9. Periodic cells, finite-size errors, and a validation project
Position in the course: Lesson 9 of 10. Complete the preceding derivation and use the explained exercises to check understanding.
Prerequisites: Bloch boundaries; DMC; statistical errors
Learning goal: explain the mathematical steps, reproduce the analytical examples, and state the conditions under which the conclusions hold.
1. A periodic cell is a finite model
An electronic QMC supercell uses a chosen number of electrons, nuclei, and periodic replicas. Coulomb interactions require a consistent long-range convention, usually an Ewald-style treatment rather than minimum-image truncation. Charge neutrality and background conventions affect electrostatic energies. The finite cell represents neither an isolated molecule nor the infinite solid automatically. Energies per particle, boundary conditions, electron number, spin state, and interaction convention must be reported.
2. Twist boundaries and one-body shell effects
For an electron moved by supercell lattice vector \(L\), a twist boundary has \(\Psi(\ldots,r_i+L,\ldots)=e^{i\theta\cdot L}\Psi(\ldots,r_i,\ldots)\). Allowed single-particle momenta become \(k=G+\theta\), shifting the finite reciprocal grid. Twist averaging reduces shell and Brillouin-zone integration effects by combining estimates over twists with declared weights. It does not automatically eliminate two-body correlation finite-size errors. Complex twists generally invoke a phase treatment, whereas certain symmetry twists permit real wavefunctions.
3. Why a missing long-wavelength mode matters
The reciprocal spacing scales as \(2\pi/L\). A finite cell omits sufficiently small nonzero wavevectors, altering long-range density fluctuations. The structure factor \(S(k)=N^{-1}\langle\rho_k\rho_{-k}\rangle\) for \(k\ne0\), with \(\rho_k=\sum_je^{ik\cdot r_j}\), provides information about those fluctuations. Kinetic and interaction finite-size corrections depend on the physical small-\(k\) behavior and estimator conventions; an empirical \(1/N\) extrapolation is not a universal theorem. Compare several sizes, shapes, and twists when the intended property demands bulk accuracy.
4. Worked grid and uncertainty examples
For a one-dimensional cell length \(L\), zero twist permits \(k_n=2\pi n/L\); half-grid twist gives \(k_n=2\pi(n+1/2)/L\). Doubling \(L\) halves grid spacing but also changes particle number at fixed density. This analytical illustration explains shell effects without claiming a simulated solid.
If independent twist estimates have common standard error \(s\) and equal weights across \(K\) twists, the average error is \(s/\sqrt K\). Unequal weights give variance \(\sum_jw_j^2s_j^2\) for independent estimates. Shared randomness creates cross covariances. A small statistical twist-average error does not bound incomplete twist quadrature or residual size bias.
Molecular and bulk calculations require different priorities. An isolated molecule needs a stated interaction boundary and possibly a large box, whereas a solid needs controlled density and commensurate cells. Shape changes reciprocal grids and image interactions even at fixed electron count. Choose convergence controls for the modeled geometry and observable; a molecular protocol cannot be transferred unchanged to a metal.
5. A practical learning project and exercises
First implement or inspect a one-dimensional oscillator VMC benchmark: check logarithmic derivatives, exact \(E_V(a)\), and constant local energy at \(a=1\). Next, study recorded output from a validated QMC package and separate optimization, VMC, and DMC sections. Finally design, rather than assume completed, convergence tests for steps, population, cell size, and twists. Orbitals may come from DFT or HF tutorials, but changing orbital sources is a trial-state choice, not proof of improved nodes.
Exercise: Is agreement of two cell sizes sufficient to certify the thermodynamic limit?
Solution
No. Shell effects can cancel accidentally, and different finite-size components can compensate. Use several controlled points, physical correction assumptions, and uncertainty-aware fits.
Exercise: Does an energy difference cancel every systematic error?
Solution
No. Cancellation depends on matched Hamiltonians, cells, trial quality, and the property. Structural changes can alter nodes or finite-size behavior; demonstrate cancellation instead of asserting it.
Original teaching schematic of the mathematics or algorithm; it is not simulation or experimental data.
6. Sources and connections
- Lin, Zong and Ceperley (2001), Twist-averaged boundary conditions in continuum quantum Monte Carlo algorithms
- Chiesa et al. (2006), Finite-size error in many-body simulations with long-range interactions
Related theory: Classical Monte Carlo · Molecular methods · Density functional theory
Software connection: CP2K · Quantum ESPRESSO · Gaussian
These software courses provide related background on energies, orbitals, or convergence management; they do not imply that the Monte Carlo or QMC examples on this page were executed there.
7. Related theory and practice
Quantum mechanics · Molecular methods · Monte Carlo · Molecular dynamics · QMCPACK