betaFeedback

WorksPhysics and Astronomy

Ab Initio Random Matrix Theory of Molecular Electronic Structure

Abstract

We use ab initio electronic-structure methods to investigate random-matrix theory (RMT) universality in molecular electronic structure. Using single-reference electronic structure methods, including Hartree-Fock, configuration-interaction singles (CIS), density functional theory, and linear-response time-dependent density-functional theory, we compute single-particle orbital energies and many-electron excitations of several representative molecules (benzene, alanine, 1-phenylethylamine, methyloxirane, and helicene chains). For generic low-symmetry geometries, the unfolded spectra of these ab initio Hamiltonians exhibit Wigner-Dyson level statistics of the Gaussian orthogonal ensemble (GOE). For extended helicene chains we explicitly restrict to bound valence excitations below the ionization threshold and still observe GOE statistics, indicating that the RMT universality is present for physical states of direct relevance to real molecules. We further explore the electric and magnetic field dependence of the molecular electronic spectra. The variance of electric polarizability (level curvature K) is predicted to be non-analytic in the magnetic field which serves as an infrared cutoff, <K^2> proportional to log(1/|B|). We observe a transition to the Gaussian unitary ensemble (GUE) by increasing the magnetic fields, although it occurs only at magnetic fields far beyond experimentally accessible scales. Our results indicate that random matrix universality provides a general framework for organizing ab initio predictions of interacting electron spectra in complex systems.

Electronic structure of molecules and materials is governed by the many-body Schrödinger equation, which in principle fixes their structure, spectroscopy, and dynamics. In practice, electron motion in all but the simplest molecules is expected to be chaotic (1). Indeed, already three-body classical Coulombic problem is non-integrable (2) and it remains such for more complex (n>3)(n > 3)-body problems, which includes the classical limit of just about all molecules of practical interest. Per the Bohigas-Giannoni-Schmit (BGS) conjecture (3) this should lead to quantum chaos and the emergence of the Wigner-Dyson (WD) level statistics of the molecular electronic spectra (4, 5). On the one hand, this suggests a practical “complexity barrier” for accurately predicting highly excited electronic levels, because uncertainties in nuclear positions and environment (6) render such levels experimentally indistinguishable beyond a certain scale. On the other hand, Random Matrix Theory (RMT) (4) provides a wealth of universal results that may be useful to computational quantum chemistry applications.

RMT is a successful theory that was originally developed to capture universal spectral fluctuations in complex nuclei (7). It has since become a useful tool for studies of complex quantum systems including mesoscopic physics (8–10), quantum circuits (11–14), photonic systems (15–17), and strongly-correlated high-energy (18) and condensed matter physics (19). In molecular systems, RMT has been tested against molecular rovibrational and vibronic spectra, where level statistics exhibit signatures of quantum chaos (20–23). RMT has since been used to characterize vibrational couplings and vibronic states involving a few low-lying electronic levels (24, 25). In parallel, molecular reaction rate theories such as Rice-Ramsperger-Kassel-Marcus (RRKM) (26) rely on assumptions of ergodicity and statistical energy redistribution within the reactant well. Because vibrational couplings in molecules are typically local and structured, local RMT models have been developed to describe vibrational energy flow and intramolecular vibrational energy redistribution [27, 28, 29]. In contrast, the statistical properties and mixing patterns of electronic states for molecular systems have been less systematically examined within a random-matrix framework, despite the potentially more extended nature of electronic Coulomb and exchange interactions.

This paper uses ab initio wavefunction and density functional electronic structure calculations to examine the BGS conjecture in quantum chemistry on the examples of several common molecules. We specifically study electronic spectra of benzene, alanine, methyloxirane, 1-phenylethylamine, and complex helicene chains. We find Wigner-Dyson Gaussian orthogonal ensemble (GOE) level statistics in the absence of external fields for all low symmetry geometries. In the case of higher-symmetry molecules (benzene and helicene chains) we recover GOE statistics by displacing the atoms from their equilibrium positions along molecular dynamic trajectories or through atom substitutions that break point group symmetries that otherwise obscure the universal distribution. In the presence of an ultrastrong magnetic field the spectra of all molecules show Gaussian unitary ensemble (GUE) statistics. We further study the random matrix structure of the spectrum in a varying external electric field and investigate the statistics of level velocities and curvatures, which can be related directly to experimental observables such as transition moments and polarizabilities correspondingly. The infrared limit of the statistical distribution of molecular polarizabilities is predicted to be universal and non-analytic in magnetic field.

Results and Discussion

Single-particle quantum chaos. We start by investigating single-particle quantum chaos in the electronic structures of a set of small molecules. Within the Born-Oppenheimer (BO) framework, [30, 31] the electronic structure for state II is determined by solving the electronic Schrödinger equation at each nuclear configuration R=(R1,⋯ ,RNn)\mathbf{R} = (\mathbf{R}_1, \cdots, \mathbf{R}_{N_n}).

H^el(r;R)ΨI(r;R)=EI(R)ΨI(r;R),with(1)\hat{H}_{\mathrm{el}}(\mathbf{r}; \mathbf{R})\Psi_I(\mathbf{r}; \mathbf{R}) = E_I(\mathbf{R})\Psi_I(\mathbf{r}; \mathbf{R}), \qquad\text{with} \tag*{(1)}
H^el(r;R)=∑i=1Nep^i22me+V^ee(r)+V^en(r,R)+Vnn(R)\hat{H}_{\mathrm{el}}(\mathbf{r}; \mathbf{R}) = \sum_{i=1}^{N_e}\frac{\hat{\mathbf{p}}_i^2}{2m_e} + \hat{V}_{ee}(\mathbf{r}) + \hat{V}_{en}(\mathbf{r},\mathbf{R}) + V_{nn}(\mathbf{R})

