓD=1+(xGD2WD+2lx2GD4WD2)ΓDPD=GD2ΓDWD=1+lxPDWDsD=GDWDΓDGD=1+xsDGD(15)\begin{aligned} \Gamma_D &= 1 + \left(x\mathcal{G}_D^2\mathcal{W}_D + 2lx^2\mathcal{G}_D^4\mathcal{W}_D^2\right)\Gamma_D \\ \mathcal{P}_D &= \mathcal{G}_D^2\Gamma_D \\ \mathcal{W}_D &= 1 + lx\mathcal{P}_D\mathcal{W}_D \\ s_D &= \mathcal{G}_D\mathcal{W}_D\Gamma_D \\ \mathcal{G}_D &= 1 + xs_D\mathcal{G}_D \tag*{(15)} \end{aligned}

Where we have introduced dimensionless parameters: ΓD=Γ\Gamma_D = \Gamma, PD=Plg2\mathcal{P}_D = \frac{P}{lg^2}, WD=Wv\mathcal{W}_D = \frac{W}{v}, sD=Σgvs_D = \frac{\Sigma}{gv}, GD=Gg\mathcal{G}_D = \frac{G}{g}. Iterating Eq. (15) starting at ΓD=1,PD=0,WD=1,sD=0,GD=1\Gamma_D = 1,\mathcal{P}_D = 0,\mathcal{W}_D = 1,s_D = 0,\mathcal{G}_D = 1 with l=1l = 1 (the fermion loop counting parameter is set to one - so we just count the total number of diagrams), we obtain that:

sD(x)=1+3x+16x2+103x3+733x4+5556x5+…(16)s_D(x) = 1 + 3x + 16x^2 + 103x^3 + 733x^4 + 5556x^5 + \ldots \tag*{(16)}

We note that each coefficient in Eq. (16) counts the number of diagrams of that order in the diagrammatic expansion of the self energy captured by the diagrammatic vertex correction approach. That is, the diagrammatic expansion in Eq. (15) captures one diagram at first oder of perturbation theory (note that Hartree insertions are included in gg - so there is only one diagram at first order in perturbation theory), three diagram at second order perturbation theory, sixteen diagrams at third order perturbation theory, 103 diagrams at fourth order perturbation theory, 733 diagrams at fifth order perturbation theory and 5556 diagrams at sixth order in perturbation theory in the bare couplings for the self energy. This is true for arbitrary Hamiltonians of the form in Eq. (1) in arbitrary dimensions.

Self energy for Hedin approximation II

Introducing the dimensionless variables:

ΓII=Γ,PII=Plg2,WII=Wv,sII=Σgv,GII=Gg,\Gamma_{II} = \Gamma,\mathcal{P}_{II} = \frac{P}{lg^2},\mathcal{W}_{II} = \frac{W}{v},s_{II} = \frac{\Sigma}{gv},\mathcal{G}_{II} = \frac{G}{g},
δΓII=g∂Γ∂G,δPII=glg∂P∂G,δWII=gv∂W∂G,δsII=∂Σ∂G(17)\delta\Gamma_{II} = g\frac{\partial\Gamma}{\partial G},\delta\mathcal{P}_{II} = \frac{g}{lg}\frac{\partial P}{\partial G},\delta\mathcal{W}_{II} = \frac{g}{v}\frac{\partial W}{\partial G},\delta s_{II} = \frac{\partial\Sigma}{\partial G} \tag*{(17)}

We get that the Hedin approximation II equations reduce to:

