Purified Pseudofermions in Fermionic Reservoirs

• Purified pseudofermions are ancillary damped fermionic modes that exactly reproduce one-sided reservoir correlators, enabling precise modeling of fermionic environments.
• They employ a one-to-one mapping from rational approximations of correlation functions, separating particle/hole sectors and time directions for enhanced numerical accuracy.
• The approach yields a Lindblad-like effective dynamics compatible with modern solvers, facilitating efficient simulations of quantum transport and interacting systems.

Purified pseudofermions are ancillary damped fermionic modes introduced to model fermionic reservoirs by matching the exact two-time reservoir correlators with a finite set of one-sided exponentials. In the formulation of “Purified pseudofermion approach for the exact description of fermionic reservoirs” (Liang et al., 17 Sep 2025), the Gaussian continuum reservoir is replaced by an effective environment of discrete ancillary modes whose free correlations are highly unusual: each mode contributes only on one time half-axis and only in one sector, particle or hole. This yields an easily implementable and accurate scheme for constructing effective models of fermionic reservoirs, and it is designed to interface directly with modern exponential-decomposition algorithms for reservoir correlation functions.

1. Open-system setting and the reservoir-correlation problem

The construction starts from a generic fermionic open system

$H = H_S + H_R + H_I,$

with

$$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$

where the reservoir is initially in a thermal state at $(\beta,\mu)$. The reduced dynamics depends only on the bath correlators

$$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$

which can be written as

$$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$

These satisfy the time-reversal relation

$C^\sigma(t)=C^{\sigma *}(-t).$

The central technical difficulty is that interacting transport and impurity problems require long-time access to $C^\sigma(t)$, often for structured hybridization functions $J(\epsilon)$ with finite bands, sharp edges, or low-temperature Fermi surfaces. The continuum description is then costly or unusable in many real-time methods. Purified pseudofermions address precisely this bottleneck by re-expressing the reservoir influence in terms of ancillary damped fermions whose free correlators reproduce elementary pieces of $C^\sigma(t)$ (Liang et al., 17 Sep 2025).

2. Definition of purified pseudofermions

A purified pseudofermion is an ancillary fermionic mode with a Lindblad-like generator that is highly asymmetric between left and right actions on the density matrix, an initial state that is either fully occupied or empty, and free correlations with support only either for $t>0$ or $$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$0, and only either in the particle correlator $$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$1 or the hole correlator $$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$2. For each type $$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$3, the free correlation matrix

$$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$4

has only one nonzero entry.

This sharply distinguishes purified pseudofermions from ordinary auxiliary fermions. Ordinary damped fermions have correlations of the form $$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$5 and simultaneously encode particle and hole sectors, with occupations fixed by a thermal parameter. Earlier pseudofermion constructions allow complex parameters but remain symmetric in time. Purified pseudofermions instead arise from an extreme analytical continuation and adiabatic elimination procedure, and each mode represents exactly one clean piece of reservoir influence: one time direction, one sector, one exponential.

The four basic types are summarized by their single nonzero correlators and initial states.

Type	Nonzero free correlator	Initial state
I	$$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$6	$$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$7
II	$$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$8	$$H_R = \sum_k \epsilon_k\, c_k^\dagger c_k,\qquad H_I = \sum_k (g_k s\, c_k^\dagger + g_k^* c_k s^\dagger),$$9
III	$(\beta,\mu)$0	$(\beta,\mu)$1
IV	$(\beta,\mu)$2	$(\beta,\mu)$3

These correlators are non-thermal and one-sided. Their significance is structural rather than merely formal: because each ancillary mode contributes a single directional exponential, the mapping from a fitted correlation-function term to a mode is one-to-one (Liang et al., 17 Sep 2025).

3. Correlator decomposition and one-to-one mode assignment

The reservoir correlators are decomposed separately on the positive and negative time half-axes,

$(\beta,\mu)$4

$(\beta,\mu)$5

using rational approximation algorithms such as AAA and Prony, applied to spectral functions $(\beta,\mu)$6. Each exponential term is then implemented by one purified pseudofermion whose free correlation is exactly that term.

