Ab Initio Few-Mode Quantization Scheme

• Ab initio few-mode quantization is a method that reduces a continuum electromagnetic environment to a minimal set of interacting, damped modes derived from first-principles data.
• It employs spectral density fitting and Green tensor computations to capture open-system features like loss, interference, and spatial inhomogeneity in nanophotonic systems.
• The approach integrates QM/MM molecular dynamics with ab initio photonic modeling to accurately simulate light–matter interactions and reproduce experimental spectral responses.

An ab initio few-mode quantization scheme is a reduced open-system formulation in which a continuum electromagnetic environment is replaced by a finite set of discrete modes and residual dissipation, while the parameters of that reduced model are derived from first-principles wave, Green-tensor, or scattering data rather than postulated phenomenologically. In the nanophotonic realization developed for molecular emitters in plasmonic resonators, arbitrary, lossy, and spatially inhomogeneous photonic environments are represented by a minimal set of interacting damped modes fitted to the exact spectral density from Maxwell simulations, and the molecular subsystem is propagated with on-the-fly QM/MM molecular dynamics and excited-state electronic-structure calculations (Tichauer et al., 5 Nov 2025). Closely related formulations establish exact system-bath decompositions for non-interacting scattering problems (Lentrodt et al., 2018), spectral-density fitting for arbitrary electromagnetic environments (Medina et al., 2020), ab initio few-mode cavity descriptions for thin-film x-ray cavity QED (Lentrodt et al., 2020), and QNM-based reduced-mode quantization with ab initio dissipation in broadband and ultrastrong regimes (Gustin et al., 29 Jul 2025).

1. Conceptual foundations and meaning of “ab initio”

The central objective of few-mode quantization is to retain the practical economy of a cavity-mode description without discarding the openness, loss, interference, and spatial structure of the underlying electromagnetic problem. In the general ab initio formulation for quantum potential scattering, the one-particle Hilbert space is decomposed exactly into a discrete “system” sector and a bath sector by Feshbach projectors $P$ and $Q$, yielding an exact few-mode Hamiltonian for non-interacting scattering problems and an exact reconstruction of the physical scattering matrix once background scattering is included (Lentrodt et al., 2018). In thin-film x-ray cavity QED, the same logic appears in a layered open scattering problem: a small set of cavity-like degrees of freedom is retained, the remainder is treated as a continuum bath, and the empty-cavity response is recovered exactly through the product $S=S_{\mathrm{bg}}S_{\mathrm{io}}$ (Lentrodt et al., 2020).

Within electromagnetic open systems, “ab initio” is not uniform across all implementations. In the spectral-density approach, the input is the dyadic Green tensor $\mathbf G$ or the associated spectral density $J(\omega)$, computed from classical electrodynamics for the actual geometry and material response, after which a compact open quantum model is obtained by fitting a few-mode ansatz (Medina et al., 2020). In the QNM approach, the reduced cavity sector is derived from macroscopic QED, the material permittivity $\epsilon(\mathbf r,\omega)$, the Green tensor, and the classical QNMs $\tilde{\mathbf f}_\mu(\mathbf r)$ with complex eigenfrequencies $\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$, so that the cavity-emitter coupling and the cavity-reservoir spectral density are inherited from Maxwell theory rather than inserted by hand (Gustin et al., 29 Jul 2025).

The nanophotonic multiscale framework makes this distinction explicit. On the molecular side, the construction is genuinely ab initio in the standard QM/MM quantum-chemistry sense because potential energies and transition dipoles are recomputed on the fly along atomistic trajectories. On the photonic side, the environment is first-principles-informed and systematically reduced: it is computed from Maxwell’s equations for the actual nanostructure through $\mathbf G$, but the final few-mode representation is a fitted effective model rather than a unique mode-by-mode quantization (Tichauer et al., 5 Nov 2025). A common misconception is therefore that “ab initio few-mode quantization” must mean a unique set of discrete cavity eigenmodes. Across this literature, the term more precisely denotes first-principles input at the field level, followed by a controlled reduced-mode representation.