being the non-relativistic electron Hamiltonian, where r=(r1,⋯ ,rNe)\mathbf{r} = (\mathbf{r}_1, \cdots, \mathbf{r}_{N_e}) denotes the electronic coordinates. p^i=−iℏ∇ri\hat{\mathbf{p}}_i = -i\hbar\nabla_{\mathbf{r}_i} is the momentum operator for the ii-th electron and V^ee(r)\hat{V}_{ee}(\mathbf{r}), V^en(r,R)\hat{V}_{en}(\mathbf{r}, \mathbf{R}), and Vnn(R)V_{nn}(\mathbf{R}) denote the electron-electron, the electron-nuclear, and the nuclear-nuclear Coulomb interactions, respectively. At the Hartree-Fock (HF) level, the ground state many-electron wavefunction Ψ0(r;R)\Psi_0(\mathbf{r}; \mathbf{R}) is approximated by a single Slater determinant of single particle orbitals {ψi(r)}\{\psi_i(\mathbf{r})\}.

of single particle orbitals {ψi(r)}\{\psi_i(\mathbf{r})\}. Applying the Rayleigh-Ritz variational principle [32] yields a set of effective single particle-equations for the orbitals ψi(r)\psi_i(\mathbf{r}) and their corresponding energies ϵi\epsilon_i, which need to be solved self-consistently.

[p^22me+Veff(r)]ψi(r)=ϵiψi(r),(2)\left[\frac{\hat{\mathbf{p}}^2}{2m_e}+V_{\mathrm{eff}}(\mathbf{r})\right]\psi_i(\mathbf{r})=\epsilon_i\psi_i(\mathbf{r}), \tag*{(2)}

Veff(r)V_{\mathrm{eff}}(\mathbf{r}) is the effective mean-field potential due to electron-nuclear and electron-electron interactions.

We calculated the HF orbital energies for three small chiral molecules (alanine, methyloxirane, and 1-phenylethylamine), a high-symmetry D6hD_{6h} benzene structure, and a fully symmetry-broken C1C_1 benzene structure generated by perturbing the molecule along a combination of its vibrational modes 1, 3 and 6. In Figure 1, we plot histograms of the nearest-neighbor spacings of the unfolded single-particle energies of these molecules (lower panel). We observe WD GOE statistics for the three chiral molecules in Figure 1 (a)–(c) and for the symmetry-broken benzene in Figure 1 (d), but not for the highly symmetric benzene in Figure 1 (e). This is because the Hamiltonian of the high-symmetry benzene structure decomposes into decoupled symmetry blocks, each of which is expected to exhibit WD symmetry, whose overlapping spectra obscure level repulsion. Displacing the atoms mixes them up and restores a single WD block.

Image not extracted

Figure 1. Nearest-neighbor level spacing histograms of the unfolded single-particle energies for (a) methyloxirane, (b) alanine, (c) 1-phenylethylamine, (d) C1C_{1} benzene, and (e) D6hD_{6h} benzene. The molecular structures are shown from left to right in the same order and share the same Cartesian system (x,y,z) with identical axis orientations. The magenta arrows represent the atomic displacement in benzene, which reduces its group symmetry from D6hD_{6h} to C1C_{1}. Carbon atoms are shown in dark gray, hydrogen in light gray, nitrogen in blue, and oxygen in red.

Random wave conjecture. We now examine the structure of single-particle chaotic wave-functions. In Ref. [33], Berry proposed an appealing heuristic microcanonical picture of the Wigner function, W(r,p)=∫ψ∗(r+p/2)ψ(r−p/2)eip⋅rd3pW(\mathbf{r},\mathbf{p}) = \int\psi^*(\mathbf{r} + \mathbf{p}/2)\psi(\mathbf{r} - \mathbf{p}/2)e^{i\mathbf{p}\cdot\mathbf{r}}d^3p, which was conjectured to be equidistributed on the classical manifold, ϵ=H(r,p)\epsilon= H(\mathbf{r},\mathbf{p}). For chaotic billiards [34, 35], this leads to a ring-shaped momentum space structure of the high-energy states. The molecular problem is formulated in non-compact space, includes both discrete and continuum states and hence is qualitatively different from billiards. To our knowledge there is no widely accepted theory of random eigenfunctions in this regime, which should be informed by the structure of classical electron trajectories in the self-consistent Hartree or Hartree-Fock potential. If the latter changes strongly around a classically chaotic orbit, we expect the Berry spherical momentum-shell to be smeared out. We found this to be the case for most excited states of the small molecules we studied. See, Figure 2, where we observe regular structures, some of which may represent quantum scars [36]. Ring-shaped structures consistent with Berry’s random wave conjecture can also be seen in the projected momentum-space wave-functions of select high-energy states. However these data should be interpreted with care, because the pseudocontinuum states may mimic random waves.

Image not extracted

Figure 2. Three representative eigenfunctions of the perturbed benzene (n=697n = 697 - top panel, n=704n = 704 - middle panel, and n=861n = 861 - lower panel). The left panel displays the isosurface plots (∣ψ(r)∣2=0.01|\psi(\mathbf{r})|^2 = 0.01 a.u.) of molecular virtual orbital wavefunction, with red and blue colors representing opposite signs. The middle panel displays the position space ∫∣ψ(r)∣2dr\int|\psi(\mathbf{r})|^2 d\mathbf{r} and the right panel is the momentum space projection ∫∣ψ(p)∣2dp\int|\psi(\mathbf{p})|^2d\mathbf{p}. The red circle plotted in the right panel represents equimomentum contour defined by the average kinetic energy of the orbital px2+py2=2me(ϵ−⟨Veff⟩)p_x^2 + p_y^2 = 2m_e(\epsilon- \langle V_{\mathrm{eff}}\rangle).