ΓII=1+xδsIIGII2ΓIIPII=GII2ΓIIWII=1+lxPIIWIIsII=GIIWIIΓIIGII=1+xsIIGIIδΓII=xGII2δsIIΓII+2xδsIIGIIΓIIδPII=2GIIΓII+lGII2δΓIIδWII=lxWII2δPIIδsII=WIIΓII+GIIδWIIΓII+GIIWIIδΓII(18)\begin{aligned} \Gamma_{II} &= 1 + x\delta s_{II}\mathcal{G}_{II}^2\Gamma_{II} \\ \mathcal{P}_{II} &= \mathcal{G}_{II}^2\Gamma_{II} \\ \mathcal{W}_{II} &= 1 + lx\mathcal{P}_{II}\mathcal{W}_{II} \\ s_{II} &= \mathcal{G}_{II}\mathcal{W}_{II}\Gamma_{II} \\ \mathcal{G}_{II} &= 1 + xs_{II}\mathcal{G}_{II} \\ \delta\Gamma_{II} &= x\mathcal{G}_{II}^2\delta s_{II}\Gamma_{II} + 2x\delta s_{II}\mathcal{G}_{II}\Gamma_{II} \\ \delta\mathcal{P}_{II} &= 2\mathcal{G}_{II}\Gamma_{II} + l\mathcal{G}_{II}^2\delta\Gamma_{II} \\ \delta\mathcal{W}_{II} &= lx\mathcal{W}_{II}^2\delta\mathcal{P}_{II} \\ \delta s_{II} &= \mathcal{W}_{II}\Gamma_{II} + \mathcal{G}_{II}\delta\mathcal{W}_{II}\Gamma_{II} + \mathcal{G}_{II}\mathcal{W}_{II}\delta\Gamma_{II} \tag*{(18)} \end{aligned}

We iterate Eq. (18) starting with:

ΓII=1,PII=0,WII=1,sII=0,GII=1,δΓII=0,δPII=0,δWII=0,δsII=0(19)\Gamma_{II} = 1,\mathcal{P}_{II} = 0,\mathcal{W}_{II} = 1,s_{II} = 0,\mathcal{G}_{II} = 1,\delta\Gamma_{II} = 0,\delta\mathcal{P}_{II} = 0,\delta\mathcal{W}_{II} = 0,\delta s_{II} = 0 \tag*{(19)}

We get that for l=1l = 1 we have that:

sII(x)=1+3x+18x2+146x3+1385x4+14344x5+…(20)s_{II}(x) = 1 + 3x + 18x^2 + 146x^3 + 1385x^4 + 14344x^5 + \ldots \tag*{(20)}

We note that already Hedin II captures many more Feynman diagrams then the leading order vertex correction scheme at every order in perturbation theory (see also table I).

Self energy within Hedin approximation III

Introducing the dimensionless variables:

ΓIII=Γ,PIII=Plg2,WIII=Wv,sIII=Σgv,GIII=Gg,\Gamma_{III}=\Gamma,\mathcal{P}_{III}=\frac{P}{lg^{2}},\mathcal{W}_{III}=\frac{W}{v},s_{III}=\frac{\Sigma}{gv},\mathcal{G}_{III}=\frac{G}{g},
δΓIII=g∂Γ∂G,δPIII=∂2Pl∂G,δWIII=gv∂W∂G,δsIII=∂Σ∂G,\delta\Gamma_{III}=g\frac{\partial\Gamma}{\partial G},\delta\mathcal{P}_{III}=\frac{\partial^{2}P}{l\partial G},\delta\mathcal{W}_{III}=\frac{g}{v}\frac{\partial W}{\partial G},\delta s_{III}=\frac{\partial\Sigma}{\partial G},
δ2ΓIII=g2∂2Γ∂G2,δ2PIII=∂2Pl∂G2,δ2WIII=g2v∂2W∂G2,δ2sIII=g∂2Σv∂G2(21)\delta^{2}\Gamma_{III}=g^{2}\frac{\partial^{2}\Gamma}{\partial G^{2}},\delta^{2}\mathcal{P}_{III}=\frac{\partial^{2}P}{l\partial G^{2}},\delta^{2}\mathcal{W}_{III}=\frac{g^{2}}{v}\frac{\partial^{2}W}{\partial G^{2}},\delta^{2}s_{III}=g\frac{\partial^{2}\Sigma}{v\partial G^{2}} \tag*{(21)}

then we have that the Hedin approximation III can be written as:

ΓIII=1+xδsIIIGIII2ΓIIIPIII=GIII2ΓIIIWIII=1+lxPIIIWIIIsIII=GIIIWIIIΓIIIGIII=1+xsIIIGIIIδΓIII=xGIII2δsIIIδΓIII+2xδsIIIGIIIΓIII+xδ2sIIIGIII2ΓIIIδPIII=2GIIIΓIII+lGIII2δΓIIIδWIII=lxWIII2δPIIIδsIII=WIIIΓIII+GIIIδWIIIΓIII+GIIIWIIIδΓIIIδ2ΓIII=2xδ2sIIIGIII2δΓIII+4xδsIIIGIIIδΓIII+xδsIIIGIII2δ2ΓIII+4xδ2sIIIGIIIΓIII+2xδsIIIΓIIIδ2PIII=2ΓIII+4GIIIδΓIII+GIII2δ2ΓIIIδ2WIII=2l2xWIII3δPIII2+lxWIII2δ2PIIIδ2sIII=2δWIIIΓIII+2WIIIδΓIII+2GIIIδWIIIδΓIII+GIIIδ2WIIIΓIII+GIIIWIIIδ2ΓIII(22)\begin{aligned} \Gamma_{III}&=1+x\delta s_{III}\mathcal{G}_{III}^{2}\Gamma_{III}\\ \mathcal{P}_{III}&=\mathcal{G}_{III}^{2}\Gamma_{III}\\ \mathcal{W}_{III}&=1+lx\mathcal{P}_{III}\mathcal{W}_{III}\\ s_{III}&=\mathcal{G}_{III}\mathcal{W}_{III}\Gamma_{III}\\ \mathcal{G}_{III}&=1+xs_{III}\mathcal{G}_{III}\\ \delta\Gamma_{III}&=x\mathcal{G}_{III}^{2}\delta s_{III}\delta\Gamma_{III}+2x\delta s_{III}\mathcal{G}_{III}\Gamma_{III}+x\delta^{2}s_{III}\mathcal{G}_{III}^{2}\Gamma_{III}\\ \delta\mathcal{P}_{III}&=2\mathcal{G}_{III}\Gamma_{III}+l\mathcal{G}_{III}^{2}\delta\Gamma_{III}\\ \delta\mathcal{W}_{III}&=lx\mathcal{W}_{III}^{2}\delta\mathcal{P}_{III}\\ \delta s_{III}&=\mathcal{W}_{III}\Gamma_{III}+\mathcal{G}_{III}\delta\mathcal{W}_{III}\Gamma_{III}+\mathcal{G}_{III}\mathcal{W}_{III}\delta\Gamma_{III}\\ \delta^{2}\Gamma_{III}&=2x\delta^{2}s_{III}\mathcal{G}_{III}^{2}\delta\Gamma_{III}+4x\delta s_{III}\mathcal{G}_{III}\delta\Gamma_{III}+x\delta s_{III}\mathcal{G}_{III}^{2}\delta^{2}\Gamma_{III}+4x\delta^{2}s_{III}\mathcal{G}_{III}\Gamma_{III}+2x\delta s_{III}\Gamma_{III}\\ \delta^{2}\mathcal{P}_{III}&=2\Gamma_{III}+4\mathcal{G}_{III}\delta\Gamma_{III}+\mathcal{G}_{III}^{2}\delta^{2}\Gamma_{III}\\ \delta^{2}\mathcal{W}_{III}&=2l^{2}x\mathcal{W}_{III}^{3}\delta\mathcal{P}_{III}^{2}+lx\mathcal{W}_{III}^{2}\delta^{2}\mathcal{P}_{III}\\ \delta^{2}s_{III}&=2\delta\mathcal{W}_{III}\Gamma_{III}+2\mathcal{W}_{III}\delta\Gamma_{III}+2\mathcal{G}_{III}\delta\mathcal{W}_{III}\delta\Gamma_{III}+\mathcal{G}_{III}\delta^{2}\mathcal{W}_{III}\Gamma_{III}+\mathcal{G}_{III}\mathcal{W}_{III}\delta^{2}\Gamma_{III} \tag*{(22)} \end{aligned}

Whereby we obtain for l=1l=1:

sIII(x)=1+3x+20x2+186x3+2153x4+29024x5+…(23)s_{III}(x)=1+3x+20x^{2}+186x^{3}+2153x^{4}+29024x^{5}+\ldots \tag*{(23)}

We shall see below that this is a nearly perfect match to the exact series with nearly all Feynman diagrams captured (see also table I).

