Recovering Exact Vibrational Energies Within a Phase Space Electronic Structure Framework

Identifying vibrational energies is one of the chemist’s most useful tools, which aids in identifying functional groups¹, molecular structure and local temperature². Moreover, the theory of how to calculate vibrational modes is effectively as old as quantum mechanics itself.³ Consider a Hamiltonian for a molecular or material system composed of nuclei and electrons:

Here, we let be the momentum of electron , while is the charge of nucleus , is the position of nucleus , and is the mass of nucleus . The electron mass is Due to the extreme computational cost in diagonalizing , it is standard to investigate chemical systems through the Born-Oppenheimer framework^{4, 5}. Within this framework, one begins any analysis by diagonalizing the electronic Hamiltonian () to generate a basis of electronic adiabatic states:

Here, denotes the ground state potential energy surface for the nuclei. Thereafter, in the basis of adiabatic electronic states, the full Hamiltonian takes the form:

where the matrix of derivative couplings are defined as:

Next, according to the BO approximation, one ignores the derivative couplings and generates the stationary states of the system by solving the nuclear Schrodinger equation on a single surface. In particular, for the ground state (0) energy, one diagonalizes the operator:

One of the most important features of the BO framework is the capacity to calculate and characterize quantum states according to the nature of the excitation, e.g. electronic, vibrational, and rotational excitations. In particular, for many chemical systems, one can diagonalize (rather than ) so as to generate vibrational excitations that agree well with experiment⁶. Thus, the agreement between BO vibrational energies and experimental spectra has always been a resounding endorsement of the notion of potential energy surfaces. That being said, however, it must be noted that within the realm of high-resolution spectroscopy, it has long been appreciated that one can measure deviations from the BO approximation⁷. Indeed, a host of corrections must be introduced if one seeks to match BO vibrational energies with high-resolution spectroscopy.^{8, 9, 10, 11} In other words, the BO approximation is (like all approximations) approximate; it is impossible to absolutely separate electronic and vibrational excitations. Indeed, this statement is not controversial for anyone familiar with the curl conditions for diabatic states.^{12, 13}

At this point, one must wonder: if one seeks theory to match high-resolution spectroscopy, is there a meaningful and efficient alternative BO framework? More bluntly, if one seeks to characterize vibrational energies within a single surface approximation, are we confident that the BO surfaces are the optimal surfaces? Is there perhaps a richer and more complete potential energy surface? Even more broadly, are there other possible basis sets and frameworks for expanding coupled nuclear-electronic wavefunctions ¹⁴?

To that end, over the last few years, our research group has vigorously investigated the possibility of replacing the BO framework and the BO electronic Hamiltonian () with a phase space framework and a phase space (PS) electronic Hamiltonian which is parameterized by both nuclear and . To be specific, we have posited a phase space Hamiltonian of the form

where is a one-electron operator that satisfies key translation and rotational conditions that are obeyed by the adiabatic derivative coupling. For the exact form of , please see Refs. 15, 16, as well as Eq. 120 below. (Note, though, that all formal work below will not depend on the particular form of .)

In all cases to date, a phase space electronic Hamiltonian of the form in Eq. 8 has outperformed the BO electronic Hamiltonian, often with very meaningful implications. In particular, a PS approach outperforms BO as far as:

1. 1.

   PS theory offers meaningful (nonzero) electronic momenta for dynamics along the ground state^{15, 17}, whereas BO theory predicts zero electronic momenta, a fact which has motivated the theory for quite some time^{18, 19, 20}.

2. 2.

   PS conserves the total linear and angular momentum^{15, 21}. Formally BO theory also conserves the total nuclear momentum²² and classical BO dynamics should carry a Berry force²³, but for systems with an even number of electrons and a non-degenerate time-reversal ground state, the Berry force vanishes and BO sets the electronic momentum to zero (incorrectly). As such, within BO theory, in certain cases, the total momentum is no different from the nuclear momentum.

3. 3.

   PS theory easily recovers an accurate VCD signal with only a single perturbation ^{24, 25}, whereas one must perform magnetic field perturbation theory with a double perturbation within BO theory²⁶.

4. 4.

   PS offers improved vibrational energies. Often times, these improvements are small, but in cases of artificially large electronic masses (which break the BO approximation), the improvements can be substantial. ¹⁶

5. 5.