Many-particle quantum chaos. In real molecules, the energies of single-electron excitations are not the integrals of motion and only the many-body spectrum constrained by the total energy conservation is meaningful. To observe the emergence of many-body quantum chaos, we solve the electronic Hamiltonian in Eq. (1) within the subspace of singly excited configurations built on the HF ground state, ∣Ψia⟩=a^a†a^i∣Ψ0⟩|\Psi_i^a\rangle= \hat{a}_a^\dagger\hat{a}_i|\Psi_0\rangle, where a^i\hat{a}_i and a^a†\hat{a}_a^\dagger denote the annihilation and creation operators that act on the occupied orbital ii and the virtual orbital aa. The excited wavefunction is then approximated as a linear combination of these singly excited configurations

∣ΨJ⟩=∑i,aXiaJ∣Ψia⟩(3)|\Psi_J\rangle= \sum_{i,a} X_{ia}^J |\Psi_i^a\rangle \tag*{(3)}

where the excitation amplitudes XiaJX_{ia}^J are solved by diagonalizing the electronic Hamiltonian in Eq. (1) in the single-excitation basis,

∑j,bHia,jbXjbJ=EJXiaJ,Hia,jb=(ϵa−ϵi)δijδab+⟨ai∥bj⟩.(4)\sum_{j,b} H_{ia,jb}X_{jb}^J = E_J X_{ia}^J,\quad H_{ia,jb} = (\epsilon_a-\epsilon_i)\delta_{ij}\delta_{ab} + \langle ai\Vert bj\rangle. \tag*{(4)}

The Coulomb and exchange two-electron integrals are

⟨ai∥bj⟩=⟨ai∣bj⟩−⟨ai∣jb⟩(5)\langle ai\Vert bj\rangle= \langle ai|bj\rangle-\langle ai|jb\rangle \tag*{(5)}
⟨pq∣rs⟩=∫dr1dr2ψp∗(r1)ψq∗(r2)ψr(r1)ψs(r2)∣r1−r2∣.(6)\langle pq|rs\rangle= \int d\mathbf{r}_1d\mathbf{r}_2 \frac{\psi_p^*(\mathbf{r}_1)\psi_q^*(\mathbf{r}_2)\psi_r(\mathbf{r}_1)\psi_s(\mathbf{r}_2)}{|\mathbf{r}_1-\mathbf{r}_2|}. \tag*{(6)}

Here indices ijij label occupied orbitals, abab label virtual orbitals, and pqrspqrs denote general orbital indices. This corresponds to the standard configuration interaction singles (CIS) approximation in quantum chemistry [37].

If we ignore interactions for excited electron states by setting ⟨ai∥bj⟩=0\langle ai\Vert bj\rangle= 0, there are no thermalization processes and each electron’s energy is conserved. The model is chaotic in the single-particle sense, but integrable in the many-body sense [38]. We compute Slater determinants of the single-particle orbitals and observe the Poisson statistics of the many-body energies as expected, see Figure 3(f)–(j).

Image not extracted

Figure 3. Nearest-neighbor level spacing statistics for (a)–(e) interacting and (f)–(j) non-interacting many particle excitations of (a)(f) methyloxirane, (b)(g) alanine, (c)(h) 1-phenylethylamine, (d)(i) C1_1 benzene, and (e)(j) D6h_{6h} benzene. The three chiral molecules and the perturbed benzene molecule have no spatial symmetries and the nearest-neighbor level spacings show GOE distributions. The symmetric benzene molecule is known to have D6h_{6h} group symmetry and mixing excited states across symmetry groups disrupt the GOE distribution.

Including interactions in earnest effectively makes the CIS Hamiltonian Hia,jbH_{ia,jb} a random matrix. Our numerical analysis of the many-body spectrum shows its excellent convergence to Wigner-Dyson GOE statistics, Figure 3, for all molecules apart from D6h_{6h} benzene, which shows an accumulation of low-energy states, see Figure 3e. As before, this is due to high symmetry of benzene, whose breaking restores Wigner-Dyson statistics - Figure 3d.

While, we have not examined the structure of chaotic many-electron eigenstates in detail, the eigenstate thermalization hypothesis (ETH) [39, 40, 41] predicts that they should behave as pseudo-random vectors in the configuration basis. When expanded in a DD-dimensional basis {∣Φi⟩}\{|\Phi_i\rangle\}, the coefficients cic_i of an eigenstate ∣ΨJ⟩=∑i=1Dci∣Φi⟩|\Psi_J\rangle= \sum_{i=1}^{D} c_i |\Phi_i\rangle are expected to have zero mean and variance ∼1/D\sim1/D.

Spectral form-factor and quantum dynamics

The spectral form-factor (SFF) is a more refined measure of level statistics than the level spacing histogram and is defined as a Fourier transform of the two-point level correlation function or equivalently [42, 43] [4]

K(t)=1(Tr⁡1^)2⟨∣Tr⁡e−iH^t∣2⟩=1D2∑n,m=1D⟨e−i(En−Em)t⟩(7)K(t) = \frac{1}{(\operatorname{Tr}\hat{1})^2}\left\langle\left|\operatorname{Tr}e^{-i\hat{H}t}\right|^2\right\rangle= \frac{1}{D^2}\sum_{n,m=1}^{D}\left\langle e^{-i(E_n-E_m)t}\right\rangle \tag*{(7)}

where DD is the size of the relevant Hilbert space. In a finite-dimensional Hilbert space, the choice of a basis is irrelevant. Otherwise, including our case, it is dictated by physics. For example, if the molecule is kicked into an excited state spanned over a certain energy window, the latter determines the relevant subspace [44].

Figure 4 displays the molecular SFFs, which collapse into the canonical “quantum chaotic” structure [45, 46] for all geometries except the symmetric benzene: see, the dip, ramp, and plateau regions. At t=0t = 0, the SFF is trivially K(0)=1K(0) = 1. The early-time dynamics involves a sharp drop to ∼1/D\sim1/D - the dip - which occurs at the Thouless time. This perturbative regime is determined by the non-universal density of states (DoS), which is highly system specific and is not expected to follow the Wigner semicircle law [47].

