HefeiNAMD Series - Part I
When Does Inter-k NAC Vanish?
A teaching statement that nonadiabatic couplings between different k points “cannot be calculated” raises a physical question: is the coupling really zero, or has the calculation excluded the motion that would produce it? The precise result concerns a trajectory that preserves the primitive translations at every time. Along that restricted trajectory, the time-derivative coupling between different primitive crystal momenta is exactly zero.
Part 0 introduced the workshop calculation chain. This first technical article isolates the problem behind one of its choices: which electronic transitions can a periodic nuclear trajectory represent? We first distinguish wavefunction orthogonality from derivative coupling, prove the symmetry restriction, and identify what is missing from primitive-cell molecular dynamics. Part II then develops the real-space supercell route.
The general NAC formulas already have their own foundation note. Here NAC means nonadiabatic coupling, and the new question is how translation symmetry constrains its scalar time-derivative form.
1. Begin with the time-derivative coupling
Let the nuclei follow a differentiable prescribed path \(\mathbf R(t)\), and let the instantaneous electronic eigenstates satisfy
We use \(d_{mn}\) for the scalar coupling along that path, rather than for the full nuclear-coordinate derivative vector:
This quantity has units of inverse time. Differentiating the eigenvalue equation and projecting onto a distinct state gives
The quotient requires nondegenerate states and a differentiable choice of eigenvectors. At a degeneracy, one must treat the coupled subspace; dividing by a vanishing gap is not a prescription for the coupling. The numerator identifies the relevant question: what spatial symmetry does the time-dependent perturbation \(\dot H_e\) have?
2. Orthogonality alone does not prove zero NAC
Two orthogonal states can rotate while remaining orthogonal. For example, in a fixed orthonormal two-state basis, define
Their overlap is always zero, but their derivative overlap is
More generally, differentiating orthonormality proves \(d_{mn}=-d_{nm}^{*}\), not \(d_{mn}=0\). Thus Bloch-state orthogonality needs an additional argument before it can say anything about NAC.
3. The strict zero follows from a symmetry preserved in time
Take a fixed lattice and require the electronic Hamiltonian to retain every primitive translation \(\mathbf L\) at every time. Choose symmetry-adapted instantaneous states, with crystal momenta fixed modulo primitive reciprocal lattice vectors:
For a nonlocal Hamiltonian, the first condition means commutation with the lattice-translation operator. Because the translation and its Bloch phase are time independent, differentiating the second relation leaves \(\partial_t\psi_{n\mathbf k}\) in the same momentum sector. Translating the full-crystal inner product therefore yields
If the two momenta differ modulo a reciprocal lattice vector, some translation has a phase different from one. Consequently,
This statement is exact within the stated trajectory model. It does not require a harmonic approximation or a nonzero energy gap: symmetry keeps the sectors separate even if energies in different sectors cross. At degeneracies we retain a basis adapted to translation symmetry. Arbitrary rotations mixing degenerate momentum sectors can obscure their labels without creating physical transfer between the sectors.
The inner product is over a Born-von Kármán periodic crystal, not an arbitrary integral of two cell-periodic factors over one primitive cell. An overlap code returns zero for these cross-sector overlaps, including adjacent-time overlaps along the same symmetry-preserving path. “Cannot be calculated” is therefore better read as “this trajectory contains no such coupling to extract.” A different overlap formula cannot recover a displacement pattern that was never present.
4. What primitive-cell molecular dynamics leaves out
Write the displacement of atom \(\kappa\) in primitive cell \(l\) as \(\mathbf u_{l\kappa}(t)\). Primitive-cell periodic molecular dynamics (MD) requires every repeated cell to undergo the same motion:
Only displacement patterns with primitive wavevector \(\mathbf q=0\) are available. Internal optical motion can still change the electronic states and couple bands at the same \(\mathbf k\). What this trajectory lacks is a displacement pattern whose phase varies from one primitive cell to the next.
Adding electronic k points does not add such a nuclear pattern. The electronic mesh samples more states of the same primitive-periodic Hamiltonian, while the nuclei remain constrained to repeat after one primitive cell. Anharmonic forces can change the time dependence of the permitted motion, but cannot create a spatial wavelength excluded by the boundary condition.
The full derivative-coupling vector needs a further distinction. Differentiating with respect to one independently displaced atom in an extended crystal can break primitive periodicity, so its matrix elements need not vanish between different reference momenta. The strict zero here concerns the contraction of that derivative with a velocity field that preserves primitive translation symmetry.
5. A strict zero within a restricted physical model
The zero is not a small number lost to floating-point precision, and the obstacle is not an overlap routine that needs a better formula. It is an exact symmetry statement for the chosen nuclear path. In that model, the cross-momentum process is absent.
It is also not a universal prohibition on momentum scattering in a real crystal. A finite-wavevector lattice distortion does not retain every primitive translation. Once that condition changes, the preceding proof no longer forces coupling between different reference primitive momenta to vanish. Removing the restriction permits coupling; other symmetries can still make a particular matrix element zero.
The practical question is therefore how to admit the missing spatial patterns into a simulation while retaining a tractable electronic calculation. Before discussing transition probabilities or lifetimes, we must establish that the nuclear and electronic state spaces contain the process of interest. Part II takes up that construction through supercell expansion, wavevector folding, MD, NAC, and surface hopping.
Sources and further reading
- Official HefeiNAMD code and teaching page.
- Z. Zheng et al., discussion of the periodic-MD momentum restriction, Introduction and the beginning of the real-space methodology.
- MQC 0: NAC and overlap formulas, for the general derivative-coupling foundation.