![Arxiv:1811.04069V3 [Quant-Ph] 23 Jan 2020 in Both fields, the Majority of Investigations Have Focused in Both Time Dependent and Independent Contexts](http://data.docslib.org/img/90077052536cd81d65f6c50125139456-1.webp)
Digital quantum simulation of molecular vibrations
Sam McArdle,1 Alexander Mayorov,1, 2 Xiao Shan,3 Simon Benjamin,1 and Xiao Yuan1, ∗ 1Department of Materials, University of Oxford, Parks Road, Oxford OX1 3PH, United Kingdom 2Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge CB2 1EW, United Kingdom 3Physical and Theoretical Chemical Laboratory, University of Oxford, South Parks Road, Oxford OX1 3QZ, United Kingdom (Dated: January 24, 2020) Molecular vibrations underpin important phenomena such as spectral properties, energy transfer, and molecular bonding. However, obtaining a detailed understanding of the vibrational structure of even small molecules is computationally expensive. While several algorithms exist for efficiently solving the electronic structure problem on a quantum computer, there has been comparatively little attention devoted to solving the vibrational structure problem with quantum hardware. In this work, we discuss the use of quantum algorithms for investigating both the static and dynamic vibrational properties of molecules. We introduce a physically motivated unitary vibrational coupled cluster ansatz, which also makes our method accessible to noisy, near-term quantum hardware. We numerically test our proposals for the water and sulfur dioxide molecules.
I. INTRODUCTION be used to treat larger systems, these tend to be less accu- rate than experiments [12]. Although recently proposed Simulating many-body physical systems enables us to analog quantum algorithms [13–19] are capable of sim- study chemicals and materials without fabricating them, ulating molecular vibrations using resources which scale saving both time and resources. The most accurate sim- polynomially with the size of the molecule, the long term ulations require a full quantum mechanical treatment scalability of these approaches has yet to be established. — which is exponentially costly for classical computers. In this work, we discuss a general method for effi- While many approximations have been developed to solve ciently simulating molecular vibrations on a universal this problem, they are often not sufficiently accurate [1]. quantum computer. Our method targets the eigenfunc- One possible route to more accurate simulations is to use tions of a vibrational Hamiltonian with potential terms quantum computers. Quantum computation can enable beyond quadratic order (‘anharmonic potentials’). These us to solve certain problems asymptotically more quickly wavefunctions can then be used to efficiently calculate than with a ‘classical’ computer [2–4]. While the quan- properties of interest, such as absorption spectra at fi- tum computers that we currently possess are small and nite temperatures. We can also use our method to per- error prone, it is hoped that we will one day be able to form simulations of vibrational dynamics, enabling the construct a universal, fault-tolerant quantum computer investigation of properties such as vibrational relaxation. — widely expected to be capable of outperforming its classical counterparts on certain tasks. One example of such tasks is simulating quantum systems on quantum II. VIBRATIONS computers [5–7]. In particular, simulating chemical sys- tems, such as molecules [8], has received significant at- A consequence of quantum mechanics is that molecules tention. This may stem from the commercial benefits are never at rest, possessing at least the vibrational zero- of being able to investigate and design such systems in point energy correction to the ground-state energy [20– silico [9]. The development of quantum computational 22]. As a result, vibrations affect all chemical calcula- chemistry has arguably echoed its classical counterpart. tions, to a greater or lesser extent. They are important
arXiv:1811.04069v3 [quant-ph] 23 Jan 2020 In both fields, the majority of investigations have focused in both time dependent and independent contexts. From on the electronic structure of molecules [10]. This has a dynamics perspective, vibrational structure affects high resulted in a wealth of well established methods for solv- frequency time-resolved laser experiments [23], reaction ing problems of electrons. However, methods concerned dynamics [24–26], and transport [27, 28]. In a static with the nuclear degrees of freedom are comparatively context, vibrations underpin spectral calculations, such less well established. Understanding vibrations is critical as: infrared and Raman spectroscopy [29] and fluores- for obtaining the most accurate models of real physical cence [30]. These calculations determine the performance systems [10]. Unfortunately, the most detailed classical of solar cells [31, 32] and industrial dyes [33, 34], as well as simulations of vibrations are limited to small molecules, the susceptibility of molecules to photodamage [13, 35]. consisting of a few atoms [11]. While approximations can Despite their importance for accurate results, studying vibrations has proven difficult. There are several possible routes to obtaining an accurate description of vibrational behaviour. Real-space, grid based methods, which treat ∗ [email protected] the electronic and nuclear degrees of freedom on an equal 2 footing, are limited to systems of a few particles. While The computational difficulties described above make algorithms to efficiently solve this problem on a universal accurate vibrational calculations on large systems very quantum computer exist [36, 37], it will take many years challenging for classical computers. To overcome these to develop a quantum computer with the required num- challenges, quantum solutions have been suggested for ber of qubits [38]. Alternatively, one may separate the the vibrational structure problem [13–19, 47]. To date, electronic and nuclear degrees of freedom. We can solve the majority of suggestions have focused upon analog for the electronic energy levels of the system as a func- quantum simulation of vibrations. In analog simulations, tion of the nuclear positions, which enables us to map the simulator emulates a specific system of interest, but out potential energy surfaces for the system. A number cannot in general be programmed to perform simulations of approximate classical methods have been developed to of other, different systems. Huh et al. proposed using bo- solve this problem [1, 39], as well as several quantum al- son sampling circuits to determine the absorption spectra gorithms [8, 40–42]. These electronic potential surfaces of molecules [13]. These boson sampling circuits consist can then be viewed as the nuclear potential, determining of photons passing through an optical network. This ini- the vibrational energy levels. This is known as the vibra- tial proposal relied on the harmonic oscillator approxi- tional structure problem. The accuracy of the nuclear mation at zero temperature, but does take into account potential is determined by the accuracy of the electronic bosonic mode mixing due to nuclear structural changes structure calculation, as well as the number of points that result from electronic excitation. This method has obtained for the potential energy surface. Once this po- since been experimentally demonstrated [15, 16], and ex- tential has been obtained, a number of classical methods tended to finite temperature spectra [14, 19]. The main can be used for solving both the time dependent and in- limitation of these simulations is the