Image not extracted

Figure 4. Spectral form factors calculated with unfolded energy levels for alanine (blue), C1C_{1} benzene (orange), D6hD_{6h} benzene (green), methyloxirane (red), 1-phenylethylamine (purple). The molecules that exhibit WD nearest-level statistics (all except D6hD_{6h} benzene, shown in green) also display a ramp following the initial dip, consistent with a hallmark signature of quantum chaos.

Interestingly, the time it takes for the system to “scramble” (or “forget” the initial state, see Refs. [43, 44, 47] for details) is approximately the Thouless time. While solving the time-dependent Schrödinger equation is often a prohibitively difficult task, using the RMT approach circumvents the need to rederive statistical mechanics from scratch for each individual system, and ties quantum dynamics to relatively simple non-universal features of the bare SFF.

The initial dip is followed by a ramp, which is the hallmark feature of quantum chaos related to interference of unstable periodic orbits and time-translation symmetry [45, 48] [1]. The ramp rises to a plateau value at the Heisenberg time and the SFF stays there. It corresponds to energy scales below the inverse level spacing where correlations are absent. We observe the standard dip-ramp-plateau structure, which is accompanied by wild fluctuations representing finite-size effects. We emphasize that the ramp is a stronger evidence of quantum chaos than nearest-level repulsion, as it probes a more general pair correlation function. The latter appears in the distributions of unrelated structures from eigenenergies of billiards and disordered metals [49] to non-trivial zeroes of the Riemann zeta function [50].

Effect of the magnetic field

Next, we consider level statistics in the presence of an external magnetic field by including it in Eq. (1) via minimal substitution p^i→p^i+eA(r^i)\hat{\boldsymbol{p}}_i \to\hat{\boldsymbol{p}}_i + e\boldsymbol{A}(\hat{\boldsymbol{r}}_i), where A(r^i)=12B×(r^i−G)\boldsymbol{A}(\hat{\boldsymbol{r}}_i) = \frac{1}{2}\boldsymbol{B} \times(\hat{\boldsymbol{r}}_i - \boldsymbol{G}) is the vector potential and r^i\hat{\boldsymbol{r}}_i is the position operator of the iith-electron. In quantum chemistry, it is customary to include the “gauge origin,” G\boldsymbol{G}, even though the additional term represents a pure gauge. Obviously, any gauge transform including a shift of G\boldsymbol{G} must have no effect on observables. In a finite computational basis of Gaussian atomic orbitals however, ensuring gauge invariance is non-trivial as discussed in Refs. [51, 52]. We ensured gauge invariance, including ∂E∂G=0\frac{\partial E}{\partial\boldsymbol{G}} = 0, by employing the standard procedure in quantum chemistry based on the London orbitals [51, 52]. This entails attaching an atomic-centered U(1)U(1) gauge phase-factor to each atomic basis function; i.e., ϕLondon(r)≡e−i2ℏB×(RM−G)⋅rϕ(r−RM)\phi_{\mathrm{London}}(\boldsymbol{r}) \equiv e^{-\frac{i}{2\hbar}\boldsymbol{B}\times(\boldsymbol{R}_M-\boldsymbol{G})\cdot\boldsymbol{r}}\phi(\boldsymbol{r}-\boldsymbol{R}_M), where MM labels the nuclear center of the basis function.

The magnetic field breaks time-reversal symmetry and is expected to change the WD statistics from the Gaussian orthogonal to Gaussian unitary ensemble (GUE) in the absence of other antiunitary symmetries (as present for magnetic hydrogen [53] and we also expect this symmetry asymptotically restored for highly excited molecular Rydbergs). However, capturing the GUE is non-trivial for small molecules because the magnetic energy scale is often smaller than the relevant level spacing. Figure 5 plots the nearest-neighbor many-body level spacing statistics for alanine and C1C_1 benzene molecules where the statistics remains GOE for physically meaningful out-of-plane field. To enforce the transition to GUE in simulations, we applied an extremely large magnetic field, Bz>10−2B_z > 10^{-2} a.u. This scale can be qualitatively understood by comparing the flux through an electronic orbital Φ∼πBz⟨r^2⟩\Phi\sim\pi B_z\langle\hat{r}^2\rangle to the flux quantum.

Image not extracted

Figure 5. Nearest-neighbor level spacing statistics for interacting CIS excitations of (a)-(b) alanine, (c)-(d) perturbed benzene in an external magnetic field along the zz direction with Bz=0.001B_z = 0.001 a.u. in (a)(c) and Bz=0.01B_z = 0.01 a.u. in (b)(d). The xx, yy, and zz axes are shown in the molecular frames in orange, green, and blue. The transition from GOE to GUE becomes more apparent with Bz=10−2B_z = 10^{-2} a.u.

This however does not mean that GUE is out of reach for realistic electronic spectra. Spectra of larger molecular structures (e.g., multiple-ring helicene) are expected to turn GUE for smaller orbital fields. Furthermore, application of a circularly polarized light may mimic the effect of an orbital magnetic field [54, 55], albeit in non-equilibrium with A(t)=F0/Ω(±cos⁡Ωt,sin⁡Ωt,0)\mathbf{A}(t) = F_0/\Omega(\pm\cos\Omega t,\sin\Omega t,0), where F0F_0 is the amplitude of the periodic electric field, F(t)=−∂tA\mathbf{F}(t) = -\partial_t\mathbf{A} and Ω=2π/T\Omega= 2\pi/T is its frequency. The solution to the time-dependent Schrödinger equation for the time-evolution operator, i∂tU^(t)=H^(t)U^(t)i\partial_t\hat{U}(t) = \hat{H}(t)\hat{U}(t) satisfies the Floquet theorem [56], U^(t)=P^(t)e−iF^t\hat{U}(t) = \hat{P}(t)e^{-i\hat{F}t}, where P^(t)=P^(t+T)\hat{P}(t) = \hat{P}(t+T) is a periodic unitary which resets every period, P^(nT)=1^\hat{P}(nT) = \hat{1} and F^\hat{F} is the time-independent Floquet Hamiltonian, which determines effective electron properties [54, 55, 57]. Since the spectrum of F^\hat{F} is defined on a circle modulo Ω\Omega, we expect the Floquet unitary operator to crossover from the circular orthogonal ensemble (COE) to the circular unitary ensemble (CUE) as the drive strength increases.