2. Photonic reduction from Green tensors, spectral densities, and mode networks

In the nanophotonic implementation for multiple emitters, the exact photonic object is the dyadic Green tensor $\mathbf{G}(\mathbf{r}_i,\mathbf{r}_j,\omega)$. For reciprocal media, the relevant spectral density is

$Q$0

This tensorial $Q$1 controls cavity-mediated emitter dynamics and interactions. The few-mode construction replaces the continuum environment by a finite interacting-mode model whose response reproduces this spectral density as accurately as possible. Before the RWA, the effective Hamiltonian is

$Q$2

with losses added by Lindblad damping. The model spectral density is

$Q$3

The real fitting parameters $Q$4, $Q$5, and $Q$6 are optimized nonlinearly so that the model reproduces the full spectral density extracted from the Green tensor (Tichauer et al., 5 Nov 2025).

This construction is not a set of independent cavity resonances. The retained photonic degrees of freedom form an interacting mode network with diagonal frequencies, off-diagonal intermode couplings $Q$7, and mode-dependent losses $Q$8. That point is structurally important because noninteracting lossy modes generate only sums of Lorentzians, whereas interacting modes can reproduce interference-induced non-Lorentzian structures, including Fano-like features (Medina et al., 2020). In the single-emitter formulation, the same logic leads to the master equation

$Q$9

with a model spectral density

$S=S_{\mathrm{bg}}S_{\mathrm{io}}$0

where off-diagonal mode couplings are the mechanism that captures broad backgrounds, overlapping resonances, and asymmetric lineshapes (Medina et al., 2020).

The silver-dimer application illustrates the practical reduction. The Green tensor is computed with SCUFF-EM using a Drude model for silver, and the numerically computed spectral density is fit with a 40-mode representation until a relative fitting error of $S=S_{\mathrm{bg}}S_{\mathrm{io}}$1 is reached (Tichauer et al., 5 Nov 2025). A central finding is that a single-mode reduction fails even when the dipolar resonance is spectrally well isolated and can be fitted by a single Lorentzian, because the remaining fitted modes and their mutual couplings encode the spatially inhomogeneous field structure in the nanogap (Tichauer et al., 5 Nov 2025). This directly opposes the standard intuition that spectral isolation alone guarantees single-mode adequacy.

3. Coupling to realistic matter: QM/MM molecular dynamics and geometry-dependent Hamiltonians

The defining extension in the 2025 nanophotonic framework is the incorporation of ab initio QM/MM molecular dynamics into the few-mode light-matter description. Each chromophore is treated as a geometry-dependent electronic subsystem. At each MD step, the ground-state energy $S=S_{\mathrm{bg}}S_{\mathrm{io}}$2, excited-state energy $S=S_{\mathrm{bg}}S_{\mathrm{io}}$3, excitation gap

$S=S_{\mathrm{bg}}S_{\mathrm{io}}$4

and transition dipole $S=S_{\mathrm{bg}}S_{\mathrm{io}}$5 are computed on the fly (Tichauer et al., 5 Nov 2025).

After the RWA, the working Hamiltonian is

$S=S_{\mathrm{bg}}S_{\mathrm{io}}$6

This compactly combines ab initio molecular energetics and dipoles with a fitted few-mode photonic environment. In the one-excitation sector and with only photonic losses, the dynamics are propagated with the equivalent non-Hermitian Hamiltonian

$S=S_{\mathrm{bg}}S_{\mathrm{io}}$7

since the Lindblad quantum jumps repopulate the ground manifold and do not affect amplitudes within the one-excitation subspace (Tichauer et al., 5 Nov 2025).

The nuclear dynamics are classical and coupled to the quantum electron-photon subsystem in Ehrenfest mean field. The hybrid wavefunction is expanded as

$S=S_{\mathrm{bg}}S_{\mathrm{io}}$8

where the first $S=S_{\mathrm{bg}}S_{\mathrm{io}}$9 basis states are singly excited molecular states and the remaining $\mathbf G$0 states are one-photon occupations of the effective modes. The additional Hellmann-Feynman force on atom $\mathbf G$1 in molecule $\mathbf G$2 contains geometry-dependent ground- and excited-state gradients and a term involving $\mathbf G$3, so the force is weighted by electronic-photonic amplitudes rather than fixed-state occupations (Tichauer et al., 5 Nov 2025).

The implementation is chemically specific. For Rhodamine, the ground state is RHF/3-21G on the three fused rings in the QM region and the excited state is CIS/3-21G; for Methylene Blue, DFT with $\mathbf G$4B97X-D/6-31G* and TD-DFT at the same functional and basis are used; for HBQ, DFT with CAM-B3LYP/6-31G* and TD-DFT with the same functional and basis are used. Rhodamine and Methylene Blue are solvated in water with TIP3P, HBQ in cyclohexane with GROMOS-2016H66, the MD time step is $\mathbf G$5 fs, and temperature is controlled by velocity-rescaling (Tichauer et al., 5 Nov 2025). The resulting matter sector contains geometry-dependent diagonal disorder, dynamic disorder, broadening, and, for HBQ, ultrafast ESIPT, none of which can be represented by fixed identical two-level emitters.

4. Relations to pseudomodes, QNMs, system-bath theory, and x-ray few-mode models

Several distinct ab initio few-mode traditions converge on the same structural problem: how to extract the relevant discrete degrees of freedom from an open continuum without reintroducing phenomenological cavity assumptions. The potential-scattering formalism provides the most explicit exact statement for non-interacting systems: the free/open problem is handled exactly, the few-mode Hamiltonian is exact, and the only approximation in the interacting extension is the neglect of direct matter-bath coupling after the system-bath split (Lentrodt et al., 2018). The thin-film x-ray formulation similarly treats the free field through an exact system-bath partitioning, introduces geometry-derived couplings $\mathbf G$6, $\mathbf G$7, and $\mathbf G$8, and shows that the cavity propagator is generally a matrix with off-diagonal elements rather than a scalar linewidth (Lentrodt et al., 2020).

The spectral-density-fitting scheme is related in spirit to pseudomode and reaction-coordinate approaches because a structured continuum is represented by damped discrete auxiliaries, but operationally it uses fitting of the macroscopic-QED response rather than direct pseudomode decomposition or reaction-coordinate mapping (Medina et al., 2020). The nanophotonic multiscale formulation is explicit on this point: it is not derived via explicit quasinormal-mode normalization, not via a reaction-coordinate mapping from a microscopic system-bath Hamiltonian, and not by direct pseudomode decomposition in the usual open-system sense, even though the language of “pseudomode” appears when discussing a broad spectral feature produced by higher-order nanoparticle modes (Tichauer et al., 5 Nov 2025).

The QNM formulation addresses a different but adjacent issue: how dissipation and gauge consistency constrain reduced-mode cavity QED in broadband and ultrastrong regimes. There the continuum polariton operators are projected onto a discrete QNM subspace plus a residual reservoir, the cavity-reservoir coupling is derived from geometry rather than appended as $\mathbf G$9, and the single-mode dissipative master equation acquires a frequency-dependent spectral density

$J(\omega)$0

which depends on the local complex QNM phase at the emitter position (Gustin et al., 29 Jul 2025). A recurrent misconception in reduced-mode modeling is therefore that photon loss can always be represented by a flat bath with a guessed operator. The QNM analysis shows that both the bath-coupling operator and the spectral function are constrained by Maxwell/QNM physics, especially in broadband dissipative and USC settings (Gustin et al., 29 Jul 2025).

Across these approaches, the phrase “few mode” does not mean that the physical field literally contains only a few modes. It means that a small, judiciously chosen discrete sector captures the structured part of the field relevant in the interaction region while the remainder is retained as a continuum bath or compressed into effective losses (Lentrodt et al., 2020). The retained discrete modes should therefore be interpreted as reduced dynamical degrees of freedom, not automatically as canonical physical cavity eigenmodes.

5. Representative systems and quantitative behavior

The multiscale nanophotonic applications use a silver nanosphere dimer aligned along $J(\omega)$1, with sphere radius $J(\omega)$2 nm and a $J(\omega)$3 nm gap. Five emitters are placed in the gap: one at the center and four peripheral molecules displaced by $J(\omega)$4 nm in a square arrangement in the equatorial plane, all at least $J(\omega)$5 nm from the metal surface (Tichauer et al., 5 Nov 2025). For the bright dipolar mode in the first application, the fitted resonance is

$J(\omega)$6

corresponding to a decay time $J(\omega)$7. When the five emitters are first modeled as ideal identical TLSs with excitation energy $J(\omega)$8 eV and transition dipole $J(\omega)$9 D, the few-mode QM/MM implementation reproduces Lindblad-master-equation population dynamics with error below $\epsilon(\mathbf r,\omega)$0, establishing numerical consistency (Tichauer et al., 5 Nov 2025).

The physically significant behavior emerges when realistic molecular dynamics are included. For Rhodamine in gas phase at $\epsilon(\mathbf r,\omega)$1, $\epsilon(\mathbf r,\omega)$2, $\epsilon(\mathbf r,\omega)$3, $\epsilon(\mathbf r,\omega)$4, and $\epsilon(\mathbf r,\omega)$5 K, and in water at $\epsilon(\mathbf r,\omega)$6 K, coherent population exchange between molecular excitations and the dipolar plasmon remains visible, so strong coupling persists in the presence of molecular motion and disorder for the chosen system (Tichauer et al., 5 Nov 2025). Dynamic disorder lifts the degeneracy of the four peripheral emitters: coherent equality among them is lost after about $\epsilon(\mathbf r,\omega)$7, $\epsilon(\mathbf r,\omega)$8, $\epsilon(\mathbf r,\omega)$9, and $\tilde{\mathbf f}_\mu(\mathbf r)$0 fs at $\tilde{\mathbf f}_\mu(\mathbf r)$1, $\tilde{\mathbf f}_\mu(\mathbf r)$2, $\tilde{\mathbf f}_\mu(\mathbf r)$3, and $\tilde{\mathbf f}_\mu(\mathbf r)$4 K, respectively, and in solution at $\tilde{\mathbf f}_\mu(\mathbf r)$5 K the peripheral populations remain superimposed only for roughly $\tilde{\mathbf f}_\mu(\mathbf r)$6 fs. Averaging over 101 solution-phase trajectories, coherent oscillations persist for about $\tilde{\mathbf f}_\mu(\mathbf r)$7–$\tilde{\mathbf f}_\mu(\mathbf r)$8 fs (Tichauer et al., 5 Nov 2025). The idealized symmetry-protected degeneracies of the corresponding TLS model are therefore not small perturbative corrections; they are qualitatively broken by geometry fluctuations, solvent disorder, and nonidentical local electromagnetic response.

The cavity-mediated energy-transfer calculations further show that the inclusion of nuclear motion and solvent can enhance transfer relative to a fixed-parameter TLS treatment. Starting from excitation localized on the central Rhodamine, the TLS model yields peripheral populations of only about $\tilde{\mathbf f}_\mu(\mathbf r)$9–$\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$0 by $\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$1 fs, decaying to $\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$2, whereas the full QM/MM plus few-mode dynamics yields individual realizations with up to $\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$3 peripheral population and $\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$4 on average at $\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$5–$\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$6 fs, remaining around $\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$7–$\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$8 on average at $\tilde\omega_\mu=\omega_\mu-i\gamma_\mu$9 fs (Tichauer et al., 5 Nov 2025). The mechanism identified in the paper is enhanced spectral overlap between broadened molecular absorption and the dipolar cavity mode.

The HBQ–Methylene Blue donor-acceptor example emphasizes that chemically induced energy shifts can reorganize the relevant photonic channel during the dynamics. Here HBQ is placed at the center and four Methylene Blue molecules at the peripheral sites, while the dimer spectral density is shifted so that the MeB absorption is resonant with the dipolar mode at $\mathbf G$0 (Tichauer et al., 5 Nov 2025). A fixed-energy TLS model would predict negligible transfer because of poor donor-acceptor spectral overlap, but in the full simulation HBQ undergoes ultrafast ESIPT and a large Stokes shift. Around $\mathbf G$1 fs, coincident with proton-transfer completion, the pseudo-mode population near $\mathbf G$2 eV drops while the bright dipolar mode rises by nearly an order of magnitude over $\mathbf G$3 fs; after HBQ shifts into resonance with the dipolar mode, the MeB ensemble becomes strongly coupled and their populations rise coherently with the dipolar mode on another $\mathbf G$4 fs timescale. By $\mathbf G$5 fs, MeB molecules carry about $\mathbf G$6 of the total excitation, with one peripheral MeB reaching up to $\mathbf G$7 (Tichauer et al., 5 Nov 2025).

Earlier validations show the broader relevance of the few-mode program. In a hybrid plasmonic-photonic structure consisting of a GaP microsphere of radius $\mathbf G$8 embedding a silver nanorod dimer with a $\mathbf G$9 gap, an interacting fit with $\mathbf{G}(\mathbf{r}_i,\mathbf{r}_j,\omega)$0 modes gives “almost perfect agreement” with the target spectral density over the full frequency range, while the noninteracting model with the same number of modes fails especially in the background and asymmetric features; the interacting model also reproduces spontaneous-emission population dynamics and electric-field intensity dynamics with very high accuracy (Medina et al., 2020).

6. Validity regime, limitations, and current constraints

The multiscale nanophotonic formulation has a clearly stated validity window. The environment is assumed linear and reciprocal, the fitted few-mode model must reproduce the relevant photonic response over the frequency range sampled by the molecular dynamics, the working Hamiltonian uses the rotating-wave approximation, the molecular subsystem is reduced to few-level systems and in the applications effectively to $\mathbf{G}(\mathbf{r}_i,\mathbf{r}_j,\omega)$1 electronic subspaces, the propagation is restricted to the single-excitation manifold, and the nuclei are treated semiclassically through Ehrenfest dynamics rather than as fully quantum nuclear wave packets (Tichauer et al., 5 Nov 2025). The method does not include explicit free-space molecular radiative decay on long timescales, explicit beyond-dipole terms, or ultrastrong-coupling gauge analysis (Tichauer et al., 5 Nov 2025).

The spectral-density approach has additional structural limits. For a single emitter, the environment is fully encoded by $\mathbf{G}(\mathbf{r}_i,\mathbf{r}_j,\omega)$2, but for multiple emitters one needs the full matrix of self- and cross-spectral densities or, equivalently, the Green-tensor information between emitter positions (Medina et al., 2020). The mapping from $\mathbf{G}(\mathbf{r}_i,\mathbf{r}_j,\omega)$3 to the fitted parameters $\mathbf{G}(\mathbf{r}_i,\mathbf{r}_j,\omega)$4 is not unique, depends on the chosen frequency window, and can become numerically difficult as the number of modes increases (Medina et al., 2020). The x-ray few-mode theory isolates the approximation differently: the free field is partitioned exactly, but direct coupling of nuclei to the residual bath is neglected after the decomposition, and the thin-layer approximation fails when the field varies appreciably across a resonant layer (Lentrodt et al., 2020).

The QNM formulation sharpens the limitations of single-mode reduced descriptions under dissipation. In the one-mode case, positivity of all decay rates requires $\mathbf{G}(\mathbf{r}_i,\mathbf{r}_j,\omega)$5, and the heuristic upper bound

$\mathbf{G}(\mathbf{r}_i,\mathbf{r}_j,\omega)$6

is presented as a rough single-mode breakdown estimate, although full Maxwell simulations deviate from the single-QNM model before that bound is reached (Gustin et al., 29 Jul 2025). The “spatially specified” correction that restores the correct phase-sensitive spectral density has not yet been generalized to several QNMs and several emitters with different local phases (Gustin et al., 29 Jul 2025). This implies that a general ab initio few-mode quantization scheme for open cavity QED is still method-dependent: exact for certain free scattering problems, highly effective and practical for spectral-density fitting, chemically rich in the QM/MM nanophotonic framework, and formally rigorous but not yet fully generalized in the QNM dissipative USC setting.

Taken together, these works define the ab initio few-mode quantization scheme as a family of reduction strategies rather than a single universal construction. Their common core is the replacement of phenomenological cavity models by reduced-mode Hamiltonians whose parameters are inherited from Maxwell theory, scattering theory, or atomistic electronic structure. Their main technical consequence is that openness, disorder, multimode interference, and geometry-dependent matter response can be retained without reverting to a full continuum simulation at every stage.