use of the harmonic dependent Schr¨odingerequations. oscillator approximation for the vibrational wavefunc- The most simple methods uses the ‘harmonic approx- tion. It is in general difficult to engineer ground states imation’. This treats the nuclear potential in the vicin- of anharmonic Hamiltonians using an optical network, ity of the equilibrium geometry as a harmonic oscillator as non-linear operations, such as squeezing, are required. potential, resulting in energy eigenstates which are har- Optical networks have also been used for simulating vi- monic oscillator eigenfunctions. brational dynamics [17]. These simulations investigated vibrational transport, adaptive feedback control, and an- Alternatively, one may consider higher order expan- harmonic effects. sions of the nuclear potential, resulting in more accurate The aforementioned schemes make use of the analogy calculations [43]. One common route towards obtaining between the vibrational energy levels in molecules in the the nuclear potential is to first carry out many electronic Harmonic oscillator approximation, and the bosonic en- structure calculations on the system, in the vicinity of ergy levels accessible to photons and ions. One advantage the minimum energy configuration. Each of these elec- of this is that the bosonic modes are in principle able to tronic structure calculations is approximate, and so the store an arbitrary number of excitations. As these analog cost of each one scales polynomially with the system size. simulators are relatively simple to construct (when com- However, if one proceeds to obtain the nuclear poten- pared to a universal, fault-tolerant quantum computer), tial using this simple grid based method, then a num- they will likely prove useful for small calculations in the ber of grid points scaling exponentially with the num- near-term. However, it is not yet known how to suppress ber of modes is required [44]. In practice one can of- errors to an arbitrarily low rate in analog simulators. As ten instead construct an approximate nuclear potential a result, if we are to simulate the vibrational behaviour of by considering a reduced number of mode couplings, or larger quantum systems, we will likely require error cor- using interpolation, or using adaptive methods. A re- rected universal quantum computers. This motivates our view of these, and other state-of-the-art methods can be work on methods for vibrational simulation on universal found in Ref. [44]. The requirement to first perform mul- quantum computers. tiple electronic structure calculations to obtain the an- The rest of this paper is organised as follows. In harmonic nuclear potential makes calculating vibrational Sec. III, we introduce the vibrational stucture problem energy levels expensive [45], even if only mean-field vibra- for molecules and show how this problem can be mapped tional calculations are then performed. If the correlation onto a quantum computer. In Sec. IV, we show how to between different vibrational modes is included in the solve both static and dynamic problems of molecular vi- calculation, then the simulation becomes even more ex- brations. Finally, in Sec. V, we present the results of pensive. While most of the existing classical vibrational numerical simulations of the H O and SO molecules. simulation methods scale polynomially with the number 2 2 of modes in the system (e.g. vibrational self-consistent field methods [12], or vibrational coupled cluster the- ory [46]), and are sufficiently accurate for some systems, III. ENCODING they only provide approximations to the true full config- uration interaction vibrational wavefunction, which can a. Vibrational Hamiltonian. Under the Born- be exponentially costly to obtain. A similar hierarchy of Oppenheimer approximation, nuclear variables are accuracy also exists for dynamics simulations. treated as parameters in the electronic structure prob- 3 lem and are restored as quantum nuclear variables In contrast, we show below that it is possible to at the level of the full problem. In the following, we efficiently encode the kth order nuclear Hamiltonian neglect the rotational degrees of freedom from negligible into a Hamiltonian acting on qubits. We can then use rotational-vibrational couplings for rigid molecules quantum algorithms to efficiently calculate the static (rigid rotator approximation). After diagonalising the and dynamic properties of the nuclear Hamiltonian. electronic Hamiltonian and neglecting nonadiabatic couplings, the molecular Hamiltonian becomes b. Mapping to qubits We first discuss mapping the molecular Hamiltonian into qubits. We work in the basis X Hmol = |ψsihψs|e ⊗ Hs, (1) of harmonic oscillator eigenstates, as these can be easily s mapped to qubits. The direct mapping presented below was originally suggested in the context of simulating where |ψsie are the electronic energy eigenstates. The general bosonic systems by Somma et al. [48]. It has effective nuclear Hamiltonian Hs is been used recently in the context of quantum simulation of nuclear physics to investigate the binding energy of a p2 Hs = + Vs(q), (2) deuteron nucleus [49]. The compact mapping discussed 2 below was proposed by Veis et al., in the context of using quantum computers to simulate ‘nuclear orbital where q = (q1, q2 ... ) are nuclear coordinates, p = plus molecular orbital (NOMO)’ theory, which uses −i∂/∂q are nuclear momenta with ~ = 1, and Vs(q) is the effective nuclear potential. This potential is deter- Gaussian orbitals for the nuclei, and treats them on an mined by the corresponding electronic potential energy equal footing to the electrons [50]. This differs from our work, which separates the nuclear and electronic surface of |ψsie. As described in Appendix A, we work in mass-weighted normal coordinates and decouple the degrees of freedom, and predominantly considers a harmonic oscillator basis for the vibrational modes. rotational and vibrational modes. The potential Vs(q) can be approximated as While this tailors our method for vibrational problems, it means we are limited to solving problems for which 1 X the Born-Oppenheimer approximation is valid, unlike V (q) ≈ ω2q2, (3) s 2 i i Ref. [50]. i
th Focusing first on one harmonic oscillator, hˆ = ωa†a, where ωi is the harmonic frequency of the i vibrational we consider the truncated eigenstates with the lowest d normal mode. Thus, the nuclear Hamiltonian Hs can be approximated by a sum of independent harmonic oscilla- energies, |si with s = 0, 1, . . . , d − 1. A direct mapping of tors, the space {|si} is to encode it with d qubits as s−1 d−1 X † |si = ⊗j=0 |0ij |1is ⊗j=s+1 |0ij , (6) Hs ≈ ωiai ai, (4) i with creation operator
† d−2 with ai and ai being the creation and annihilation op- √ th † X erator of the i harmonic oscillator. This is the com- a = s + 1 |0ih1|s ⊗ |1ih0|s+1 . (7) monly used ‘harmonic approximation’. Even for accu- s=0 rate potentials and rigid molecules, the harmonic approx- The annihilation operator can be obtained by taking the imation is less accurate than modern spectroscopic tech- Hermitian conjugate of a†. As an alternative to the direct niques [12]. This approximation becomes inadequate for mapping, we can use a compact mapping, using K = large and ‘floppy’ molecules [12]. Improved results can be blog dc qubits, obtained by including anharmonic effects which requires information of higher order potential terms in the Hamil- |si = |bK−1i |bK−2i ... |b0i , (8) tonian [44]. For example, we can expand the potential K−1 K−2 as with binary representation s = bK−12 +bK−22 + 0 . . . b02 . The representation of the creation operator is ∞ X X d−2 Vs(q) = ki1,i2,...,ij qi1 qi2 . . . qij , (5) √ † X j=2 ij a = s + 1 |s + 1ihs| . (9) s=0 where ki ,i ,...,i are the coefficients of the expansion 1 2 j These binary projectors can then be mapped to Pauli qi1 qi2 . . . qij , and the harmonic frequencies are ωi = p operators; 2ki,i. In general, the eigenstates of these Hamiltoni- ans are entangled states, when working in a basis of har- 1 1 monic oscillator eigenstates. Consequently, solving the |0i h0| = (I + Z), |1i h1| = (I − Z), 2 2 (10) higher order vibrational Hamiltonian is a hard problem 1 1 |1i h0| = (X − iY ), |0i h1| = (X + iY ). for classical computers. 2 2 4