Level velocities and curvatures. Changing the Hamiltonian in Eq. (1) by either displacing the atoms due to atomic motion, R(t)\mathbf{R}(t),1 or the application of the electric field changes the realization of the electronic spectrum – energy levels move. Specifically, for the electric field, F\mathbf{F}, the Hamiltonian reads.

H^F(r,R,F)=H^el(r,R)−μ^e⋅F.(8)\hat{H}_{\mathbf{F}}(\mathbf{r},\mathbf{R},\mathbf{F}) = \hat{H}_{\mathrm{el}}(\mathbf{r},\mathbf{R}) - \hat{\boldsymbol{\mu}}_{\mathrm{e}}\cdot\mathbf{F}. \tag*{(8)}

Using perturbation theory, its eigenstates {ψ}\{\psi\} can be written in terms of the eigenstates {ϕ}\{\phi\} of the unperturbed Hamiltonian, ∣ψI(F)⟩=∣ϕI⟩+∑J≠I⟨ϕJ∣μ^e⋅F∣ϕI⟩EI−EJ∣ϕJ⟩+O(∣F∣2)|\psi_I(F)\rangle= |\phi_I\rangle+ \sum_{J\ne I}\frac{\langle\phi_J|\hat{\boldsymbol{\mu}}_{\mathrm{e}}\cdot\mathbf{F}|\phi_I\rangle}{E_I-E_J}|\phi_J\rangle+ O(|F|^2). Hence, the level velocity for the state II with respect to the field (corresponding to the electric dipole matrix element) is ∂EI∂F=−⟨ψI∣μ^e∣ψI⟩\frac{\partial E_I}{\partial F} = -\langle\psi_I|\hat{\mu}_{\mathrm{e}}|\psi_I\rangle by the Hellmann-Feynman theorem. The level curvature (the diagonal components of the electric polarizability tensor) is

∂2EI∂F2=−2∑J≠I∣⟨ψJ∣μ^e∣ψI⟩∣2EJ−EI(9)\frac{\partial^2 E_I}{\partial F^2} = -2\sum_{J\ne I}\frac{\left|\langle\psi_J|\hat{\mu}_{\mathrm{e}}|\psi_I\rangle\right|^2}{E_J-E_I} \tag*{(9)}

Figure 6a shows the flow of random matrix spectra as a function of the electric field. Figure 6b is the same but under nuclear motion parameterized by time, tt, where the nuclei trace out a realistic trajectory, R(t)\mathbf{R}(t) calculated using molecular dynamics simulations. In both cases, we observe a highly “chaotic” structure, which illustrates the sensitivity of electron spectra and suggests that Dyson Brownian motion (6) due to environmental noise may render individual highly excited electronic levels ill-defined in practice for complex molecules and materials.

Image not extracted

Figure 6. Unfolded energies ϵI=EIΔ\epsilon_I = \frac{E_I}{\Delta} with a mean level spacing Δ\Delta as a function of the unfolded parameter (a) x=C0(F−F0)x = \sqrt{C_0}(F-F_0), where C0=⟨(∂ϵ∂F)2⟩C_0 = \left\langle\left(\frac{\partial\epsilon}{\partial F}\right)^2 \right\rangle is the averaged level velocity variance and FF is the external electric field. (b) x=C0tx = \sqrt{C_0}t, where C0=⟨(∂ϵ∂t)2⟩C_0 = \left\langle\left(\frac{\partial\epsilon}{\partial t}\right)^2 \right\rangle is the averaged level velocity variance. The energy levels are varied as a function of time, tt, due to atomic motion Ra(t)\mathbf{R}_a(t).

However, this level dynamics has a hidden statistical structure [58, 59, 60] [61]. For the canonical Gaussian random matrix ensemble, the distribution of level velocities is Gaussian. However, this Gaussianity is non-universal being sensitive to the scar structure of more realistic physical systems. As a representative example, we show the velocity statistics of the canonical spectra of perturbed benzene in Figure 7(a-Figure 7c), where small velocity structure does differ from a Gaussian. Within the GOE RMT, the full distribution of level curvatures for unfolded eigenvalues was conjectured by Zakrzewski and Delande [59] to take the form

Image not extracted

Figure 7. Level velocity and curvature statistics of the C1 benzene in a varying external electric field FF. Histograms of unfolded level velocities vσ=∂E∂Fσ/⟨(∂E∂Fσ)2⟩v_\sigma= \frac{\partial E}{\partial F_\sigma}/\sqrt{\left\langle\left(\frac{\partial E}{\partial F_\sigma}\right)^2\right\rangle} along σ=x,y,z\sigma= x,y,z axis are shown in (a)-(c), respectively, which approximates the transition electric dipole moment within the CIS framework. The histogram and tail statistics for the unfolded isotropic curvature κ=13∑σ∂2E∂Fσ2/⟨(13∑σ∂E∂Fσ)2⟩\kappa= \frac{1}{3}\sum_\sigma\frac{\partial^2 E}{\partial F_\sigma^2}/\left\langle\left(\frac{1}{3}\sum_\sigma\frac{\partial E}{\partial F_\sigma}\right)^2\right\rangle is shown in (d) and (e), which approximates the isotropic polarizability within the CIS framework. The variance of the curvature diverges at B→0B \to0 is shown schematically in (f).