   Polkovnikov et al have shown that PS approaches have analogies in the classical world, offering a new twist on how to derive smooth (rather than noisy) slow variables^{27, 28}.

For the purposes of this paper let us now discuss item 4 in more detail. Starting from a potential energy surface that depends on both and , , there is no obvious path to generate a nuclear quantum Hamiltonian. After all, and do not commute and thus the and in cannot be regarded as quantum operators. That being said, in Ref. 16, in order to extract vibrational energies, we performed a Weyl transform,

and then diagonalized the resulting .

Despite the successes described above (and especially diagonalizing Eq. 9 for vibrational energies ¹⁶), the introduction of a phase space electronic Hamiltonian inevitably raises two criticisms. First, unlike in Eq. 4, there is no unique phase space electronic Hamiltonian because in Eq. 8 is not unique. Second, even though diagonalization of Eq. 8 on a single surface yields better performing vibrational energies than BO theory does, one could argue that diagonalizing Eq. 8 is ad hoc because the result is not systematically improvable. In other words, no one has yet shown how to generate an exact coupled nuclear-electronic eigenfunction of Eq. 1 starting from phase-space electronic states (whereas one can work with BO electronic states for an exact diagonalization).

In what follows below, we will now address the second criticism above. Even without a unique phase space electronic Hamiltonian, we will show below that one can use a phase space electronic Hamiltonian approach as a starting point for an exact diagonalization of . Our approach will follow ideas set forth by Blount²⁹ originally, and later formalized by Littlejohn and Flynn^{30, 31, 32} and Teufel^{33, 34}. The basic idea is to use perturbation theory in tandem within a Wigner representation to generate exact eigenstates with the desired properties. Within the theoretical chemistry community, Wigner-Weyl transforms are very well known in the dynamics community and can be used to develop nonadiabatic dynamics techniques^{35, 36, 37, 38, 39, 40, 41, 42}; here, however, we will show that the same transform can be incredibly helpful for designing new techniques in electronic structure. Thus, in the end, the present paper justifies using a phase space electronic Hamiltonian, putting the latter on a very rigorous (and even practical) footing.

An outline of this paper is as follows. In Sec.2 of this paper, we will review BO theory, showing first how the standard expansion can be derived within Wigner theory and then deriving the Littlejohn-Flynn perturbative expansion. In Sec.3, we will then review phase space electronic structure theory and formally derive a perturbative expansion that allows one to recover an exact eigenstate starting within a phase space framework. In Sec.4, we demonstrate that the present formalism indeed works numerically using a model Hamiltonian. In Sec.5, we discuss two relevant nuances – the mass polarization term and the harmonic limit. We conclude in Sec.6.

Finally, a word about notation. In this manuscript, it will be crucial to distinguish between operators and scalars, as applied to both electronic and nuclear degrees of freedom. In our introduction above, we have used reasonably standard notation whereby all operators (nuclear or electronic) are written with hats. That being said, given that we will work with Wigner transforms below and nuclei will be treated differently from electrons, the manuscript is written for a system with one nuclear degree of freedom (though extensions are obvious). Moreover, henceforward, we will change notation and use hats () exclusively for electronic operators, and boldface () exclusively for nuclear operators. While this notation is not standard, we believe it is the simplest means to communicate the final answer and intuition. Thus, going forward, our single nuclear degrees of freedom has classical phase space coordinates and the corresponding quantum operators are . Lastly, a word about superscripts and perturbation theory. If we perform perturbation theory, one typically writes the final energy as:

For our purposes, it will also be helpful to have a shorthand for the intermediate sums. Thus, we will use the tilde notation to signify such intermediate sums:

and so forth.

For a chemical problem with one nuclear degree of freedom, the adiabatic representation from Eq. 5 above has the following matrix elements:

Note that the sum over states in Eqs. 5 and 14 must extend over a complete set of states for an exact representation of the Hamiltonian.

Now, note that we can consider Eq. 5 formally as an expansion in :

with corresponding matrix elements:

Note that one can develop a slightly different expansion by writing the matrix elements in Eq. 14 as:

where is the second derivative coupling,

Eq. 14 and Eq. 19 are identical as can be shown by using the identities (with be an arbitrary trial function)

and plugging into Eq. 14:

For our purposes, Eq. 14 is preferable to Eq. 19 because, with the former, each of the three terms in the Hamiltonian is hermitian. In any event, the BO approximation arises by neglecting the linear and quadratic terms in :

All of the theory above might have appeared to be unnecessary so far. After all, one can work in the BO representation without Wigner-Weyl transforms; and unless one seeks second order energies, Littlejohn-Flynn theory is equivalent to BO theory. Thus, one might wonder why did we work out all of the calculations above?

