9.10.6 Quasi-Classical Molecular Dynamics

So-called “quasi-classical” trajectories ⁶⁰² Karplus M., Porter R. N., Sharma R. D.
J. Chem. Phys.
(1965), 43, pp. 3259. Link ^{, 1006} Porter R.
Annu. Rev. Phys. Chem.
(1974), 25, pp. 317. Link ^{, 1005} Porter R., Raff L., Miller W. H.
J. Chem. Phys.
(1975), 63, pp. 2214. Link (QCT) put vibrational energy into each mode in the initial velocity setup step, which can improve on the results of purely classical simulations, for example in the calculation of photoelectron ⁶⁷⁶ Lambrecht D. S. et al.
J. Phys. Chem. A
(2011), 115, pp. 5928. Link or infrared spectra. ¹⁰²⁹ Ramos-Cordoba E., Lambrecht D. S., Head-Gordon M.
Faraday Discuss.
(2011), 150, pp. 345. Link Improvements include better agreement of spectral linewidths with experiment at lower temperatures and better agreement of vibrational frequencies with anharmonic calculations.

The improvements at low temperatures can be understood by recalling that even at low temperature there is nuclear motion due to zero-point motion. This is included in the quasi-classical initial velocities, thus leading to finite peak widths even at low temperatures. In contrast to that the classical simulations yield zero peak width in the low temperature limit, because the thermal kinetic energy goes to zero as temperature decreases. Likewise, even at room temperature the quantum vibrational energy for high-frequency modes is often significantly larger than the classical kinetic energy. QCT-MD therefore typically samples regions of the potential energy surface that are higher in energy and thus more anharmonic than the low-energy regions accessible to classical simulations. These two effects can lead to improved peak widths as well as a more realistic sampling of the anharmonic parts of the potential energy surface. However, the QCT-MD method also has important limitations which are described below and that the user has to monitor for carefully.

In our QCT-MD implementation the initial vibrational quantum numbers are generated as random numbers sampled from a vibrational Boltzmann distribution at the desired simulation temperature. In order to enable reproducibility of the results, each trajectory (and thus its set of vibrational quantum numbers) is denoted by a unique number using the AIMD_QCT_WHICH_TRAJECTORY variable. In order to loop over different initial conditions, run trajectories with different choices for AIMD_QCT_WHICH_TRAJECTORY. It is also possible to assign initial velocities corresponding to an average over a certain number of trajectories by choosing a negative value. Further technical details of our QCT-MD implementation are described in detail in Appendix A of Ref.  676 Lambrecht D. S. et al.
J. Phys. Chem. A
(2011), 115, pp. 5928. Link .

Below is a simple example input for running a QCT-MD simulation of the vibrational spectrum of water. Most input variables are the same as for classical MD as described above. The use of quasi-classical initial conditions is triggered by setting the AIMD_INIT_VELOC variable to QUASICLASSICAL.

Other types of spectra can be calculated by calculating spectral properties along the trajectories. For example, we observed that photoelectron spectra can be approximated quite well by calculating vertical detachment energies (VDEs) along the trajectories and generating the spectrum as a histogram of the VDEs. ⁶⁷⁶ Lambrecht D. S. et al.
J. Phys. Chem. A
(2011), 115, pp. 5928. Link We have included several simple scripts in the $QC/aimdman/tools subdirectory that we hope the user will find helpful and that may serve as the basis for developing more sophisticated tools. For example, we include scripts that allow to perform calculations along a trajectory (md_calculate_along_trajectory) or to calculate vertical detachment energies along a trajectory (calculate_rel_energies).

Another application of the QCT code is to generate random geometries sampled from the vibrational wave function via a Monte Carlo algorithm. This is triggered by setting both the AIMD_QCT_INITPOS and AIMD_QCT_WHICH_TRAJECTORY variables to negative numbers, say $-m$ and $-n$, and setting AIMD_STEPS to zero. This will generate $m$ random geometries sampled from the vibrational wave function corresponding to an average over $n$ trajectories at the user-specified simulation temperature.

For systems that are described well within the harmonic oscillator model and for properties that rely mainly on the ground-state dynamics, this simple MC approach may yield qualitatively correct spectra. In fact, one may argue that it is preferable over QCT-MD for describing vibrational effects at very low temperatures, since the geometries are sampled from a true quantum distribution (as opposed to classical and quasi-classical MD). We have included another script in the $QC/aimdman/tools directory to help with the calculation of vibrationally averaged properties (monte_geom).

It is also possible make some modes inactive, i.e., to put vibrational energy into a subset of modes (all other are set to zero). The list of active modes can be specified using the $qct_active_modes section. Furthermore, the vibrational quantum numbers for each mode can be specified explicitly using the $qct_vib_distribution input section. It is also possible to set the phases using $qct_vib_phase (allowed values are 1 and $-$1). Below is a simple sample input:

Finally we turn to a brief description of the limitations of QCT-MD. Perhaps the most severe limitation stems from the so-called “kinetic energy spilling” problem, ²⁶⁶ Czako G., Kaledin A. L., Bowman J. M.
J. Chem. Phys.
(2010), 132, pp. 164103. Link which means that there can be an artificial transfer of kinetic energy between modes. This can happen because the initial velocities are chosen according to quantum energy levels, which are usually much higher than those of the corresponding classical systems. Furthermore, the classical equations of motion also allow for the transfer of non-integer multiples of the zero-point energy between the modes, which leads to different selection rules for the transfer of kinetic energy. Typically, energy spills from high-energy into low-energy modes, leading to spurious “hot” dynamics. A second problem is that QCT-MD is actually based on classical Newtonian dynamics, which means that the probability distribution at low temperatures can be qualitatively wrong compared to the true quantum distribution. ⁶⁷⁶ Lambrecht D. S. et al.
J. Phys. Chem. A
(2011), 115, pp. 5928. Link

Q-Chem implements a routine to monitor the kinetic energy within each normal mode along the trajectory and that is automatically switched on for quasi-classical simulations. It is thus possible to monitor for trajectories in which the kinetic energy in one or more modes becomes significantly larger than the initial energy. Such trajectories should be discarded. (Alternatively, see Ref.  266 Czako G., Kaledin A. L., Bowman J. M.
J. Chem. Phys.
(2010), 132, pp. 164103. Link for a different approach to the zero-point leakage problem.) Furthermore, this monitoring routine prints the squares of the (harmonic) vibrational wave function along the trajectory. This makes it possible to weight low-temperature results with the harmonic quantum distribution to alleviate the failure of classical dynamics for low temperatures.