7.2 From a dipole trajectory to an IR spectrum
These are unexecuted teaching inputs and starting models. Original diagrams are schematics, not calculated results. Validate version-specific syntax, licensed or authorized data, numerical convergence and the scientific model before using this workflow.
7.2.1 Model, units and provenance
Keep basis/potential files and executable versions traceable. Grid controls use Ry; common energy/force outputs use hartree and hartree/bohr. Read the unit in each output heading.
Shared inputs, conventions and evidence
7.2.2 Unexecuted inputs and explicit deltas
Use the accompanying instructions to identify the parent calculation and placement of every delta; a snippet is not automatically a standalone input. Preserve all blank-line and file-provenance requirements.
7.2.2.1 Input block 1
7.2.3 Worked investigation
Question and intuition. Normal modes provide a local, zero-temperature harmonic picture. A molecular-dynamics trajectory instead samples fluctuating environments and anharmonic motion, within the chosen classical-nuclear model. IR absorption is related to the autocorrelation of the total dipole derivative, C(t) = ⟨μ̇(0)·μ̇(t)⟩. Its Fourier transform reveals frequencies at which charge motion remains correlated. A velocity autocorrelation yields a vibrational density-of-states-like quantity; it is not generally an IR spectrum because it lacks dipole selection rules. This case uses neutral, isolated methanol so that molecular dipoles can be treated before the additional complications of periodic polarization.
Prerequisites. Prepare a stable Born–Oppenheimer MD setup with validated time step and SCF convergence. Equilibrate at the chosen temperature, save coordinates and velocities, then examine a production segment whose thermostat influence is understood. A short NVE segment initialized from an equilibrated NVT state is a useful diagnostic; NVE is not a universal requirement, and energy drift must be acceptable. Use the same Hamiltonian as the intended static comparison. A DFTB trajectory and a PBE Hessian compare both method and sampling effects, so do not attribute every difference to temperature. The ETH Zürich MD infrared exercise motivates this comparison.
Original input delta: valid syntax. Insert in FORCE_EVAL/DFT/PRINT of the existing nonperiodic MD input. Keep GLOBAL/RUN_TYPE MD and the separately validated MOTION/MD setup. This block records dipoles; spectral analysis is external.
See input block 1 above.
Print every MD step for the initial test. If the MD step is Δt, the saved interval is Δts = stride × Δt. Use the time and dipole units printed by the output format; convert once and preserve the conversion in the analysis manifest. The 2025.2 MOMENTS reference warns that periodic moments can jump between polarization branches. PERIODIC F is not an acceptable shortcut for bulk periodic liquid water.
Numbered analysis procedure. 1. Parse the actual file format into a table containing time, μx, μy, μz. Check uniform intervals, missing frames, duplicated restart frames, and changes of units or cell convention. Keep the original file. 2. Remove the equilibration segment. Plot all three components and their first differences before Fourier analysis. A discontinuity may be an output/restart artifact, not a chemical event. 3. Differentiate with a documented central finite-difference stencil, discarding unsupported end points. Differentiation amplifies high-frequency noise, so tighten SCF rather than hiding noise with aggressive filtering. 4. Estimate C(k) by averaging dot products across multiple time origins, with the appropriate count of pairs at each lag. Limit the maximum lag before the estimate becomes dominated by too few origins. 5. Apply a documented lag window and transform the correlation using a consistent one-sided/two-sided convention. An external Python/NumPy or specialist spectroscopy workflow can do this; no automatic CP2K “IR FFT” is implied here. 6. Compare independent blocks and longer trajectories. Retain the time-domain correlation, the unnormalized spectrum, and all preprocessing choices alongside a display-normalized figure.
Units and interpretation. For saved times in femtoseconds, a Fourier frequency f in fs^−1 becomes a wavenumber f/(2.99792458×10^−5) in cm^−1. This conversion uses ordinary frequency, not angular frequency; an extra 2π would be wrong. The Nyquist limit is 1/(2Δts), while the rough resolution is set by the correlation duration, approximately 1/(cTcorr). Zero-padding interpolates a curve without creating finer physical resolution. Windowing trades reduced ringing against broader peaks. Absolute absorption needs the correct volume, temperature, electromagnetic prefactors, normalization, and any justified quantum correction; a normalized transform alone supplies relative spectral information.
Convergence, pitfalls, and exercise. Compare time steps, SCF thresholds, correlation lengths, and block uncertainties separately. Molecular rotation contributes to a laboratory-frame dipole signal; arbitrary rotational alignment changes the observable and must be justified. For periodic insulators, consistently unwrap Berry-phase branches or use an appropriate current/polarization analysis; wrapping atomic positions is a different problem. Classical nuclei miss quantum populations and zero-point motion, and a multiplicative correction cannot repair all anharmonic errors. Exercise: calculate the Nyquist bound for a chosen saved interval, then deliberately downsample one already-computed dipole record and identify aliasing without presenting it as a new simulation. Explain why a smoother spectrum may have less information.
7.2.4 Related calculations
- 7.1 Is this optimized molecule really a minimum?
- 7.3 A vertical excitation is a response calculation