Final results

We note that the exact series in xx for the Hedin equations for the self energy is given by [26]:

sE(x)=1+3x+20x2+189x3+2232x4+31130x5+…(24)s_{E}(x)=1+3x+20x^{2}+189x^{3}+2232x^{4}+31130x^{5}+\ldots \tag*{(24)}

We now emphasize that Hedin approximation III captures one out of one leading order diagrams, three out of three second order diagrams, twenty out of twenty third order diagrams, 186 out of 189 fourth order diagrams, 2153 out of 2232 fifth order diagrams, 29024 out

MethodSelf Energy Series
GW1+2x+7x2+30x3+143x4+728x5+…1+2x+7x^{2}+30x^{3}+143x^{4}+728x^{5}+\ldots
Vertex1+3x+16x2+103x3+733x4+5556x5+…1+3x+16x^{2}+103x^{3}+733x^{4}+5556x^{5}+\ldots
Hedin II1+3x+18x2+146x3+1385x4+14344x5+…1+3x+18x^{2}+146x^{3}+1385x^{4}+14344x^{5}+\ldots
Hedin III1+3x+20x2+186x3+2153x4+29024x5+…1+3x+20x^{2}+186x^{3}+2153x^{4}+29024x^{5}+\ldots
Exact1+3x+20x2+189x3+2232x4+31130x5+…1+3x+20x^{2}+189x^{3}+2232x^{4}+31130x^{5}+\ldots

Table 1. A table of the power series expansions (enumeration of the number of Feynman diagrams) of the dimensionless self energy for various methods: GW, diagrammatic vertex corrections, Hedin approximation II, Hedin approximation III and the exact solutions.

of 31130 sixth order diagrams for the self energy and as such is a near perfect match to the exact results.

We note that the GW series (Hedin approximation I) is given by [26]:

sGW(x)=1+2x+7x2+30x3+143x4+728x5+…(25)s_{GW}(x)=1+2x+7x^{2}+30x^{3}+143x^{4}+728x^{5}+\ldots \tag*{(25)}

We see that Hedin II captures many more diagrams then the GW approximation this is further supported by the curves Figure 1 and the series in table I). We note again that Hedin II is already sufficient to beat the state of the art vertex corrections as it captures many more Feynman diagrams, see Figure 1. Furthermore as we see in Figure 1 the Hedin III approximation is a near perfect match to the exact self energy curves (see Table I and Figure 1).

Self energy graphs for the GW series, diagrammatic vertex correction, Hedin approximation II, Hedin approximation III and the exact series

Figure 1. Self energy graphs for the GW series, diagrammatic vertex correction, Hedin approximation II, Hedin approximation III and the exact series. We notice that Hedin approximation II is already better (captures the self energy more closely) then the state of the art diagrammatic expansion. Furthermore Hedin III is nearly on top of the exact self energy in zero dimensions and captures nearly all Feynman diagrams.

