A Dyson equation for non-equilibrium Green's functions in the partition-free setting
aa r X i v : . [ m a t h - ph ] J a n A Dyson equation for non-equilibrium Green’s functions in thepartition-free setting
H.D. Cornean , V. Moldoveanu , C.-A. Pillet Department of Mathematical SciencesAalborg UniversityFredrik Bajers Vej 7G, 9220 Aalborg, Denmark National Institute of Materials PhysicsP.O. Box MG-7 Bucharest-Magurele, Romania Aix Marseille Univ, Universit´e de Toulon, CNRS, CPT, Marseille, FranceJanuary 24, 2019
Abstract.
We consider a small interacting sample coupled to several non-interacting leads.Initially, the system is at thermal equilibrium. At some instant t the system is set into the socalled partition-free transport scenario by turning on a bias on the leads. Using the theory ofVolterra operators we rigorously formulate a Dyson equation for the retarded Green’s functionand we establish a closed formula for the associated proper interaction self-energy. The backbone of many-body perturbation theory (MBPT) is the interaction self-energy Σ which appearsin the Dyson equation for equilibrium or non-equilibrium Green’s function (NEGF). At equilibrium, thestructure of Σ is guessed by systematically using Wick’s theorem and by analysing the resulting expansioninto Feynman diagrams [1]. Approximation schemes (e.g. mean-field approach or RPA) correspond topartial resummation of series of diagrams contributing to Σ.In the finite temperature non-equilibrium regime of interacting systems the initial state cannot be ingeneral connected to a non-interacting state in the remote past [2] and writing down statistical averagesof time-dependent observables becomes cumbersome. The remedy for these technical difficulties is tocombine the chronological T and anti-chronological T time-ordering operators into a single operator T C which allows an unambiguous book-keeping of time arguments on the two-branch Schwinger-Keldyshcontour [3, 4, 5]. This construction comes with a price: the non-equilibrium GFs turn to contour-ordereduantities as well and the various identities among them are not easy to recover. At a formal level oneassumes the existence of a well-defined self-energy and then the contour-ordered Dyson equation splitsvia the Langreth rules [6] into the Keldysh equation for the lesser/greater GFs and the Dyson equationfor the retarded/advanced GFs (see the textbook [7]).The existence of a self-energy for the contour-ordered GF is argued by the formal analogy betweenequilibrium and non-equilibrium quantum averages. Then a complete interaction self-energy can bedefined [8]. In more recent formulations [9] one starts from the differential equations of motions relatinghigher n -particle Green-Keldysh functions and then truncates the so-called Martin-Schwinger hierarchy[10] to identify various approximate interaction self-energies.Nowadays, the NEGFs formalism has grown up as a remarkable machinery, being extensively used formodelling quantum transport in mesoscopic systems [11], molecules [12] or even nuclear reactions [13].Nonetheless, some fundamental theoretical questions were only recently answered by fully exploiting themathematical structure of the theory and without making any approximations. We refer here to: (i) theexistence of non-equilibrium steady-state (NESS) in interacting open systems and (ii) the independence ofthe steady-state quantities from the initial state of the sample [14, 15, 16, 17, 18] both in the partitioning[19] and partition free [20, 21] settings. We recall here that in the partitioned case the system and thebiased leads are initially decoupled.In our recent work [22] the NEGF formalism for open systems in the partitioning transport settingwas rigorously treated in great detail and generality. In particular, we derived the Jauho-Wingreen-Meirformula (JWM) [23] for the time-dependent current through an interacting sample by using only real-timequantities.In this short note we are interested in the partition-free regime which was adapted for interacting systemsby Stefanucci and Almbladh [24]. Recently, the long-time limit of the energy current in the partition-freesetting was discussed in Ref.