A. Vibrational energy levels 250 Direct mapping C H 6 14 Compact mapping An important, but classically difficult problem, is to 200 obtain accurate energy levels for the vibrational Hamil-
C H N O tonian. The spectrum of the vibrational Hamiltonian 5 6 2 2 provides corrections to the electronic eigenstates used to 150 predict reaction rates [22]. Moreover, we will show how these energy levels can be used to calculate the absorp- 100 tion spectrum of molecules in Sec. IV B. Of particular
Number of qubits interest are the lowest lying energy levels at low tem- C H 2 4 perature. Using a universal quantum computer, we can 50 first prepare an initial state that has a large overlap with H O 2 the ground state of the vibrational Hamiltonian. We can 0 then use the phase estimation algorithm [7, 51] to prob- 0 102030405060 abilistically obtain the ground state and ground state Number of modes m energy. A possible initial ground state is the lowest en- ergy product state of the harmonic oscillator basis states, FIG. 1. Number of qubits required for the direct and compact |ψ i = ⊗ |s i . However, we note that the overlap mappings with d = 4 energy levels for each mode. 0 m m m between this state and the true ground state may de- crease exponentially with the size of the molecule. This so-called ‘orthogonality catastrophe’ has been discussed When decomposing a† and a into local Pauli matrices, previously in the context of electronic structure calcula- there are O(d) and O(d2) terms for the direct and com- tions on a quantum computer [52, 53]. As a result, for pact mapping, respectively. In Fig. 1, we show the num- large systems it may be more efficient to use an initial ber of qubits required to describe the vibrational Hamil- state obtained from a classical vibrational self-consistent tonians of several molecules, for both mappings. field (VSCF) calculation. VSCF is the vibrational ana- As p and q can both be represented by a linear logue of the Hartree-Fock method in electronic structure combination of creation and annihilation operators, we theory. VSCF optimises the basis functions to minimise can thus map the nuclear vibrational Hamiltonian to a the energy of the Hamiltonian with a product state. qubit Hamiltonian. If the molecule has n atoms, it has Another route to a state with a large overlap with the M = 3n − 6 vibrational modes for a nonlinear molecule, ground state, is to prepare the VSCF state, and then and M = 3n − 5 for a linear molecule. The vibrational adiabatically evolve under a Hamiltonian that changes wavefunction can then be represented with Md (direct slowly from the VSCF Hamiltonian, to the full vibra- mapping) or M log(d) (compact mapping) qubits. This tional Hamiltonian H. This approach has received sig- can be contrasted with the exponentially scaling classical nificant attention within quantum computing approaches memory required to store the wavefunction. If the poten- to the electronic structure problem, since it was first pro- tial is expanded to kth order (with M k), the Hamilto- posed in the context of quantum computational chem- nian contains O(M kdk) (direct) or O(M kd2k) (compact) istry in Ref. [8]. However, both adiabatic state prepara- terms. These terms are strings of local Pauli matrices. tion and phase estimation typically require long circuits, In this work, we take d to be a small constant. This ap- with a large number of gates. As a result, quantum er- proximation constrains us to the low energy subspace of ror correction is required to suppress the effect of device the Hamiltonian, which should be valid for calculations imperfections. It is therefore helpful to introduce varia- of ground and low-lying excited states. The applicabil- tional methods, which may make these calculations feasi- ity of this approximation to the simulation of dynamics ble for near-term, non-error corrected quantum comput- is discussed in Sec. VI. We set k = 4 to investigate the ers. Variational methods replace the long gate sequences Hamiltonian to quartic order. The resulting Hamiltonian required by phase estimation with a polynomial number has O(M 4) terms. of shorter circuits [42, 54]. This dramatically reduces the coherence time required. As a result, quantum error cor- rection may not be required, if the error rate is sufficiently low, in the context of the number of gates required. The IV. SIMULATING MOLECULAR VIBRATIONS circuits used consist of a number of parametrised gates which seek to create an accurate approximation of the Once the vibrational modes have been mapped to desired state. The parameters are updated using a clas- qubits, we can use quantum algorithms to obtain the sical feedback loop, in order to produce better approxi- static and dynamic properties of the system. We can mations of the desired state. The circuit used is known P as the ‘ansatz’ circuit. write the qubit Hamiltonian as Hs = i λihi, where hi are coefficients determining the strength of each term in Inspired by classical methods for the vibrational struc- the Hamiltonian. ture problem, we introduce the unitary vibrational cou- 5 pled cluster (UVCC) ansatz. This is a unitary analogue B. Franck-Condon factors of the VCC ansatz introduced in Refs. [12, 46]. We note that a similar pairing exists for the electronic struc- In addition to focusing on the eigenstates or thermal ture problem, where the unitary coupled cluster (UCC) states of a single vibrational Hamiltonian, we can also ansatz [55, 56] has been suggested as a quantum ver- consider vibronic (vibrational and electronic) transitions sion of the classical coupled cluster method. The UVCC between the vibrational levels resulting from different ansatz is given by electronic potential energy surfaces. Consider two elec- E tronic states, |ii and |fi . The molecular Hamiltonian Ψ(θ~) = exp Tˆ − Tˆ† |Ψ i , (11) e e 0 is where the initial state |Ψ0i can be either the ground Hmol = |iihi|e ⊗ Hi + |fihf|e ⊗ Hf , (14) state |ψ0i of the harmonic oscillators or the VSCF state ˆ |ΨVSCFi, T is the sum of molecular excitation operators where Hi and Hf are vibrational Hamiltonians, with en- ~ i f E truncated at a specified excitation rank, and θ are the ergy eigenstates ψ and ψ , respectively. Using parameters defined below. Similar to the unitary cou- vib vib Fermi’s Golden Rule, the probability of a photon-induced pled cluster in electronic structure problems, the single i i transition between two wavefunctions ψ = |ii ⊗ ψvib and double excitation operators are E f f and ψ = |fi ⊗ ψvib is proportional to the square Tˆ = Tˆ1 + Tˆ2 + ..., (12) of the transition dipole moment P 2 = | ψf µˆ ψi |2, with using first order time-dependent perturbation theory. M d−1 X X Within the Condon approximation,µ ˆ =µ ˆe +µ ˆN , the Tˆ = θ |s iht | , 2 1 sm,tm m m transition probability becomes proportional to P = m=1 sm,tm=0 D E 2 D E 2 (13) ψf ψi · | hf|µˆ |ii|2. Here ψf ψi are re- M d−1 vib vib e vib vib ˆ X X ferred to as Franck-Condon integrals. Without the Con- T2 = θsm,tm,pm,qn |smpnihtmqn| .