Conclusions and Discussions

In this work we have studied a sequence of integral equation approximations to the solutions of the Hedin equations: Hedin approximation I, II, III, IV etc. We have shown that Hedin approximation I is equivalent to the GW approximation while the solutions to Hedin approximations n→∞n \to\infty are equivalent to the exact solutions to the Hedin equations (as such we have introduced a systematic method to improve the GW approximation without resorting to functional derivatives or Feynman diagrams). The main idea for this sequence of approximations was to promote functional derivatives (with respect to the propagator GG) to independent variables, study the exact set of equations thereby obtained (that is differentiate the original Hedin equations, in Eq. (2), using product rule inside the integral sign and generate many new equations) and truncate at some order of derivative kept (that is drop all terms with higher derivatives in GG). This leads to purely integral equations with no functional derivatives - which are known to be significantly easier to solve. We presented the example of zero dimensional field theory and showed explicitly for Hedin approximations I, II and III that higher Hedin approximations are closer to the exact solutions of the Hedin equations and retain more terms in a power series expansion in the dimensionless interaction strength. We also showed that Hedin approximation II already captures more Feynman diagrams then the state of the art diagrammatic vertex correction calculation [23, 24, 25], while Hedin III is nearly a perfect match to the exact Hedin equations.

In the future it would be of interest to extend these calculations to the uniform electron gas, Hubbard model, simple compounds and elements as well as small molecules. For small molecules the GW approximation (Hedin approximation I) is known to work very well [30] so it would be of great interest to see the improvement of the higher Hedin approximations: II, III, IV etc. It would also be of interest to combine Hedin approximations II, III, IV etc. with Dynamical Mean Field Theory (DMFT) much like GW+eDMFT [31]. It would also be of interest to extend these ideas to phonons as in the Hedin-Baym equations [2] and to spin-orbit coupling [12, 13, 14].

We would like to further note that in this work we are proposing a new method to solve systems of differential equations. Indeed suppose we have a differential equation of the form:

0=F(x,X,∂Xi∂xj)(26)0 = \mathbf{F}\left(\mathbf{x},\mathbf{X},\frac{\partial X_i}{\partial x_j}\right) \tag*{(26)}

That is X\mathbf{X} is a vector field over x\mathbf{x} that satisfies the differential equation in Eq. (26) for some vector field F\mathbf{F}. Higher derivative versions are also possible. We can simplify this equation by writing a series of approximations of the form:

0=F(x,X,∂Xi∂xj→0)(27)0 = \mathbf{F}\left(\mathbf{x},\mathbf{X},\frac{\partial X_i}{\partial x_j} \to0\right) \tag*{(27)}

and call it approximation I, and of the form

0=F(x,X,∂Xi∂xj)0 = \mathbf{F}\left(\mathbf{x},\mathbf{X},\frac{\partial X_i}{\partial x_j}\right)
0=∂Fi∂xj(x,X,∂Xk∂xl,∂2Xm∂xj∂xl→0)(28)0 = \frac{\partial F_i}{\partial x_j}\left(\mathbf{x},\mathbf{X},\frac{\partial X_k}{\partial x_l},\frac{\partial^2 X_m}{\partial x_j\partial x_l} \to0\right) \tag*{(28)}

and call it approximation II, and of the form:

0=F(x,X,∂Xi∂xj)0 = \mathbf{F}\left(\mathbf{x},\mathbf{X},\frac{\partial X_i}{\partial x_j}\right)
0=∂Fi∂xj(x,X,∂Xk∂xl,∂2Xm∂xj∂xl)0 = \frac{\partial F_i}{\partial x_j}\left(\mathbf{x},\mathbf{X},\frac{\partial X_k}{\partial x_l},\frac{\partial^2 X_m}{\partial x_j\partial x_l}\right)
0=∂2Fi∂xj∂xk(x,X,∂Xp∂xl,∂2Xm∂xp∂xl,∂3Xm∂xj∂xk∂xl→0)(29)0 = \frac{\partial^2 F_i}{\partial x_j\partial x_k}\left(\mathbf{x},\mathbf{X},\frac{\partial X_p}{\partial x_l},\frac{\partial^2 X_m}{\partial x_p\partial x_l},\frac{\partial^3 X_m}{\partial x_j\partial x_k\partial x_l} \to0\right) \tag*{(29)}

which is approximation III and so forth. In the future it would be of interest to see applications of this technique to the solutions of differential or functional derivative equations, in particular to see which equations besides the Hedin ones are particularly amenable to this technique. Based on our experience with Hedin equations these should be singular differential equations - where the highest derivative terms are multiplied by functions that vanish at specific points and therefore contribute little to the equations (may be dropped to leading order near those points).