The answer, of course, is that for PS theory, where we parameterize electronic states by both nuclear position and nuclear momentum , one is forced to use Wigner-Weyl transforms. Moreover, the LF perturbative expansion above can be easily adapted to a PS reference so as to extract improved eigenvalues, and in this scenario, we do find first order corrections.

Let us begin by defining the phase space electronic Hamiltonian

For now, the operator can be any electronic operator parametrized by nuclear position , . Intuitively, one can consider to be the Legendre transform of if one thinks of as “”. Next, just as the case for BO theory, we diagonalize :

Note that we can also ignore the term and define

which is then diagonalized:

Thus far, we have not focused on the mass polarization term defined by

in Eq. 115. For exact diagonalization calculations above, we have included this term. However, within a phase-space approach, we have not included such a mass polarization term above (for reasons that will soon be clear). In this discussion section we will now address the question of mass polarization in the context of phase space electronic structure theory. In short, for the impatient reader, we find that including mass polarization can change absolute ground state energies but has a much more modest affect on vibrational energies (that require only relative ground state energy information). This finding is encouraging because most calculations will include mass polarization if only to maintain size consistency.

To begin our analysis, in Figs. 5-7, we plot the absolute ground state energy, the relative error in the ground state energy (on an absolute energy scale), and the relative error of the ground state energy (on logarithmic energy scale) when we do not include the mass polarization term; for Figs. 8-10, we plot the same results where we do include the mass polarization term. As far as analyzing these results, one must be careful regarding the observable. If one seeks the most accurate ground state energy, results and give the best answer without the mass polarization term. However, with the mass polarization term, and seem to give the best answer.

As far as scaling is concerned, note that the goes to zero as in Fig. 10; for this method, we include all anharmonicity and mass polarization, while the inaccuracy is dominated by the term (which arises from the second order correction in Eq. 63); for more details, see Ref. 32. Interestingly, for this set of data and for Figs. 8-10, the scaling of the PS data never scales higher than , likely because our second order correction (in Eq. 111) is only approximate.

We have developed a formal but practical theory of phase space electronic Hamiltonians, whereby at zeroth order one diagonalizes an electronic operator parameterized by both and , and thereafter one can continue to improve the resulting eigenvalues perturbatively approaching the exact answer. Unlike the standard BO representation, where the expansion of the total Hamiltonian terminates at second order in , with a phase space formalism, the exact Hamiltonian must be written down as a non-terminating perturbative expansion in (as in Eq. 91) following Littlejohn-Flynn theory. At the end of the day, although the formal theory in Sec. 3 above does not make any statement regarding how to choose a operator, our analysis does show how to mathematically ground the hypothesized phase space electronic Hamiltonians from Ref. 15 within a rigorous framework. Even without a unique , phase space electronic structure theory is not an ad hoc approach.

Turning to our results, in Sec. 4, we have shown that a first order correction to phase space theory can strongly improve vibrational energies – as long as we go beyond the harmonic limit, relative energies can improve by a factor of 10 when we go to first order in , though at the cost of taking derivatives with respect to and . As a practical matter, we are unsure whether this first-order correction will be of interest for larger systems given the need for such derivatives. Another important question surrounds the expansion in Eq. 91. Note that Eq. 91 is perturbative, as can be ascertained from the denominator in Eq. 107. Thus, Eq. 91 is only expected to be accurate in the limit that the ground electronic state is reasonably well separated from the higher electronic states. Looking forward, it will also be crucial in the future to figure out how to rigorously work with non-perturbatively coupled, nearly degenerate systems (e.g. spin systems and systems with multiconfigurational character), where one needs to avoid the denominators in Eq. 107.

In the end, though a great deal of future work remains to be done, the present paper shows how to recover exact quantum eigenvalues starting with phase space electronic states. As such, this paper represents an important numerical validation of the entire phase space approach and is a strong endorsement of developing non-Born Oppenheimer electronic structure theory.