PGOE(K)=12γ01[1+(K/γ0)2]3/2,(10)P_{\mathrm{GOE}}(K) = \frac{1}{2\gamma_0}\frac{1}{\left[1+(K/\gamma_0)^2\right]^{3/2}}, \tag*{(10)}

where γ0\gamma_0 is a curvature scale fixed by the mean level spacing and the variance of the level velocities. The low-

KK regime corresponds to an adiabatic level dynamics, while the universal high-KK limit [62] [61] is related to the statistics of avoided level-crossings. They are shadows of the nearby conical intersections [64] [63] (or equivalently “diabolical points” (65)) accessible through two-parameter flows or possibly higher-dimensional degeneracy structures in the larger parameter space (66). However, the exact degeneracies have zero probability to occur within a one-parameter flow without special symmetries or fine-tuning. This is the case in our simulations, Figure 7, where we recover the universal P(K)∝∣K∣−3P(K) \propto\lvert K\rvert^{-3} tail.

Because of the slow 1/∣K∣31/\lvert K\rvert^3 decay, the second moment of PGOE(K)P_{\mathrm{GOE}}(K) diverges logarithmically. The infrared regularization is dictated by physics of the problem. For example a weak field, B=Bez\mathbf{B}=B\mathbf{e}_z introduces off-diagonal terms in the minimal 2×22\times2 matrix model of the avoided crossing and cuts off the variance of the curvature:

⟨K2⟩∼γ2ln⁡(B∗∣B∣),B→0.(11)\left\langle K^2\right\rangle\sim\gamma^2 \ln\left(\frac{B_*}{\lvert B\rvert}\right), \qquad B\to0. \tag*{(11)}

where B∗B_* is a constant. Eq. (11) shows that the ensemble-averaged curvature in the orthogonal regime grows logarithmically as B→0B\to0. This is reminiscent of the weak localization correction to conductivity, which scales the same way as Eq. (11) in two dimensional systems, and is ubiquitously observed in experiment (15).

Character of the electronic RMT states. In the single-particle description, the electronic states can be classified in increasing energy as core orbitals, valence orbitals, valence virtual orbitals, Rydberg orbitals, and continuum states. The continuum states have no energy gaps and hence have no simple notion of level statistics in terms of discrete level spacings. Yet, the finite Gaussian atomic orbital basis effectively introduces pseudocontinuum states

as discretized artifacts in the simulations. The Rydberg orbitals are hydrogen-like and there are an infinite number of them (they should still yield WD statistics in the presence of strong interactions due to multi-electron excitations, but observing it in finite-basis simulations is challenging). To connect imperfect simulations to the spectra of real molecules, it is therefore important to remove the pseudocontinuum and Rydberg states and isolate the valence orbitals which tend to couple strongly. To collect meaningful statistics of these discrete states, we study long-chain helicene molecules with broken C2C_2 symmetry through atomic substitution and atomic displacement along the C2C_2-symmetry-breaking vibrational mode. In Figure 8, we used density functional theory (DFT) method to plot the neighboring orbital energy differences for the bound states, excluding the localized core orbitals. Here, we will see clear tendency towards the WD statistics which is a more quantitatively reliable result.

Image not extracted

Figure 8. Nearest-neighbor unfolded level spacing statistics for (a)-(b) bound states and (e)-(f) full spectra of non-interacting orbital energies (black) of [36]-helicene and (c)-(d) bound states and (g)-(h) both bound and unbound states of interacting excited states (blue) of [18]-helicene. The C2C_2 symmetry of these helicene chains is broken either through substitution of nitrogen atoms as shown in (a)(c)(e)(g) or displacement of nuclear Cartesian coordinates along the 6π-torsional mode as shown in (b)(f)(d)(h). The histograms constructed from both bound and unbound states exhibit Wigner-Dyson GOE statistics. Similar characteristics is observed when only the bound states are considered, with the degree of agreement depending on the extent to which the C2C_2 symmetry is broken.

We also perform a study of the many-body level statistics of the valence states (by explicitly excluding the Rydberg orbitals, core excitations, resonances and continuum states for the long helicene molecules). Similarly, we observe many-body quantum chaos generated by these bound states. Figure 8 plots the neighboring level statistics for the excited states below the first ionization threshold calculated using linear-response time-dependent DFT within the Tamm-Dancoff Approximation (LR-TDDFT/TDA). (67) We performed all the calculations in a developmental branch of Q-Chem software (68).

Conclusions

We have shown that ab initio electronic structure methods effectively generate random matrices. Although we restrict our analysis to the singlet single excitations, the lower-energy bound eigenstates of these matrices can provide accurate predictions of the molecular energetics. In finite basis sets, high-lying states are not necessarily individually accurate, but their statistical properties are expected to be robust, universal, and experimentally testable. This conceivably may explain why increasing the basis size and summing over highly excited electron levels often improves convergence of quantum chemistry computations and their correspondence to experimental data [69, 70]. In both cases, there may be two key contributions at play: low-energy non-chaotic levels (captured accurately by ab initio numerics) and high-energy RMT levels. To verify these conjectures, we propose measuring universal statistics of level crossings through electric and magnetic polarizabilities [71]. Experimentally probing the Wigner-Dyson statistics of the molecular electronic structure directly should also be within reach. Even though separating electronic excitations from vibronic states may be difficult in practice, the combined spectrum may still exhibit RMT-like statistics in regimes of strong mixing. Finally, beyond the bulk Wigner-Dyson statistics studied here, RMT also predicts universal behavior at the spectral edge, where the fluctuations of the smallest eigenvalue follow the Tracy-Widom distribution [72, 73].

This work raises follow-up questions about the signatures of quantum chaos arising from different physical interactions in chemical systems. On the one hand, stronger electronic correlation can be further investigated using the state-of-the-art electronic structure methods that account for higher-order excitations and multireference character [74, 75]. On the other hand, vibronic mixing becomes increasingly important for highly excited electronic states in regimes of large density of states, which can give rise to chaotic features beyond the Born-Oppenheimer picture [76, 77, 78]. Finally, although spin-orbit coupling is generally weak in organic molecules and we neglected it here, stronger spin-orbit-coupled molecules are expected to exhibit features of the third canonical RMT ensemble - Gaussian symplectic ensemble (GSE) - with Kramers-degenerate levels and a distinct pattern of level repulsion in molecular spectra.

Acknowledgments

This work was supported by startup funds from the University of Rhode Island (Z.T.), the U.S. Department of Energy, Office of Science Basic Energy Sciences under Award No. DE-SC0001911 and the Julian Schwinger Foundation (V.G.). We acknowledge the use of the Unity Cluster for computational resources. Z.T. would like to thank Tanner Culpitt for helpful discussions and assistance with the magnetic field calculations.

References

  1. [27]Peter G. Wolynes. Randomness and complexity in chemical physics. Accounts of Chemical Research, 24(11):513–519, 1992. ISSN 0001-4842. · URL https://doi.org/10.1021/ar00023a005.
  2. [28]David M. Leitner. Quantum ergodicity and energy flow in molecules. Advances in Physics, 64(4):445–517, 2015. ISSN 0001-8742. · URL https://doi.org/10.1080/00018732.2015.1059817.
  3. [29]David M. Leitner. Molecules and the Eigenstate Thermalization Hypothesis. Entropy, 20(9), 2018. ISSN 1099-4300. · URL https://www.mdpi.com/1099-4300/20/9/673.DOI
  4. [30]Born, M., & Oppenheimer, R. (1927). Zur Quantentheorie der Molekeln. Annalen der Physik, 389, 457–484.
  5. [31]Max Born, Kun Huang, and M Lax. Dynamical theory of crystal lattices. American Journal of Physics, 23(7):474–474, 1955. .DOI
  6. [32]Szabo, A., & Ostlund, N. S. (1989). Modern Quantum Chemistry. In Dover Publications. https://doi.org/ISBN 978-0-486-13459-8
  7. [33]M. V. Berry. Regular and irregular semiclassical wavefunctions. Journal of Physics A: Mathematical and General, 10(12):2083–2091, 1977. .DOI
  8. [34]L. A. Bunimovich. On ergodic properties of certain billiards. Functional Analysis and Its Applications, 8(3):254–255, 1974. .DOI
  9. [35]Rozembaum, E. B., Bunimovich, L. A., & Galitski, V. (2020). Early-time exponential instabilities in nonchaotic quantum systems. Physical Review Letters, 125, 014101.
  10. [36]Eric J. Heller. Bound-state eigenfunctions of classically chaotic Hamiltonian systems: Scars of periodic orbits. Physical Review Letters, 53(16):1515–1518, 1984. .DOI
  11. [37]Andreas Drew and Martin Hegedorn. Single-reference ab initio methods for the calculation of excited states of large molecules. Chem. Rev., 105(11):4009–4037, 2005. ISSN 0009-2665. . URL https://doi.org/10.1021/cr0505627.
  12. [38]Yunxiang Liao, Amit Vikram, and Victor Galitski. Many-body level statistics of single-particle quantum chaos. Physical Review Letters, 125:250601, 2020. .DOI
  13. [39]J. M. Deutsch. Quantum statistical mechanics in a closed system. Phys. Rev. A, 43:2046–2049, 1991.pubmed.ncbi.nlm.nih.gov/9905246
  14. [40]M. Srednicki. Chaos and quantum thermalization. Phys. Rev. E, 50:888–901, 1994.DOI
  15. [41]L. D’Alessio, Y. Kafri, A. Polkovnikov, and M. Rigol. From quantum chaos and eigenstate thermalization to statistical mechanics and thermodynamics. Adv. Phys., 65:239–362, 2016.DOI
  16. [42]R. E. Prange. The spectral form factor is not self-averaging. Physical Review Letters, 78:2280–2283, 1997. .DOI
  17. [43]Amit Vikram and Victor Galitski. Exact universal bounds on quantum dynamics and fast scrambling. Physical Review Letters, 132:040402, 2024. .DOI
  18. [44]Peter Reimann. Typical fast thermalization processes in closed many-body systems. Nature Communications, 7:10821, 2016. .DOI
  19. [45]M. V. Berry. Semiclassical theory of spectral rigidity. Proceedings of the Royal Society of London A, 400:229–251, 1985.DOI
  20. [46]Yunxiang Liao and Victor Galitski. Universal dephasing mechanism of many-body quantum chaos. Physical Review Research, 4:L012037, 2022. .DOI
  21. [47]Amit Vikram, Laura Shou, and Victor Galitski. Proof of a universal speed limit on fast scrambling in quantum systems. Physical Review Letters, 2026. To appear.DOI
  22. [48]N. Argaman, F.-M. Dittes, E. Doron, J. P. Keating, A. Y. Kitaev, M. Sieber, and U. Smilansky. Correlations in the actions of periodic orbits derived from quantum chaos. Physical Review Letters, 71:4326–4329, 1993. .pubmed.ncbi.nlm.nih.gov/10055218
  23. [49]Alexander Altland and Alex Kamenev. Wigner-dyson statistics from the keldysh σ-model. Physical Review Letters, 85:5615–5618, 2000. .pubmed.ncbi.nlm.nih.gov/11136060
  24. [50]Hugh L. Montgomery. The pair correlation of zeros of the zeta function. In Analytic Number Theory, volume 24 of Proceedings of Symposia in Pure Mathematics, pages 181–193. American Mathematical Society, Providence, RI, 1973.DOI
  25. [51]F. London. Théorie quantique des courants interatomiques dans les combinaisons aromatiques. Journal de Physique et le Radium, 8(10):397–409, 1937. . URL https://hal.science/jpa-00233534.DOI
  26. [52]Helgaker, T., & Jo/rgensen, P. (1991). An electronic hamiltonian for origin independent calculations of magnetic properties. J. Chem. Phys., 95(4), 2595–2601. https://doi.org/10.1063/1.460912
  27. [53]B. D. Simons, A. Hashimoto, D. Kleppner, and B. L. Altshuler. New class of universal correlations in the spectra of hydrogen in a magnetic field. Physical Review Letters, 71(18):2899–2902, 1993. .pubmed.ncbi.nlm.nih.gov/10054807
  28. [54]Y. H. Wang, H. Steinberg, P. Jarillo-Herrero, and N. Gedik. Observation of floquet-bloch states on the surface of a topological insulator. Science, 342(6157):453–457, 2013. .DOI
  29. [55]Netanel H. Lindner, Gil Refael, and Victor Galitski. Floquet topological insulator in semiconductor quantum wells. Nat. Phys., 7:490–495, 2011. .DOI
  30. [56]Gaston Floquet. Sur les équations différentielles linéaires à coefficients périodiques. Annales Scientifiques de l’École Normale Supérieure, 12:47–88, 1883. .DOI
  31. [57]Z. Zhou, Y. Wu, X. Bian, and J. E. Subotnik. Nonadiabatic dynamics in a continuous circularly polarized laser field with flouquet phase-space hopping. J. Chem. Theory Comput., 19(3):718–732, 2023. ISSN 1549-9618. .DOI
  32. [58]B. D. Simons and B. L. Altshuler. Universalities in the spectra of disordered and chaotic systems. Physical Review B, 48(8):5422–5431, 1993. .pubmed.ncbi.nlm.nih.gov/10009061
  33. [59]Jakub Zakrzewski and Dominique Delande. Parametric motion of energy levels in quantum chaotic systems. I. curvature distributions. Physical Review E, 47:1650–1663, 1993. .pubmed.ncbi.nlm.nih.gov/9960190
  34. [60]Jakub Zakrzewski, Dominique Delande, and Marek Kuś. Parametric motion of energy levels in quantum chaotic systems. II. avoided-crossing distributions. Physical Review E, 47:1665–1675, 1993. .pubmed.ncbi.nlm.nih.gov/9960191
  35. [62]Yan V. Fyodorov and Hans-Jürgen Sommers. “level curvature” distribution for diffusive Aharonov–Bohm systems: Analytical results. Physical Review E, 51:R2719–R2722, 1995. .pubmed.ncbi.nlm.nih.gov/9963080
  36. [64]Wolfgang Domcke and David R. Yarkony. Role of conical intersections in molecular spectroscopy and photoinduced chemical dynamics. Annual Review of Physical Chemistry, 63:325–352, 2012. .DOI
  37. [69]Marco D’Alessandro and Luigi Genovese. Locality and computational reliability of linear response calculations for molecular systems. Phys. Rev. Mater., 3(2):023805, 2019.DOI
  38. [70]Otero, N., Panagiotakis, M., & Mandado, M. (2019). A new method to analyze and understand molecular linear and nonlinear optical responses via field-induced functions: a straightforward alternative to sum-over-states (SOS) analysis. Physical Chemistry Chemical Physics, 21(11), 6274–6286. https://pubs.rsc.org/en/content/articlelanding/2019/cp/c8cp07362gpubs.rsc.org/en/content/articlelanding/2019/cp/c8cp07362g
  39. [71]U. Hohm. Experimental static dipole–dipole polarizabilities of molecules. Journal of Molecular Structure, 1054-105:282–292, December 2013. ISSN 0022-2860. . URL https://www.sciencedirect.com/science/article/pii/S002228601300834X.DOI
  40. [72]Craig A. Tracy and Harold Widom. Level-spacing distributions and the airy kernel. Communications in Mathematical Physics, 159:151–174, 1994. .DOI
  41. [73]Craig A. Tracy and Harold Widom. On orthogonal and symplectic matrix ensembles. Communications in Mathematical Physics, 177:727–754, 1996. .DOI
  42. [74]González, L., Deleuze, D., & Suárez, L. S.-A. (2012). Progress and Challenges in the Calculation of Electronic Excited States. ChemPhysChem, 13(1), 28–51. https://doi.org/10.1002/cphc.201100200
  43. [75]Lischka, H., Nachtigallová, A., Aquilino, P. G., Szalay, F., Schapiro, I., Machado, B. C., & Barbati, M.-. reference . (2018). Multireference Approaches for Excited States of Molecules. Chemical Reviews, 118(15), 7293–7361. https://doi.org/10.1021/acs.chemrev.8b00244
  44. [76]Ali Abedi, Neepa T. Maitra, and E. K. U. Gross. Correlated electron-nuclear dynamics: Exact factorization of the molecular wavefunction. The Journal of Chemical Physics, 137(22):22A530, 2012. .DOI
  45. [77]Federica Agostini and Basile F. E. Curchod. Chemistry without the Born–Oppenheimer approximation. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 380(2223):202200375, 2022. ISSN 1364-503X. . URL https://doi.org/10.1098/rsta.2020.0375.
  46. [78]Bian, X., Tousian, T., Bradbury, N., Tao, Z., Bhati, M., Qiu, T., Wu, X., Wu, Y., & Subotnik, J. E. (2021). The phase-space way to electronic structure theory and subsequently chemical dynamics. Chemical Physics Reviews, 7(1), 011303. https://doi.org/10.1063/5.0286240

Paper details

vlnc:W_zkprtpcj
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