Acknowledgements: The author would like to acknowledge Gabriel Kotliar for many useful discussions.

Some variations

First variation

Hedin Equations

The Hedin Equations may also be written as [26]:

Γ(1,2,3)=δ(1−2)δ(2−3)+∫d(4,5,6,7)δΣ(1,2)δg(4,5)g(4,3)g(5,3)P(1,2)=−i∫d(3,4)G(1,3)Γ(3,4,2)G(4,1)W(1,2)=v(1,2)+∫d(3,4)v(1,3)P(3,4)W(4,2)Σ(1,2)=i∫d(3,4)G(1,3)W(4,1)Γ(3,2,4)G(1,2)=g(1,2)+∫d(3,4)g(1,3)Σ(3,4)G(4,2)(30)\begin{aligned} \Gamma(1,2,3) &= \delta(1-2)\delta(2-3) + \int d(4,5,6,7)\frac{\delta\Sigma(1,2)}{\delta g(4,5)}g(4,3)g(5,3) \\ P(1,2) &= -i\int d(3,4)G(1,3)\Gamma(3,4,2)G(4,1) \\ W(1,2) &= v(1,2)+\int d(3,4)v(1,3)P(3,4)W(4,2) \\ \Sigma(1,2) &= i\int d(3,4)G(1,3)W(4,1)\Gamma(3,2,4) \\ G(1,2) &= g(1,2)+\int d(3,4)g(1,3)\Sigma(3,4)G(4,2) \tag*{(30)} \end{aligned}

This version of the Hedin equations is also amenable to our approximations method we call these the Hedin approximations I’, II’, III’ etc.

Hedin approximations n’

The Hedin equations for the Hedin approximation n’ are given by:

Γ=1+δΣδgggP=GGΓW=v+vPWΣ=GWΓG=g+gΣG⋯=⋯⋯=⋯⋯=⋯⋯=⋯δn′−1Γδgn′−1=δn′−1δgn′−1(δΣδggg)−δn′Σδgn′−1ggδn′−1Pδgn′−1=δn′−1δgn′−1(GGΓ)δn′−1Wδgn′−1=δn′−2δgn′−2(WδPδgW)δn′−1Σδgn′−1=δn′−1δgn′−1(GWΓ)δn′−1Gδgn′−1=δn′−1δgn′−1(g+gΣG)(31)\begin{aligned} \Gamma&= 1+\frac{\delta\Sigma}{\delta g}gg \\ P &= GG\Gamma\\ W &= v+vPW \\ \Sigma&= GW\Gamma\\ G &= g+g\Sigma G \\ \cdots&= \cdots\\ \cdots&= \cdots\\ \cdots&= \cdots\\ \cdots&= \cdots\\ \frac{\delta n'^{-1}\Gamma}{\delta g n'^{-1}} &= \frac{\delta n'^{-1}}{\delta g n'^{-1}}\left(\frac{\delta\Sigma}{\delta g}gg\right)-\frac{\delta n'\Sigma}{\delta g n'^{-1}}gg \\ \frac{\delta n'^{-1}P}{\delta g n'^{-1}} &= \frac{\delta n'^{-1}}{\delta g n'^{-1}}(GG\Gamma) \\ \frac{\delta n'^{-1}W}{\delta g n'^{-1}} &= \frac{\delta n'^{-2}}{\delta g n'^{-2}}\left(W\frac{\delta P}{\delta g}W\right) \\ \frac{\delta n'^{-1}\Sigma}{\delta g n'^{-1}} &= \frac{\delta n'^{-1}}{\delta g n'^{-1}}(GW\Gamma) \\ \frac{\delta n'^{-1}G}{\delta g n'^{-1}} &= \frac{\delta n'^{-1}}{\delta g n'^{-1}}(g+g\Sigma G) \tag*{(31)} \end{aligned}

Which is very similar to Hedin approximation nn, in particular we have that Hedin approximation n′→∞n' \to\infty converges to the exact Hedin equations. These equations may be efficiently iterated in order to obtain their solutions.

