6.4 RDF, mean-square displacement and honest uncertainty
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.
6.4.1 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
6.4.2 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.
6.4.2.1 Input block 1
&PRINT
&TRAJECTORY
FORMAT XYZ
&EACH
MD 10
&END EACH
&END TRAJECTORY
&CELL
&EACH
MD 10
&END EACH
&END CELL
&END PRINT
6.4.3 Worked investigation
Intuition. An RDF measures pair structure; an MSD measures displacement growth. They require different coordinate handling: minimum-image distances for local pair structure, unwrapped continuous trajectories for diffusion. Thousands of saved frames are not thousands of independent samples. Use an equilibrated production segment with known timestep, output stride, cell vectors and atom identities. Analysis is performed with a validated external script or package such as MDAnalysis, ASE-based code or TRAVIS, with the exact version and conventions recorded; no built-in CP2K RDF output is assumed here.
Original output delta. Under MOTION/PRINT save both TRAJECTORY and CELL at matching cadence. For a 0.5 fs MD step, EACH/MD 10 gives a 5 fs stored interval, not 10 fs. Whether that interval is suitable depends on the observable. For velocity correlation/IR one may need much finer output, and unwrapping can fail if particles move too far between saved frames.
See input block 1 above.
Workflow. 1. Remove the justified equilibration interval, parse atom identities and align cell frames with coordinates. 2. For O–O RDF, accumulate pair distances using the correct periodic minimum image. Normalize by shell volume and number density, handling same-species self exclusion and pair double counting consistently. 3. Limit isotropic RDF range conservatively relative to the cell's shortest periodic image distance; a simple half-box rule applies only to suitable orthorhombic geometry. 4. For diffusion, unwrap oxygen or molecular center trajectories continuously, with an algorithm appropriate to fixed or changing cell. 5. Compute MSD(τ)=〈|r(t+τ)−r(t)|²〉 over particles and time origins. 6. Fit the diffusive regime, where MSD≈6Dτ in isotropic three dimensions, excluding the short ballistic region and poorly sampled longest lags. 7. Repeat across independent blocks and runs.
Interpretation. An RDF peak position is not a bond probability by itself; coordination requires integrating 4πρr²g(r) to a stated cutoff. The apparent plateau or slope of MSD depends on lag window. Wrapped coordinates produce artificial jumps or bounded displacement and can make diffusion meaningless. A very short AIMD run may have no defensible diffusive window; report that limitation instead of forcing a D value. Finite-size hydrodynamic corrections require assumptions and additional properties, not an automatic universal offset.
Statistics and checks. Choose blocks longer than relevant correlation times and inspect uncertainty stability as block size changes. Overlapping time origins improve sampling but are not independent replicates. For NpT, use instantaneous cells in distance calculations and document the unwrapping convention; remove only justified global drift. Compare RDF normalization against a known test configuration or idealized random positions before analyzing scientific data. Convert diffusion units explicitly: angstrom²/ps differs from m²/s by 10^−8.
Exercise. Submit two analyses of the same trajectory: one intentionally wrong wrapped-coordinate MSD and one corrected unwrapped MSD, clearly labeled as a debugging demonstration. Explain their difference without claiming a physical diffusion coefficient until a stable fit regime and uncertainty estimate are established. Vary RDF bin width and coordination cutoff to expose analysis choices.
Diagram. Left: molecule crossing a periodic boundary, with wrapped jump and continuous unwrapped path. Right: schematic MSD with ballistic, candidate diffusive and noisy-long-lag regions. EN: minimum image, unwrapped displacement, fit window, correlated samples, uncertainty. A separate RDF shell diagram labels r and Δr.
6.4.4 Related calculations
- 6.3 Restart continuity is a state-recovery experiment
- 7.1 Is this optimized molecule really a minimum?