[25] and the transient heat currents due to a temperature gradient werecalculated in [26]. We briefly outline a rigorous formulation of the non-equilibrium Dyson equation forthe retarded Green’s function. Mathematical details are kept to a minimum while focusing on the explicitconstruction of a complete interaction self-energy.The content of the paper goes as follows: the model and the notations are introduced in Section 2, themain result and its proof are given in Section 3 while Section 4 is left for conclusions. We assume that a small sample is coupled to M leads. The one-particle Hilbert space is of tight-bindingtype and can be written as h = h S ⊕ h R where h S is finite dimensional and h R = ⊕ Mν =1 h ν describesthe (finite or not) leads. Particles can only interact in the sample. One-particle operators are denotedwith lower-case letters and their second quantized versions will be labeled by capital letters. The one-particle Hamiltonian of the decoupled system acquires a block-diagonal structure h D = h S ⊕ h R where h R = ⊕ Mν =1 h ν is supposed to be bounded. The lead-sample tunnelling Hamiltonian is defined as: h T = M X ν =1 d ν (cid:0) | f ν ih g ν | + | g ν ih f ν | (cid:1) , (1)where ν counts the particle reservoirs, f ν ∈ h ν and g ν ∈ h S are unit vectors and d ν ∈ R are couplingconstants. The one-particle Hamiltonian of the fully coupled system is then h = h D + h T .2e summarize below some useful identities from the second quantization machinery (see e.g. [27]). Thetotal Fock space admits a factorization F = F S ⊗ F R . By a ( f ) we mean either the creation operator a ∗ ( f ) or the annihilation operator a ( f ). We have a ∗ ( λf ) = λa ∗ ( f ) and a ( λf ) = λa ( f ). The general formof the canonical anticommutation relations is: { a ( f ) , a ∗ ( g ) } = h f | g i , { a ( f ) , a ( g ) } = 0 . (2)Here h f | g i denotes the scalar product in h . Also, a ( f ) is bounded on the Fock space and k a ( f ) k ≤ k f k .The interacting, coupled system, and with a potential bias v ν on lead ν is described by: K v := H + M X ν =1 v ν N ν + ξW, (3)where N ν is the particle number operator on lead ν (i.e., the second quantization of the orthogonalprojection onto h ν ), v := ( v , . . . , v M ) ∈ R M is the bias vector and W = 12 X x,y ∈S w ( x, y ) a ∗ ( | x i ) a ∗ ( | y i ) a ( | y i ) a ( | x i )is the second quantization of a two-body potential satisfying w ( x, y ) = w ( y, x ) and w ( x, x ) = 0 for all x, y ∈ S . Here ξ ∈ R stands for the interaction strength.Assume that the bias is turned on at time t = 0. Then the Heisenberg evolution of an observable A at t > τ tK v ( A ) := e i tK v A e − i tK v , t > . (4)If h is a single-particle Hamiltonian, the associated Heisenberg evolution obeys: τ tH ( a ( f )) := e i tH a ( f )e − i tH = a (e i th f ) , (5)and one has [ H, a ∗ ( f )] = a ∗ ( hf ) , [ H, a ( f )] = − a ( hf ) . (6)Along the proof of the Dyson equation we shall encounter the operators: b ( f ) := i ξ [ W, a ( f )] , b ∗ ( f ) := i ξ [ W, a ∗ ( f )] . (7)These operators vanish if f is supported in the leads. The initial state in the partition-free case is a Gibbs state characterized by the inverse temperature β > µ ∈ R . It is given by the thermodynamic (i.e., infinite leads) limit of thedensity operator ρ pf = Z − e − β ( K − µN ) where Z = Tr F e − β ( K − µN ) . In what follows we briefly explainhow it is constructed.The interacting but decoupled and unbiased Hamiltonian is denoted by: K D := H S + ξW + H R = K − H T . ρ D = Z − D e − β ( K D − µN ) where Z D = Tr F e − β ( K D − µN ) is a tensor productbetween a many-body Gibbs state ρ S = 1Tr F S e − β ( H S + ξW − µN S ) e − β ( H S + ξW − µN S ) only acting on the finite dimensional Fock space F S , and M non-interacting ( β, µ ) Fermi-Dirac quasi-freestates acting on each lead separately, where expectations can be computed with the usual Wick theorem.This special factorized initial state is denoted by (cid:10) · (cid:11) β,µ . For example, the expectation of a factorizedobservable of the type O = O S Q Mν =1 a ∗ ( ˜ f ν ) a ( f ν ) where ˜ f ν , f ν ∈ h ν is: (cid:10) O (cid:11) β,µ = Tr F S ( ρ S O S ) M Y ν =1 h f ν | (Id + e β ( h ν − µ ) ) − ˜ f ν i . Its connection with the partition-free state is as follows. Consider the operator B ( α ) := e − iαK D H T e iαK D , α ∈ R . From (1) and using (5) we see that a generic term entering B ( α ) is X ν d ν a ∗ (e − i αh ν f ν ) τ − αH S + ξW (cid:0) a ( g ν ) (cid:1) . Since h ν is bounded and F S is finite dimensional, this expression remains bounded for all complex valuesof α . Then, the initial value problemΓ ′ ( x ) = B ( ix )Γ( x ) , Γ(0) = Id , has a unique solution given by a norm convergent Picard/Dyson/Duhamel iteration, with terms containingproducts of operators either living in the sample or in the leads. Before the thermodynamic limit, theoperators Γ( β ) and e βK D e − βK satisfy the same differential equation and obey the same initial conditionat β = 0, hence they must coincide. Consequently, writing e − βK = e − βK D Γ( β ) we obtain an appropriateexpression for the thermodynamic limit: O being an arbitrary bounded physical observable, we have (cid:10) O (cid:11) pf = (cid:10) Γ( β ) O (cid:11) β,µ (cid:10) Γ( β ) (cid:11) β,µ . (8) Let 0 < T < ∞ be fixed and let C ([0 , T ]; h ) be the space consisting of time dependent vectors φ ( t ) ∈ h ,0 ≤ t ≤ T , which are continuously differentiable with respect to t , and φ (0) = 0. We also define C ([0 , T ]; h ) to be the space of vectors which are only continuous in t , with no additional condition at t = 0. We note that C ([0 , T ]; h ) is a Banach space if we introduce the norm ||| ψ ||| := sup ≤ t ≤ T k ψ ( t ) k h . We say that an operator A which maps C ([0 , T ]; h ) into itself is a Volterra operator if there exists aconstant C A < ∞ such that k ( Aψ )( t ) k h ≤ C A Z t k ψ ( t ′ ) k h dt ′ , ≤ t ≤ T.
4y induction one can prove: k ( A n ψ )( t ) k h ≤ C A ( C A T ) n − ( n − Z t k ψ ( t ′ ) k h dt ′ , n ≥ . This implies: ||| A n ψ ||| ≤ ( C A T ) n ( n − ||| ψ ||| which leads to the conclusion that the operator norm of A n is bounded by ( C A T ) n ( n − . In particular, theseries P n ≥ ( − n A n converges in operator norm and defines a Volterra operator with a constant lessthan C A e T C A . Thus, (Id + A ) − = Id + P n ≥ ( − n A n always exists and A (Id + A ) − is a Volterraoperator. Let { e j } be an arbitrary orthonormal basis in h . Define the map G : C ([0 , T ]; h ) C ([0 , T ]; h ) givenby: h e j | ( G ψ )( t ) i := − i Z t h e j | e − i( t − t ′ ) h v ψ ( t ′ ) i dt ′ , (9)where h v denotes the single-particle Hamiltonian of the non-interacting coupled and biased system. Onecan check that G is invertible and if φ ∈ C ([0 , T ]; h ):( G − φ )( t ) = i ∂ t φ ( t ) − h v φ ( t ) ∈ C ([0 , T ]; h ) . (10)By definition, the retarded non-equilibrium Green operator in the partition-free setting G ξ : C ([0 , T ]; h ) C ([0 , T ]; h ) is given by: h e j | ( G ξ ψ )( t ) i := − i Z t (cid:10) { τ t ′ K v ( a ∗ ( ψ ( t ′ ))) , τ tK v ( a ( e j )) } (cid:11) pf dt ′ . (11)Using (5) and (2) we see that G ξ coincides with G when ξ = 0. One can show that k ( G ξ ψ )( t ) k h ≤ Z t k ψ ( t ′ ) k h dt ′ , (12)so that G ξ is a Volterra operator. The integral kernel of G ξ is nothing but the more familiar retardedNEGF given by: G Rξ ( e j , t ; e m , t ′ ) := − i θ ( t − t ′ ) (cid:10) { τ t ′ K v ( a ∗ ( e m )) , τ tK v ( a ( e j )) } (cid:11) pf , (13)and h e j | ( G ξ ψ )( t ) i = X m Z t G Rξ ( e j , t ; e m , t ′ ) h e m | ψ ( t ′ ) i dt ′ . (14)The advanced NEGF can be defined as: G Aξ ( e j , t ; e m , t ′ ) := − G Rξ ( e j , t ′ ; e m , t ) = i θ ( t ′ − t ) (cid:10) { τ t ′ K v ( a ∗ ( e m )) , τ tK v ( a ( e j )) } (cid:11) pf . All properties of the advanced NEGF can be immediately read off from those of the retarded one.5
Irreducible self-energy and Dyson equation.
Here is the main result of our paper.
Theorem 3.1.
The bounded linear map e Σ ξ defined on C ([0 , T ]; h ) by h e j | ( e Σ ξ φ )( t ) i := − i Z t (cid:10) { τ t ′ K v ( b ∗ ( φ ( t ′ ))) , τ tK v ( b ( e j )) } (cid:11) pf dt ′ + i (cid:10) τ tK v ( { a ∗ ( φ ( t )) , b ( e j ) } ) (cid:11) pf , (15) obeys: G ξ = G + G e Σ ξ G . (16) Moreover, the operator G e Σ ξ is a Volterra operator, the inverse (Id + G e Σ ξ ) − exists, and by defining Σ ξ := e Σ ξ (cid:16) Id + G e Σ ξ (cid:17) − (17) we have: G ξ = G + G Σ ξ G ξ . (18) Finally, G Σ ξ is also a Volterra operator and G ξ = (Id − G Σ ξ ) − G . (19) As in the physical literature Eq.(17) defines the irreducible self-energy operator Σ ξ in terms of the reduciblepart e Σ ξ . First we will show that the identity: G − G ξ = Id + F ξ (20)holds on C ([0 , T ]; h ), where the map F ξ is given by h e j | ( F ξ ψ )( t ) i := Z t (cid:10) { τ t ′ K v ( a ∗ ( ψ ( t ′ ))) , τ tK v ( b ( e j )) } (cid:11) pf dt ′ . (21)Using (10) and (11) we have: h e m | ( G − G ξ ψ )( t ) i = h e m | ψ ( t ) i − X j h e m | h v e j ih e j | ( G ξ ψ )( t ) i + Z t (cid:10) { τ t ′ K v ( a ∗ ( ψ ( t ′ ))) , ∂ t τ tK v ( a ( e m )) } (cid:11) pf dt ′ . (22)From the antilinearity of the annihilation operators we get X j h e m | h v e j ih e j | ( G ξ ψ )( t ) i = − i Z t (cid:10) { τ t ′ K v ( a ∗ ( ψ ( t ′ ))) , τ tK v ( a ( h v e m )) } (cid:11) pf dt ′ . Also, using (4), (6) and (7) we obtain the identity: ∂ t τ tK v ( a ( e m )) = − iτ tK v ( a ( h v e m )) + τ tK v ( b ( e m )) . After introducing the last two identities into (22) we see that two terms cancel each other and we obtain(21). 6 .2 Proof: step 2.
The second step consists of showing that F ξ can be written as e Σ ξ G , with e Σ ξ as in (15). In order toidentify e Σ ξ we compute for every φ ∈ C ([0 , T ]; h ) the quantity (remember that a ∗ is linear): h e j | ( F ξ G − φ )( t ) i = i Z t (cid:10) { τ t ′ K v ( a ∗ ( ∂ t ′ φ ( t ′ ))) , τ tK v ( b ( e j )) } (cid:11) pf dt ′ − Z t (cid:10) { τ t ′ K v ( a ∗ ( h v φ ( t ′ ))) , τ tK v ( b ( e j )) } (cid:11) pf dt ′ . (23)Another key identity is: τ t ′ K v ( a ∗ ( ∂ t ′ φ ( t ′ ))) = ∂ t ′ (cid:16) τ t ′ K v ( a ∗ ( φ ( t ′ ))) (cid:17) − i τ t ′ K v ( a ∗ ( h v φ ( t ′ ))) − τ t ′ K v ( b ∗ ( h v φ ( t ′ ))) . Inserting this identity in (23), integrating by parts with respect to t ′ and using that φ (0) = 0, we obtain(15). From the first two steps we derive (16). From (15) and (12) we see that A = G e Σ ξ is a Volterra operatorfor which there exists a T -dependent constant C < ∞ such that k ( Aψ )( t ) k h ≤ C Z t k ψ ( t ′ ) k h dt ′ , ≤ t ≤ T. (24)Then (Id + A ) − exists and it is given by a norm convergent Neumann series P n ≥ ( − n A n , as long as T < ∞ . We write G = (cid:16) Id + G e Σ ξ (cid:17) − G ξ and we can choose Σ ξ as in (17), which finishes the construction of the proper self-energy. We list a few remarks concerning our main theorem.(i) The integral kernel of e Σ ξ (see (15)) is given by e Σ Rξ ( e j , t ; e m , t ′ ) := − i θ ( t − t ′ ) (cid:10) { τ t ′ K v ( b ∗ ( e m )) , τ tK v ( b ( e j )) } (cid:11) pf + i δ ( t − t ′ ) (cid:10) τ tK v ( { a ∗ ( e m ) , b ( e j ) } ) (cid:11) pf . If either e j or e m belongs to the leads, then the above matrix element equals zero. The explanation forthe first term is that at least one of the two operators b ( e j ) and b ∗ ( e m ) defined through (7) would be zeroin this case, because the self-interaction W is only supported in the sample, hence it commutes with anyobservable supported on the leads. For the second term, assume that e j is from the sample while e m isfrom the leads. Then since b ( e j ) is a sum of products of three creation/annihilation operators from thesample, it anticommutes with a ∗ ( e m ).The proper self-energy Σ ξ has the same support property. One recognizes that e Σ Rξ ( e j , t ; e m , t ′ ) is areducible self-energy . In the diagrammatic language all terms contributing to e Σ Rξ ( e j , t ; e m , t ′ ) connectto other diagrams by incoming and outgoing G -lines.7ii) If both e j = x and e m = y are located in the small sample, then from (18) we see that in order tocompute G Rξ ( x, t ; y, t ′ ) we only need to know the values of G restricted to the small sample (besides Σ ξ ,of course). From (9) we have: G R ( x, t ; y, t ′ ) = − i θ ( t − t ′ ) h x | e − i( t − t ′ ) h v y i , with x, y ∈ S . Such matrix elements can be computed from the resolvent ( h v − z ) − restricted to thesmall sample; we note that via the Feshbach formula, the biased leads appear as a non-local “dressing”potential which perturbs h S , see [17] for details.At the level of integral kernels, the Dyson equation (18) reads as: G Rξ ( x, t ; y, t ′ ) = G R ( x, t ; y, t ′ ) + X u,v ∈S Z t ds Z s ds ′ G R ( x, t ; u, s )Σ Rξ ( u, s ; v, s ′ ) G Rξ ( v, s ′ ; y, t ′ ) . (iii) Assume that we can write Σ ξ as Σ app + Σ ′ , where Σ app is an approximating Volterra operator. If G app = (Id − G Σ app ) − G is the solution of the approximate Dyson equation G app = G + G Σ app G app ,then we have: G ξ = G app + G app Σ ′ G ξ and G ξ = (Id − G app Σ ′ ) − G app .(iv) The limit T → ∞ is a difficult problem. To the best of our knowledge, the only rigorous mathematicalresults concerning the existence of a steady-state regime in partition free-systems are [17, 18]. Undercertain non-resonant conditions and for ξ small enough, one can prove that a quantity like G Rξ ( e m , t ′ + s ; e n , t ′ ), where s > t ′ → ∞ . This is definitely not guaranteed to happenin all cases, not even in non-interacting systems, due to bound states which may produce persistentoscillations.(v) One may generalize the present setting in order to allow a non-trivial time dependence of the bias, theonly difference would appear in the evolution groups which now would have time-dependent generators.Also, the notation and formulas would be more involved, but no new mathematical issues would appear. We presented a non-perturbative approach to the partition-free transport problem. Starting from theVolterra operator associated to the retarded Green’s function we establish its Dyson equation, and wederive closed formulas for the reducible and irreducible self-energies. The proof is rigorous yet elementaryin the sense that although the partition-free scenario is a genuine non-equilibrium regime we do not usecontour-ordered operators. A Keldysh equation for the lesser Green’s function should be establishedfollowing the same lines of reasoning, with the extra difficulty induced by the fact that in the partitionfree setting, the small sample is not empty at t = 0.Unravelling the connection between the closed formula (15) and the diagrammatic approach remainsan open problem. Although the anti-commutator structure (cid:10) τ tK v ( { a ∗ ( φ ( t )) , b ( e j ) } ) (cid:11) pf in Eq. (15) looksless familiar one can speculate that the systematic application of the Wick theorem should eventuallyrecover various classes of diagrams. A possible approximation in the self-energy would be to replace theinteracting propagator τ tK v ( · ) with the non-interacting one τ tH v ( · ), where K v = H v + ξW . Note however8hat the application of the Wick theorem is technically challenging due to the extra term Γ( β ) appearingin (8).Given the fact that the partition-free setting is less studied in the literature, yet more intuitive on physicalgrounds than the partitioned approach, we hope that our investigation will trigger more efforts from boththe physical and mathematical-physics communities. Our main message is that one can properly formulatesome of the central equations of the many-body perturbation theory (MBPT) in a direct way, payingclose attention to fundamental issues like convergence, existence, uniqueness, stability, and at the sametime, trying to obtain precise error bounds for a given approximation of the self-energy. The Volterratheory guarantees that for relatively small T ’s one can ”keep doing what one has been doing”; however,the large time behavior like for example the existence of steady states and the speed of convergence seemto be very much dependent on the system and no general recipe can work out in all cases. Acknowledgments.
V.M. acknowledges financial support by the CNCS-UEFISCDI Grant PN-III-P4-ID-PCE-2016-0084 and from the Romanian Core Research Programme PN16-480101. H.C. acknowledgesfinancial support by Grant 4181-00042 of the Danish Council for Independent Research | Natural Sciences.C.A.P. acknowledges financial support by the ANR, Grant NONSTOPS (ANR-17-CE40-0006),
References [1] A.L. Fetter and J.D. Walecka,
Quantum Theory of Many-Particle Systems.
Dover Publications, NewYork, 2003.[2] Such a connection would rely on the so called adiabatic assumption which however must be checkedcase by case (see Chapter 4.3 in Ref.[9]).[3] J. Schwinger, J. Math. Phys. , 407 (1961).[4] L.V. Keldysh, Zh. Eksp. Teor. Fiz. , 1515 (1964). English translation in Sov. Phys. JETP ,1018 (1965).[5] P. Danelewicz, Ann. Phys. , 239 (1984).[6] D.C. Langreth, in Linear and Nonlinear Electron Transport in Solids.
J.T. Devreese and V.E. vanDoren editors, NATO Advanced Study Institute, Series B: Physics . Plenum Press, New York,1976.[7] H.J.W. Haug and A.-P. Jauho, Quantum Kinetics in Transport and Optics of Semiconductors.
Springer Series in Solid State Sciences , 2nd Edition. Springer, Berlin, 2007.[8] See for example Eq. (3.16) in Ref.[5] or Eq. (9.10) in Ref. [9].[9] G. Stefanucci and R. van Leeuwen,
Nonequilibrium Many-Body Theory of Quantum System, a Mod-ern Introduction.
Cambridge University Press, Cambridge, 2013.[10] M. Scharnke, N. Schl¨unzen and M. Bonitz, J. Math. Phys. , 061903 (2017).[11] P. My¨oh¨anen, A. Stan, G. Stefanucci, and R. van Leeuwen, Phys. Rev. B , 115107 (2009).[12] N.E. Dahlen and R. van Leeuwen, Phys. Rev. Lett. , 153004 (2007).[13] A. Rios, B. Barker, M. Buchler and P. Danielewicz, Annals of Physics , 1274 (2011).[14] J. Fr¨ohlich, M. Merkli and D. Ueltschi, Ann. Henri Poincar´e , 897 (2004).915] W. Aschbacher, V. Jakˇsi´c, Y. Pautrat and C.-A. Pillet, in Open Quantum Systems III. RecentDevelopments.