Second variation

Hedin Equations

The Hedin equations may also be written as:

Γ(1,2∣3,4)=δ(1−3)δ(2−4)+∫d(5,6)δΣ(1,2)δG(5,6)K(5,6∣3,4)K(1,2∣3,4)=i∫d(5,6)[G(1,5)G(6,2)]Γ(5,6∣3,4)W(1,2∣3,4)=v(1,2∣3,4)+∫d(5,6,7,8)v(1,2∣5,6)K(5,6∣7,8)W(7,8∣3,4)Σ(1,2)=i∫d(3,4,5,6)G(3,4)Γ(4,2∣5,6)W(5,6∣3,1)G(1,2)=g(1,2)+∫d(3,4)g(1,3)Σ(3,4)G(4,2)(32)\begin{aligned} \Gamma(1,2 \mid3,4) &= \delta(1-3)\delta(2-4) + \int d(5,6)\frac{\delta\Sigma(1,2)}{\delta G(5,6)}K(5,6 \mid3,4) \\ K(1,2 \mid3,4) &= i\int d(5,6)[G(1,5)G(6,2)]\Gamma(5,6 \mid3,4) \\ W(1,2 \mid3,4) &= v(1,2 \mid3,4) + \int d(5,6,7,8)v(1,2 \mid5,6)K(5,6 \mid7,8)W(7,8 \mid3,4) \\ \Sigma(1,2) &= i\int d(3,4,5,6)G(3,4)\Gamma(4,2 \mid5,6)W(5,6 \mid3,1) \\ G(1,2) &= g(1,2) + \int d(3,4)g(1,3)\Sigma(3,4)G(4,2) \tag*{(32)} \end{aligned}

This version of the Hedin equations is also amenable to our approximations method we call these the Hedin approximations I”, II”, III” etc.

Hedin approximations n”

The Hedin equations for the Hedin approximation n′′n'' are given by:

Γ=1+δΣδGKK=GGΓW=v+vKWΣ=GWΓG=g+gΣG⋯=⋯⋯=⋯⋯=⋯⋯=⋯\begin{aligned} \Gamma&= 1 + \frac{\delta\Sigma}{\delta G}K \\ K &= GG\Gamma\\ W &= v + vKW \\ \Sigma&= GW\Gamma\\ G &= g + g\Sigma G \\ \cdots&= \cdots\\ \cdots&= \cdots\\ \cdots&= \cdots\\ \cdots&= \cdots \end{aligned}
δn′′−1ΓδGn′′−1=δn′′−1δGn′′−1(δΣδGK)=δn′′ΣδGn′′Kδn′′−1KδGn′′−1=δn′′−1δGn′′−1(GGΓ)δn′′−1WδGn′′−1=δn′′−1δGn′′−1(v+vKW)δn′′−1ΣδGn′′−1=δn′′−1δGn′′−1(GWΓ)(33)\begin{aligned} \frac{\delta^{n''-1}\Gamma}{\delta G^{n''-1}} &= \frac{\delta^{n''-1}}{\delta G^{n''-1}}\left(\frac{\delta\Sigma}{\delta G}K\right) = \frac{\delta^{n''}\Sigma}{\delta G^{n''}}K \\ \frac{\delta^{n''-1}K}{\delta G^{n''-1}} &= \frac{\delta^{n''-1}}{\delta G^{n''-1}}(GG\Gamma) \\ \frac{\delta^{n''-1}W}{\delta G^{n''-1}} &= \frac{\delta^{n''-1}}{\delta G^{n''-1}}(v+vKW) \\ \frac{\delta^{n''-1}\Sigma}{\delta G^{n''-1}} &= \frac{\delta^{n''-1}}{\delta G^{n''-1}}(GW\Gamma) \tag*{(33)} \end{aligned}

We note that vv and GG are independent constants therefore δvδG=0\frac{\delta v}{\delta G}=0. This slight variation is similar to the equations in the main text, in particular we have that Hedin approximation n′′→∞n'' \to\infty converges to the exact Hedin equations.