The correspondence is fixed by time direction and sector:

• Positive-time particle terms map to type II.
• Positive-time hole terms map to type III.
• Negative-time particle terms map to type I.
• Negative-time hole terms map to type IV.

Equivalently, for a decomposition

$(\beta,\mu)$7

on $(\beta,\mu)$8, one sets $(\beta,\mu)$9, takes $$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$0 and $$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$1 from the fit, and chooses the pseudofermion type according to the sign of $$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$2 and the correlator sector. The full reservoir correlator is then represented as

$$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$3

This is the fermionic analogue of bosonic pseudomode constructions, but with the particle/hole structure and the half-axis decomposition built into the elementary modes. A plausible implication is that the scheme removes much of the redundancy present in auxiliary constructions where each mode must represent both time directions simultaneously (Liang et al., 17 Sep 2025).

4. Effective Lindblad-like dynamics and derivation

The resulting effective model consists of the system Hamiltonian $$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$4, ancillary modes $$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$5, and a Lindblad-like master equation for the composite density matrix,

$$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$6

with type-dependent generators $$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$7. For example, type I obeys

$$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$8

$$C^\sigma(t) = \langle B^\sigma(t) B^{\bar\sigma}(0)\rangle,\qquad B^{+}=\sum_k g_k c_k,\quad B^{-}=\sum_k g_k c_k^\dagger,$$9

with $$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$0.

In the full model, positive-time particle modes and negative-time particle modes enter with different left/right action structures, and analogous formulas hold for hole modes. The key point is that this is not a standard Lindblad form with jump operators acting only on the right. The unusual left/right action is precisely what encodes the time ordering and fermionic parity structure of the influence functional.

The purified pseudofermion generators are obtained by starting from a conventional pseudofermion master equation with Markovian damping rate $$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$1 and occupation number $$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$2, performing the analytic continuation

$$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$3

and then carrying out adiabatic elimination in a doubled superfermion space by a Schrieffer–Wolff transformation. This leaves an effective generator acting only through the surviving sector and produces the one-sided exponential correlators exactly. The derivation matters because it explains why the generators are nonstandard while still remaining directly usable in many-body solvers (Liang et al., 17 Sep 2025).

5. Exactness, assumptions, and convergence

For noninteracting reservoirs with given $$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$4, $$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$5, and $$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$6, the correlators $$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$7 are known exactly. If the exponential decomposition is taken with $$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$8, or with enough terms for convergence,

$$C^\sigma(t) = \int d\epsilon\, \frac{J(\epsilon)}{\pi}\, e^{i\sigma\epsilon t} \left[\frac{1-\sigma}{2} + \sigma f(\epsilon)\right].$$9

then the purified pseudofermion representation is exact at the level of two-time correlations. Because the reservoir is Gaussian, exact correlation functions imply exact reduced dynamics and currents.

The assumptions are explicit: reservoirs are noninteracting and Gaussian, the initial state is thermal and factorized,

$C^\sigma(t)=C^{\sigma *}(-t).$0

and the reservoir is time stationary. There is no need for wide-band or Markovian approximations; finite bandwidth, sharp edges, and low temperatures are included directly through $C^\sigma(t)=C^{\sigma *}(-t).$1 and $C^\sigma(t)=C^{\sigma *}(-t).$2.

In practice one truncates the decomposition at finite $C^\sigma(t)=C^{\sigma *}(-t).$3. The error is then governed by the spectral approximation algorithm rather than by the pseudofermion mapping itself. The AAA fit is systematically improvable by increasing the number of poles, and typical errors can be made extremely small, including for discontinuous spectra such as zero-temperature Fermi steps and sharp band edges. This suggests that the principal numerical question is the quality of the half-axis rational approximation, not the validity of the ancillary-mode representation (Liang et al., 17 Sep 2025).

6. Numerical implementation, efficiency, and demonstrated performance

The effective dynamics is a Lindblad-like master equation for a finite system consisting of the physical degrees of freedom plus ancillary pseudofermion modes. This is naturally compatible with tensor-network methods, superfermion diagonalization for quadratic systems, and other open-system solvers that handle Lindblad or non-Hermitian generators. In the reported calculations, time-dependent DMRG with swap gates is used to manage long-range couplings between system edges and pseudofermion modes.

The claimed efficiency rests on several points. The correspondence between decomposition terms and ancillary modes is direct, so no stochastic parameter search is needed. AAA works directly on the spectral functions $C^\sigma(t)=C^{\sigma *}(-t).$4. Each purified pseudofermion is effectively a 2-dimensional fermionic site in superfermion language. Because each mode covers only one sector and one half-axis, fewer modes can reach a given accuracy than in schemes where each auxiliary mode must encode both time directions.

The numerical demonstrations are concrete. For a single impurity, currents from the purified pseudofermion model match the Landauer–Büttiker formula with high precision across a wide range of parameters, including zero temperature and sharp hybridization. For a three-site thermal engine, about $C^\sigma(t)=C^{\sigma *}(-t).$5–$C^\sigma(t)=C^{\sigma *}(-t).$6 purified pseudofermions are sufficient to achieve $C^\sigma(t)=C^{\sigma *}(-t).$7 relative error compared to Landauer–Büttiker predictions over $C^\sigma(t)=C^{\sigma *}(-t).$8, and the same decomposition is reused in the interacting case with nearest-neighbour $C^\sigma(t)=C^{\sigma *}(-t).$9. For an interacting chain of spinless fermions, chains up to $C^\sigma(t)$0 are simulated; the method captures ballistic transport, meaning current independent of length for large $C^\sigma(t)$1, and steady-state density profiles. The comparison to mesoscopic leads shows that the purified pseudofermion scheme needs fewer damped fermions for comparable or better accuracy, especially near band edges and at low temperatures (Liang et al., 17 Sep 2025).

7. Relation to adjacent literatures and scope of the term

In this usage, purified pseudofermions are not a generic synonym for pseudofermions. The term refers specifically to ancillary damped fermionic modes for the exact description of fermionic reservoirs. That usage is distinct from pseudo-fermions in pseudo-Hermitian finite-dimensional algebras and their nonlinear generalizations (Bagarello, 2017, Cherbal et al., 2016, Trifonov, 2012), from pseudofermions used as stochastic determinant estimators in lattice field theory (Abbott et al., 2022, Xiong et al., 30 Mar 2026), and from Abrikosov pseudofermions in functional renormalization-group treatments of spin models (Müller et al., 2023). It is also unrelated to the purity-controlled bosonic behavior of entangled fermion pairs discussed in composite-boson theory (Tichy et al., 2012).

Methodologically, the closest relatives are hierarchical equations of motion and influence-functional approaches, which also decompose bath correlators, as well as bosonic pseudomode and reaction-coordinate constructions. The distinctive feature here is that particle/hole sectors and positive/negative time contributions are separated at the mode level. This gives a fermionic pseudomode construction whose elementary constituents already match the output of modern half-axis decomposition algorithms.

The applications emphasized for the method include quantum transport through nanosystems, quantum thermodynamics and thermal engines, nonequilibrium phase transitions, quantum impurity problems with structured baths, and extended many-body systems coupled to leads. The stated outlook includes further optimization of the decomposition algorithms, extension to multi-lead and multi-impurity settings, generalization to more complicated reservoir initial states or driven reservoirs, and integration with advanced tensor-network and quantum-computing methods. A plausible implication is that purified pseudofermions are most valuable where reservoir accuracy, long-time real-time evolution, and extended interacting systems must be treated simultaneously without reverting to wide-band or Markovian simplifications (Liang et al., 17 Sep 2025).