m
The Franck-Condon integrals without the Condon ap- proximation can be efficiently calculated via 0.04 0.04 D 2 D 2 f i f f ˆ i ψvib µˆ(q) ψvib = Ψvib µˆ(q )UDok Ψvib . (17) 0.035 0.035 They can be both efficiently calculated with the gener- alised SWAP-test circuit [65]. Alternatively, we can obtain the Franck-Condon in- 0.03 0.03 tegrals without realising the Doktorov transform. The
E Energy (hartree) Energy (hartree) i f i qubit states Ψvib and Ψvib are obtained from Hi(q ) f i f 0.025 0.025 and Hf (q ) with normal mode coordinates q and q , respectively. Instead, we can focus on one set of nor- mal mode coordinates qi and represent the Hamiltonian i 0 i 0.02 0.02 Hf in q , Hf (q ). By solving the energy eigenstates of Two levels Four levels 0 E E 0 i f ˆ † f Hf (q ), we can directly get Ψvib = UDok Ψvib and calculate the Franck-Condon integrals without realising FIG. 2. Vibrational spectra of H2O with two and four energy the Doktorov transform. However, as the Hamiltonian levels for each mode. The solid lines are the energy levels of 0 i Hf (q ) is not encoded in the correct normal mode ba- the harmonic oscillator eigenstates and the dashed lines are sis, the ground state of the harmonic oscillators or the the vibrational spectra of the Hamiltonian with a fourth order VSCF state |ΨVSCFi may not be an ideal initial state expansion of the potential. to start with. However, this effect may be negligible if 0 E the overlap between Ψi and Ψ f is suitably large. vib vib qubitization [70, 71] in conjunction with quantum sig- i In this case, the initial state |Ψ0i for Ψvib should also nal processing [72, 73]. The product formula method is 0 E L be an ideal initial state for Ψ f . The aforementioned the most simple to realise. If H can be decomposed as vib L HL = P λLhL, the time evolution operator e−iH t can transformation can be implemented by transforming the i i i normal mode coordinates qi as described in Ref. [45]. be realised using a product formula, !N L L L −iH t Y −iλi Hi t/N 2 C. Vibrational dynamics e = e + O(t /N) (20) i In this section, we consider methods to investigate when N is chosen to be sufficiently large to suppress the the dynamic properties of vibrational Hamiltonians. Vi- error in the approximation. brational dynamics underpin phenomena including en- Alternatively, the vibrational dynamics can be realised ergy and electron transport [27, 28] and chemical reac- using a recently proposed variational algorithm [74]. tions [24–26]. Dynamical behaviour can be studied by One could use either a UVCC ansatz, or a Trotterized transforming to a single-mode basis of spatially localised ansatz [75, 76]. vibrational modes, as described in Ref. [17]. The spa- L tially localised vibrational modes ai , are related to the normal modes ai via a basis transformation V. NUMERICAL SIMULATIONS X a = U aL, (18) i i,j j In this section, we demonstrate how the techniques de- i,j scribed above can be used to calculate the vibrational with real unitary matrix Ui,j. We can obtain the corre- energy levels of small molecules. We focus on the poly- L sponding localised Hamiltonian H , using the transfor- atomic molecules H2O and SO2, which both have three mation of the normal coordinates and momenta vibrational modes. We considered both nuclear poten- tials expanded to fourth order, the coefficients of which r r X ωi X ωj p = U pL, q = U qL. (19) are shown in Table I in Appendix D. We consider the i ω i,j j i ω i,j j j j j i cases with two and four energy levels for each of the har- monic oscillator modes, yielding Hamiltonians acting on Given an initial state of the localised vibrations, the dy- 6 and 12 qubits for the direct mapping, and 3 and 6 qubits namics can be simulated by applying the time evolution for the compact mapping. We used the compact mapping L operator e−iH t. This can be achieved in a number of in our numerical simulations, as it requires fewer qubits. ways, using different Hamiltonian simulation algorithms, There are 216 and 165 terms in the Hamiltonian for H2O including: Trotterization (also referred to as product for- and SO2, respectively. mulae) [5, 66], the Taylor series method [67–69], and We first calculate the energy levels under the harmonic 7
Y 0.032 |0i0 R0 XY YX UVCC R01 R01 Y XY YX Exact energy |0i1 R1 R02 R02 XY YX 0.03 R12 R12 Y |0i2 R2 0.028 FIG. 3. The UVCC ansatz of three modes each with two en- ergy levels. There are nine gates with six parameters (joined 0.026 gates share the same parameters). The single qubit gate on
i Energy (hartree) th Y −iθσy the i qubit is Ri (θ) = e , and the two qubit gate on i j 0.024 th th AB −iθ(σA⊗σ ) the i and j is Rij (θ) = e B .
0.022 approximation. We compare this to the energy levels ob- 0 10203040506070 tained with a fourth order expansion of the potential. Iterations The results for H2O are shown in Fig. 2. We can see that although the ground state can be well approximated by FIG. 4. Solving the vibrational ground state of H2O with the the harmonic oscillators, the excited states deviate from UVCC ansatz. Here, we consider two energy levels for each the harmonic oscillators at higher energy levels. The re- mode. sults for SO2 can be found in the Appendix. These cal- culations highlight the importance of anharmonic terms in the potential for even small molecules. such as the ground state energy. However, it may not Next, we implemented the UVCC ansatz to obtain the always be possible to use this approximation when con- sidering time evolution, as this requires the exponentia- vibrational energy levels of H2O using the variational quantum eigensolver [42]. For simplicity, we considered tion of our truncated Hamiltonian. One possible route two energy levels for each mode. To implement the to overcome this challenge is to repeat simulations which UVCC ansatz, we first calculate the imaginary part of consider an increasing number of energy levels, and then Tˆ and encode it into a linear combination of local Pauli to extrapolate to the infinite energy level result. This terms, i.e., technique was used in a similar context to this work in Ref. [49]. X Tˆ − Tˆ† = i α (θ)σ . (21) Once the vibrational Hamiltonian has been mapped to i i a qubit Hamiltonian, much of the existing machinery for i quantum simulation can be applied. Static properties, such as energy levels, can be calculated using phase es- Then, as for the UCC ansatz, we realise exp Tˆ − Tˆ† timation or variational approaches. To aid variational Q by a first order Trotterisation via i exp(iαi(θ)σi). For state preparation, we proposed a unitary version of the example, the UVCC ansatz of H2O with two energy levels powerful VCC method used in classical vibrational sim- can be prepared by the circuit in Fig. 3. ulations. The resulting energy eigenstates can be used Using the UVCC ansatz, we can obtain the vibrational as an input for SWAP-test circuits. These calculate the ground state with a variational procedure. As the ground Franck-Condon factors for the molecules, which are re- ⊗3 state is close to the initial state |0i , we start with pa- lated to the absorption spectra. Alternatively, one may rameters slightly perturbed from zero. We then use gra- investigate dynamic properties, using methods for Hamil- dient descent to find the minimum energy of the system. tonian simulation to time evolve a specified state. The results are shown in Fig. 4. Compared with analog algorithms [13, 14, 16, 17], our method can easily take into account anharmonic terms in the nuclear Hamiltonian. Moreover, it could be used VI. DISCUSSION to simulate large systems by protecting the quantum computer with error correction. As our technique is In this work, we have extended many of the techniques tailored for simulating vibrational states, it makes it developed for quantum simulation to the problem of sim- simple to investigate interesting vibrational properties, ulating vibrations. Understanding vibrations is an im- such as the Franck-Condon factors. However, as our portant problem for accurately modelling chemical sys- approach uses the Born-Oppenheimer approximation, tems, yet one that is difficult to solve classically. it is not suitable for all problems in chemistry, such We have discussed ways to map between vibrational as problems including relativistic effects [77] or conical modes and qubit states. This is only possible when we intersections [78–80]. Future work will address whether consider a restricted number of harmonic oscillator en- restricting our vibrational modes to low-lying energy ergy levels. This approximation is appropriate when in- levels poses a significant challenge for problems of vestigating the low energy properties of the Hamiltonian, practical interest. 8
first step of their algorithm. As such, our two works are highly complementary, and can be compared.
NOTE ADDED
After this work was released as a preprint, a notable ACKNOWLEDGEMENTS. related work was released online [81]. That paper provides a quantum algorithm for efficiently calculating We acknowledge insightful comments from Joonsuk the entire converged Franck-Condon profile, as opposed Huh. This work was supported by BP plc and the EP- to the method we present herein, which calculates a SRC National Quantum Technology Hub in Networked single Franck-Condon factor. However, the techniques Quantum Information Technology (EP/M013243/1). X. we have discussed for finding vibrational ground states Shan acknowledges the use of the University of Oxford and thermal states provide a way to realise the crucial Advanced Research Computing (ARC) facility.
[1] T. Helgaker, P. Jorgensen, and J. Olsen, Molecular [20] R. P. de Tudela, F. J. Aoiz, Y. V. Suleimanov, and D. E. electronic-structure theory (John Wiley & Sons, 2014). Manolopoulos, J. Phys. Chem. Lett. 3, 493 (2012). [2] R. P. Feynman, International journal of theoretical [21] S. Karazhanov, M. Ganchenkova, and E. Marstein, physics 21, 467 (1982). Chem. Phys. Letters 601, 49 (2014). [3] P. W. Shor, in Foundations of Computer Science, 1994 [22] A. Gross and M. Scheffler, J. Vac. Sci. Technol. A: Vac- Proceedings., 35th Annual Symposium on (Ieee, 1994) pp. uum, Surfaces, and Films 15, 1624 (1997). 124–134. [23] T. Seideman, Forming Superposition States, in Computa- [4] L. K. Grover, in Proceedings of the twenty-eighth annual tional Molecular Spectroscopy (Chichester : Wiley, 2000) ACM symposium on Theory of computing (ACM, 1996) pp. 589–624. pp. 212–219. [24] F. F. Crim, Proc. Natl. Acad. Sci. 105, 12654 (2008). [5] S. Lloyd, Science 273, 1073 (1996). [25] D. Antoniou and S. D. Schwartz, J. Phys. Chem. Lett. B [6] D. S. Abrams and S. Lloyd, Phys. Rev. Lett. 79, 2586 105, 5553 (2001). (1997). [26] D. L. Proctor and H. F. Davis, Proc. Natl. Acad. Sci. [7] D. S. Abrams and S. Lloyd, Phys. Rev. Lett. 83, 5162 105, 12673 (2008). (1999). [27] R. Borrelli, M. D. Donato, and A. Peluso, Theor Chem [8] A. Aspuru-Guzik, A. D. Dutoi, P. J. Love, and M. Head- Acc. 117, 957 (2006). Gordon, Science 309, 1704 (2005). [28] H. Hwang and P. J. Rossky, J. Phys. Chem. A 108, 2607 [9] A. Aspuru-Guzik, R. Lindh, and M. Rei- (2004). her, ACS Central Science 4, 144 (2018), [29] C. Zhu, K. K. Liang, M. Hayashi, and S. H. Lin, Chem. https://doi.org/10.1021/acscentsci.7b00550. Phys. 358, 137 (2009). [10] O. Christiansen, Phys. Chem. Chem. Phys. 9, 2942 [30] T.-W. Huang, L. Yang, C. Zhu, and S. H. Lin, Chem. (2007). Phys. Letters 541, 110 (2012). [11] J. M. Bowman, T. Carrington, and H.- [31] S.-Y. Yue, X. Zhang, G. Qin, J. Yang, and M. Hu, Phys. D. Meyer, Mol. Phys. 106, 2145 (2008), Rev. B 94, 115427 (2016). https://doi.org/10.1080/00268970802258609. [32] L. Debbichi, M. C. Marco de Lucas, J. F. Pierson, [12] O. Christiansen, J. Chem. Phys. 120, 2140 (2004), and P. Krger, J. Phys. Chem. C 116, 10232 (2012), https://doi.org/10.1063/1.1637578. https://doi.org/10.1021/jp303096m. [13] J. Huh, G. G. Guerreschi, B. Peropadre, J. R. McClean, [33] S. Dhananasekaran, R. Palanivel, and S. Pappu, J. Adv. and A. Aspuru-Guzik, Nat Photonics 9, 615 (2015). Res. 7, 113 (2016). [14] J. Huh and M.-H. Yung, Scientific reports 7, 7462 (2017). [34] N. Biswas and S. Umapathy, J. Phys. Chem. A 101, 5555 [15] W. R. Clements, J. J. Renema, A. Eckstein, A. A. (1997), https://doi.org/10.1021/jp970312x. Valido, A. Lita, T. Gerrits, S. W. Nam, W. S. Koltham- [35] K.-W. Choi, J.-H. Lee, and S. K. Kim, J. Am. mer, J. Huh, and I. A. Walmsley, arXiv preprint Chem. Soc. 127, 15674 (2005), pMID: 16277488, arXiv:1710.08655 (2017). https://doi.org/10.1021/ja055018u. [16] Y. Shen, Y. Lu, K. Zhang, J. Zhang, S. Zhang, J. Huh, [36] I. Kassal, S. P. Jordan, P. J. Love, M. Mohseni, and and K. Kim, Chem. Sci. 9, 836 (2018). A. Aspuru-Guzik, Proc. Natl. Acad. Sci. 105, 18681 [17] C. Sparrow, E. Mart´ın-L´opez, N. Maraviglia, A. Neville, (2008). C. Harrold, J. Carolan, Y. N. Joglekar, T. Hashimoto, [37] I. D. Kivlichan, N. Wiebe, R. Babbush, and A. Aspuru- N. Matsuda, J. L. OBrien, et al., Nature 557, 660 (2018). Guzik, J. Phys. A: Math. Theor. 50 (2017). [18] S. Chin and J. Huh, arXiv preprint arXiv:1803.10002 [38] N. C. Jones, J. D. Whitfield, P. L. McMahon, M.-H. (2018). Yung, R. Van Meter, A. Aspuru-Guzik, and Y. Ya- [19] L. Hu, Y.-C. Ma, Y. Xu, W.-T. Wang, Y.-W. Ma, K. Liu, mamoto, New J. Phys. 14, 115023 (2012). H.-Y. Wang, Y.-P. Song, M.-H. Yung, and L.-Y. Sun, Sci. [39] A. Szabo and N. S. Ostlund, Modern quantum chem- Bull. 63, 293 (2018). istry: introduction to advanced electronic structure theory (Courier Corporation, 2012). 9
[40] R. Babbush, D. W. Berry, I. D. Kivlichan, A. Y. Wei, (2018), arXiv:1803.04114. P. J. Love, and A. Aspuru-Guzik, New J. Phys. 18, [66] H. F. Trotter, Proc. Amer. Math. Soc. 10, 545 (1959). 033032 (2016). [67] D. W. Berry and A. M. Childs, Quantum Info. Comput. [41] R. Babbush, D. W. Berry, Y. R. Sanders, I. D. Kivlichan, 12, 29 (2012). A. Scherer, A. Y. Wei, P. J. Love, and A. Aspuru-Guzik, [68] D. W. Berry, A. M. Childs, R. Cleve, R. Kothari, and Quantum Sci. Technol. 3, 015006 (2017). R. D. Somma, Phys. Rev. Lett. 114, 090502 (2015). [42] A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. [69] D. W. Berry, A. M. Childs, and R. Kothari, in 2015 Zhou, P. J. Love, A. Aspuru-Guzik, and J. L. Obrien, IEEE 56th Annual Symposium on Foundations of Com- Nat. Commun. 5 (2014). puter Science (2015) pp. 792–809. [43] O. Christiansen, Phys. Chem. Chem. Phys. 9, 2942 [70] G. H. Low and I. L. Chuang, arXiv:1610.06546 (2016). (2007). [71] G. H. Low, arXiv preprint arXiv:1807.03967 (2018). [44] O. Christiansen, Phys. Chem. Chem. Phys. 14, 6672 [72] G. H. Low, T. J. Yoder, and I. L. Chuang, Phys. Rev. (2012). X 6, 041067 (2016). [45] J. Huh, PhD Thesis (2011). [73] G. H. Low and I. L. Chuang, Phys. Rev. Lett. 118, [46] O. Christiansen, J. Chem. Phys. 120, 2149 (2004), 010501 (2017). https://doi.org/10.1063/1.1637579. [74] Y. Li and S. C. Benjamin, Phys. Rev. X 7, 021050 (2017). [47] S. Joshi, A. Shukla, H. Katiyar, A. Hazra, and T. S. [75] J. O’Gorman and E. T. Campbell, Phys. Rev. A 95, Mahesh, Phys. Rev. A 90, 022303 (2014). 032338 (2017). [48] R. D. Somma, G. Ortiz, E. H. Knill, and J. Gubernatis, [76] T. Jones and S. C. Benjamin, arXiv:1811.03147 (2018). Proc.SPIE 5105, 5105 (2003). [77] M. Reiher and A. Wolf, Relativistic quantum chemistry: [49] E. F. Dumitrescu, A. J. McCaskey, G. Hagen, G. R. the fundamental theory of molecular science (John Wiley Jansen, T. D. Morris, T. Papenbrock, R. C. Pooser, D. J. & Sons, 2014). Dean, and P. Lougovski, Phys. Rev. Lett. 120, 210501 [78] W. Domcke, D. R. Yarkony, and H. Kppel, (2018). Conical Intersections (WORLD SCIENTIFIC, 2004) [50] L. Veis, J. Vik, H. Nishizawa, H. Nakai, and J. Pit- https://www.worldscientific.com/doi/pdf/10.1142/5406. tner, Int. J. Quantum Chem. 116, 1328 (2016), [79] W. Domcke and D. R. Yarkony, Annu. Rev. https://onlinelibrary.wiley.com/doi/pdf/10.1002/qua.25176. Phys. Chem. 63, 325 (2012), pMID: 22475338, [51] A. Y. Kitaev, Preprint at http://arxiv. org/abs/quant- https://doi.org/10.1146/annurev-physchem-032210- ph/9511026 (1995). 103522. [52] J. R. McClean, R. Babbush, P. J. Love, and A. Aspuru- [80] I. G. Ryabinkin, L. Joubert-Doriol, and A. F. Izmaylov, Guzik, J. Phys. Chem. Lett. 5, 4368 (2014), pMID: Acc. Chem. Res. 50, 1785 (2017), pMID: 28665584, 26273989, https://doi.org/10.1021/jz501649m. https://doi.org/10.1021/acs.accounts.7b00220. [53] N. M. Tubman, C. Mejuto-Zaera, J. M. Epstein, D. Hait, [81] N. P. D. Sawaya and J. Huh, (2018), arXiv:1812.10495. D. S. Levine, W. Huggins, Z. Jiang, J. R. McClean, [82] J. T. Seeley, M. J. Richard, and P. J. R. Babbush, M. Head-Gordon, and K. B. Whaley, Love, J. Chem. Phys. 137, 224109 (2012), arXiv:1809.05523 (2018). https://doi.org/10.1063/1.4768229. [54] J. R. McClean, J. Romero, R. Babbush, and A. Aspuru- [83] M. Frisch, http://www. gaussian. com/ (2009). Guzik, New J. Phys. 18, 023023 (2016). [55] M.-H. Yung, J. Casanova, A. Mezzacapo, J. McClean, L. Lamata, A. Aspuru-Guzik, and E. Solano, Scientific reports 4, 3589 (2014). [56] J. Romero, R. Babbush, J. McClean, C. Hempel, P. Love, and A. Aspuru-Guzik, Quantum Sci. Technol. (2018), 10.1088/2058-9565/aad3e4. [57] E. B. Wilson, J. C. Decius, and P. C. Cross, Molecular vibrations: the theory of infrared and Raman vibrational spectra (Courier Corporation, 1980). [58] K. Temme, T. J. Osborne, K. G. Vollbrecht, D. Poulin, and F. Verstraete, Nature 471, 87 (2011). [59] M.-H. Yung and A. Aspuru-Guzik, Proc. Natl. Acad. Sci. 109, 754 (2012). [60] S. McArdle, T. Jones, S. Endo, Y. Li, S. Benjamin, and X. Yuan, arXiv:1804.03023 (2018). [61] X. Yuan, S. Endo, Q. Zhao, S. Benjamin, and Y. Li, (2018), arXiv:1812.08767. [62] M. Motta, C. Sun, A. T. K. Tan, M. J. O. Rourke, E. Ye, A. J. Minnich, F. G. S. L. Brandao, and G. K.-L. Chan, (2019), arXiv:1901.07653. [63] H. Kupka and P. H. Cribb, J. Chem. Phys. 85, 1303 (1986). [64] E. Doktorov, I. Malkin, and V. Man’ko, Journal of Molecular Spectroscopy 64, 302 (1977). [65] L. Cincio, Y. Suba, A. T. Sornborger, and P. J. Coles, “Learning the quantum algorithm for state overlap,” 10
Appendix A: Encoding vibrational Hamiltonians into qubits
The molecular Hamiltonian in atomic units is
2 2 X ∇i X ∇I X ZI Hmol = − − − 2 2MI |ri − RI | i I i,I (A1) 1 X 1 1 X ZI ZJ + + . 2 |ri − rj| 2 |RI − RJ | i6=j I6=J where MI , RI , and ZI are the mass, position, and charge of nuclei I, and ri is the position of electron i. Given the location of the nucleus, the electronic Hamiltonian is
2 X ∇i X ZI 1 X 1 He(RI ) = − + + , (A2) 2 |ri − RI | 2 |ri − rj| i i,I i6=j and the total Hamiltonian is
2 X ∇I 1 X ZI ZJ Hmol = − + + He(RI ). (A3) 2MI 2 |RI − RJ | I I6=J
Under the Born-Oppenheimer approximation, we assume the electrons and nuclei are in a product state,
|ψi = |ψin |ψie . (A4)
To get the ground state of the Hamiltonian, one can thus separately minimise over |ψin and |ψie,
E0 = min min hψ|n hψ|e Hmol |ψin |ψie . (A5) |ψin |ψie
As only He(RI ) depends on |ψie, the minimisation over |ψie is equivalent to finding the ground state of He(RI ). Denote
e V0 (RI ) = min hψ|e He(RI ) |ψie , (A6) |ψie then the ground state of Hmol can be found by solving the ground state of H0,
2 X ∇I 1 X ZI ZJ e H0 = − + + V0 (RI ). (A7) 2MI 2 |RI − RJ | I I6=J
P e In general, considering a spectral decomposition of He(RI ) = s Vs (RI ) |ψsihψs|e, the molecular Hamiltonian is X Hmol = |ψsihψs|e ⊗ Hs. (A8) s
Here, |ψsie are eigenstates of the electronic Hamiltonian and
2 X ∇I Hs = − + Vs(RI ), (A9) 2MI I and
1 X ZI ZJ e Vs(RI ) = + Vs (RI ). (A10) 2 |RI − RJ | I6=J
Finding the spectra of He(RI ) is called the electronic structure problem, which can be efficiently solved using a quantum computer [8]. One approach is to consider a subspace that the ground state lies in and transform the Hamiltonian He(RI ) into the second quantised formulation, with a basis determined by the subspace. As electrons are fermions, the obtained Hamiltonian is a fermionic Hamiltonian. By using the standard encoding methods, such 11 as Jordan-Wigner and Bravyi-Kitaev [82], the fermionic Hamiltonian is converted into a qubit Hamiltonian, whose spectra can be efficiently computed. Focusing on the ground state of the electronic structure Hamiltonian, we show how to redefine H0 in the mass- 1 P ZI ZJ e weighted basis and how to encode it with qubits. Denote V = + V (RI ), then one can obtain the 2 I6=J |RI −RJ | 0 mass-weighted normal coordinates qi by minimising the coupling between the rotational and vibrational degrees of freedom and diagonalising the Hessian matrix, 1 ∂2V HIJ = √ . (A11) MI MJ ∂RI ∂RJ In the mass-weighted basis, the potential can be expanded via a Taylor series truncated at fourth order 1 X X X V (4) = ω2q2 + f q q q + f q q q q , (A12) 2 i i ijk i j k ijkl i j k l i i≤j≤k i≤j≤k≤l and the total Hamiltonian becomes X ∇2 H = − i + V (4). (A13) 0 2 i If we consider the higher order terms as a perturbation, one can get the normal modes by solving the harmonic oscillator ∇2 1 hˆ = − i + ω2q2. (A14) i 2 2 i i ˆ i i We denote the eigenbasis for hi as { ψsi , ∀s = 0, 1,... }, then the nuclear wave function can be represented by
X 1 2 M |ψi = αs1s2...sM ψs1 ψs2 . . . ψsM . (A15) s1s2...sM The Hamiltonian under the normal mode basis becomes,
1 2 M 1 2 M Hs1s2...sM ,t1t2...tM = ψs1 ψs2 . . . ψsM H0 ψt1 ψt2 . . . ψtM . (A16)
i i Suppose the basis { ψsi , ∀s = 0, 1,... } is truncated to the lowest d energy levels, the space of Hs1s2...sM ,t1t2...tM is equivalent to the space of M d-level systems, or equivalently M log2 d qubits.
Appendix B: Variational quantum simulation with the unitary vibrational coupled cluster ansatz
We can make use of variational methods to find the low energy spectra of the vibrational Hamiltonian. Inspired by classical computational chemistry, we introduce the unitary vibrational coupled cluster (UVCC) ansatz
† |VCCi = exp Tˆ − Tˆ |Φ0i (B1) where the reference state |Φii is a properly chosen initial state, and T is the sum of molecular excitation operators truncated at a specified excitation rank (often single and double excitations) [56]
Tˆ = Tˆ1 + Tˆ2 + .... with
M ˆ X X m m m T1 = tam,im |ψam ihψim | , m am,im
M ˆ X X X m,n m n m n T2 = tam,an,im,jn |ψam ψan ihψim ψin | . m The initial state can be the product of the ground-state of each mode 1 2 M |Φ0i = ψ0ψ0 . . . ψ0 . (B2) Or we can also run a vibrational self-consistent field (VSCF) to get the Hartree-Fock initial state 1 2 M |Φ0i = φ φ . . . φ , (B3) which is obtained by minimising the energy of the Hamiltonian 1 2 M 1 2 M min φ φ . . . φ H0 φ φ . . . φ (B4) |φ1φ2...φM i by solving the self-consistent equation Hi |φii = Ei |φii , (B5) 1 i−1 i+1 M 1 i−1 i+1 M with Hi = φ . . . φ φ . . . φ H0 φ . . . φ φ . . . φ . Appendix C: Duschinsky transform The relation between the initial coordinates q1 and final coordinates q2 is q1 = Uq2 + d, (C1) where U is the Duschinsky rotation matrix and d is the displacement vector. The harmonic oscillator Hamiltonian with unit mass is p2 1 hˆ = − i + ω2q2, (C2) i 2 2 i i ∂ with pi = . The operators q1 and p1 can be represented by the creation and annihilation operators ∂qi r r Ω ~ † ~ 1 † (C3) q1 = a1 + a1 , p1 = i a1 − a1 . 2Ω1 2 † The transformation for p is p1 = U p2. The transformation for the creation operators are r † Ω1 i a1 = q1 − p1 , 2~ Ω1 r Ω1 i † = Uq2 + d − U p2 , 2~ Ω1 ! r r r r Ω1 ~ † i † ~Ω2 † Ω1 (C4) = U a2 + a2 − U i a2 − a2 + d, 2~ 2Ω2 Ω1 2 2~ r 1p p −1 † 1p −1 †p † Ω1 = Ω1U Ω2 a2 + a2 + Ω1 U Ω2 a2 − a2 + d, 2 2 2~ r 1 0 1 0 † Ω1 = (J − J )a2 + (J + J )a2 + d, 2 2 2~ √ √ √ √ −1 0 −1 † where J = Ω1U Ω2 and J = Ω1 U Ω2. It was shown by Doktorov et al. [64] that the Duschinsky transform can be implemented using a unitary transform inserted into the overlap integral 0 ˆ hνf | |νii = νf UDok |νii (C5) where |νi is a harmonic oscillator eigenstate in the initial coordinate q and |ν0i is a harmonic oscillator eigenstate in the final coordinate q0. The Doktorov unitary can be decomposed into a product of unitaries, which depend on the 13 ~ displacement vector d, the rotation matrix U, and matrices of the eigenenergies of the harmonic oscillator states (i, 0 √ 0 p 0 i); Ω = 1/~ diag( i) and Ω = 1/~ diag( i). It is given by ˆ ˆ ˆ † ˆ ˆ UDok = UtUs0 UsUr (C6) where 1 ~t 0 † Uˆt = exp d Ω (~a − ~a) (C7) 2~ T where ~a = (a0, ..., aM ) , and 1 † t 0 † 1 0 Uˆ 0 = exp − (~a + ~a) ln(Ω )(~a − ~a) + Tr(ln(Ω )) (C8) s 2 2 and 1 1 Uˆ = exp − (~a† + ~a)tln(Ω)(~a† − ~a) + Tr(ln(Ω)) (C9) s 2 2 and 1 Uˆ = exp ((~a†)tln(U)~a − (~a)tln(U)~a† . (C10) r 2 These exponentials could be expanded into local qubit operators using Trotterization. It is important to note that these relations are only valid when the single-mode basis functions are chosen to be harmonic oscillator eigenstates. Appendix D: Numerical simulation The coefficients of H2O and SO2 are shown in Table I, which were computed at MP2/aug-cc-pVTZ level of theory using Gaussian09 software [83]. The simulation results of the vibrational energy levels of SO2 are shown in Fig. 5. TABLE I. Coefficients of the potential energy surface of H2O and SO2. The coefficients are in the atomic units, where the unit −10 of length is a0 = 1 Bohr (0.529167 × 10 m), the unit of mass is the electron mass me, and the unit of energy is 1 Hartree 2 (1 Hartree = e /4π0a0 = 27.2113 eV). k H2O SO2 −4 −5 k1,1 0.275240 × 10 0.252559 × 10 −3 −4 k2,2 0.151618 × 10 0.125410 × 10 −3 −4 k3,3 0.161766 × 10 0.176908 × 10 −6 −8 k1,1,1 0.121631 × 10 0.316646 × 10 −6 −8 k1,1,2 0.698476 × 10 0.575325 × 10 −6 −7 k1,2,2 −0.266427 × 10 0.197771 × 10 −5 −7 k2,2,2 −0.312538 × 10 −0.668689 × 10 −6 −9 k1,3,3 −0.915428 × 10 −0.370850 × 10 −5 −6 k2,3,3 −0.964649 × 10 −0.284244 × 10 −9 −11 k1,1,1,1 −0.463748 × 10 0.330842 × 10 −7 −9 k1,1,2,2 −0.449480 × 10 −0.172869 × 10 −8 −9 k1,2,2,2 0.957558 × 10 −0.215928 × 10 −7 −9 k2,2,2,2 0.433267 × 10 0.225400 × 10 −7 −9 k1,1,3,3 −0.555026 × 10 −0.356155 × 10 −7 −9 k1,2,3,3 0.563566 × 10 −0.128135 × 10 −6 −8 k2,2,3,3 0.269239 × 10 0.220168 × 10 −7 −9 k3,3,3,3 0.462143 × 10 0.458046 × 10 −11 k2,3,3,3 0 −0.720760 × 10 14 10-3 10-3 13 12 12 11 11 10 10 9 9 Energy (hartree) Energy (hartree) 8 8 7 7 6 6 Two levels Four levels FIG. 5. Vibrational spectra of SO2 with two and four energy levels for each mode. The solid lines are the energy levels of the harmonic oscillator eigenstates and the dashed lines are the vibrational spectra of the Hamiltonian with a fourth order expansion of the potential.