References

  1. [1]I. Shavitt, and R. J. Bartlett, Many-body methods in chemistry and physics: MBPT an coupled-cluster theory (Cambridge University Press, Cambridge, 2009).
  2. [2]G. Stefanucci, and R. van Leeuwen, Nonequilibrium many body theory of quantum systems: a modern introduction (Cambridge University Press, Cambridge, 2025).
  3. [3]F. Aryasetiawan, Elements of Green’s functions and density functional theory (World Scientific, New Jersey, 2025).DOI
  4. [4]S. M. Blinder and J. E. House, Mathematical physics in theoretical chemistry (Elsevier Inc., Amsterdam, 2019).DOI
  5. [5]F. Becca and S. Sorella, Quantum Monte Carlo approaches for correlated systems (Cambridge University Press, Cambridge, 2017).DOI
  6. [6]V. M. Galitskii, and A. B. Migdal, Sov. Phys. JETP 7, 96 (1958).
  7. [7]F. Aryasetiawan and F. Nilsson, Downfolding methods in many-electron theory (AIP Publishing, Melville, 2022).DOI
  8. [8]R. Kubo, J. of the Soc. Phys. of Japan 12, 570 (1957).
  9. [9]L. Hedin, Phys. Rev. 139, A796 (1965).
  10. [10]B. I. Lundqvist, Phys. Kondens. Matt. 6, 193 (1967).
  11. [11]B. I. Lundqvist, Phys. Kondens. Matt. 6, 206 (1967).
  12. [12]F. Aryasetiawan, and S. Biermann, Phys. Rev. Lett. 100, 116402 (2008).
  13. [13]C. Lane, arXiv 2506.07302DOI
  14. [14]F. Aryasetiawan, and S. Biermann, J. of Phys. Cond. Matt. 21, 064232 (2009).
  15. [15]P. E. Trevisanutto, M. Milletarl, Phys. Rev. B 92, 235303 (2015).
  16. [16]V. I. Yukalov, Phys. Rev. E 94, 012106 (2016).
  17. [17]V. I. Yukalov and E. P. Yukalova, Laser Phys. 35, 045501 (2025).
  18. [18]D. Golze, M. Dvorak, and P. Rinke, Front. Chem. 7, 377 (2019).
  19. [19]F. Aryasetiawan, and O. Gunnarsson, arXiv 971203.
  20. [20]A. Marie, A. Ammar and P.-F. Loos, arXiv 2311.05351.
  21. [21]A. L. Kuteppov, Comp. Phys. Comm. 257 107502 (2020).
  22. [22]A. L. Kuteppov, arXiv 1809.06654.
  23. [23]A. L. Kuteppov and G. Kotliar, Phys. Rev. B 96, 035108 (2017).
  24. [24]A. L. Kuteppov, Phys. Rev. 95, 195120 (2017).
  25. [25]R. W. Schindlmayr and R. W. Godby, arXiv 9710295.
  26. [26]L. G. Molinari, Phys. Rev. B 71, 131102 (2005).
  27. [27]Y. Pavlyukh, and W. Hübner, J. Math. Phys. 48, 052109 (2007).
  28. [28]L. G. Molinari, and N. Manini, Eur. Phys. J. B 51, 331 (2006).
  29. [29]E. Fradkin, Quantum field theory: an integrated approach (Princeton University Press, Princeton, 2021).
  30. [30]F. Bruneval, N. Dattani, and M. J. van Setten, Front. in Chem. 9, 749779 (2021).
  31. [31]G. Kotliar, S. Y. Savrasov, K. Haule, V. S. Oudovenko, O. Parcollet, and C. A. Marianetti, Rev. of Mod. Phys. 78, 865 (2006).

Paper details

vlnc:W_286y8ytm
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents