7. Fermion signs, fixed nodes, and fixed phase
Position in the course: Lesson 7 of 10. Complete the preceding derivation and use the explained exercises to check understanding.
Prerequisites: Antisymmetry; DMC; variational bounds
Learning goal: explain the mathematical steps, reproduce the analytical examples, and state the conditions under which the conclusions hold.
1. Cancellation is the obstacle
Fermionic exchange changes the sign of a real wavefunction. Positive and negative contributions can be individually large while their difference is small. If a signed integral is sampled using absolute weights, an observable becomes \(\langle A\rangle=\langle sA\rangle_{|w|}/\langle s\rangle_{|w|}\), where \(s=\pm1\). Small average sign makes the denominator noisy and the ratio difficult. Under conventional extensive free-energy scaling, average sign behaves as \(e^{-\beta N\Delta f}\) in many finite-temperature formulations, so maintaining relative precision can require exponentially increasing work. This behavior is representation and system dependent; it is not a claim that every QMC problem has the same severity.
2. Fixed nodes define a boundary-value problem
The trial nodal surface is \(\mathcal N_T=\{R:\Psi_T(R)=0\}\). Fixed-node projection solves the local Schrödinger problem within its pockets with Dirichlet zero boundary conditions. When pockets and signs are stitched consistently with fermionic antisymmetry, the allowed trial domain is restricted. For an appropriate real local Hamiltonian this restriction yields
Exact nodes recover the fermionic ground energy in the ideal projection limit. Sampling longer cannot move the boundary. Improving positive Jastrow amplitudes improves efficiency but, by itself, does not change fixed nodes. A nodal error is a systematic approximation, not part of a conventional sampling error bar.
3. A one-dimensional boundary check
For the harmonic oscillator, the antisymmetric odd sector has lowest state \(\psi_1=x e^{-x^2/2}\) with energy \(3/2\). Its node is \(x=0\). A zero boundary at the origin separates two half-lines; the lowest Dirichlet solution on either half-line has that energy. The unconstrained positive oscillator ground state has energy \(1/2\) and even symmetry. This example illustrates how boundaries impose a symmetry sector, not a numerical QMC measurement. In many-electron space, nodes have far more complex geometry.
4. Complex states require phase constraints
Write \(\Psi=\rho e^{i\theta}\). In a simple zero-vector-potential Hamiltonian, the kinetic quadratic form contains \(|\nabla\rho|^2+\rho^2|\nabla\theta|^2\). Holding a trial phase fixed turns its gradient into an effective extra potential while optimizing \(\rho\). This fixed-phase construction generalizes fixed nodes; magnetic vector potentials require a gauge-consistent covariant derivative. One should not label a complex twist-boundary calculation fixed-node without identifying how phase is handled.
Nonlocal pseudopotentials and localization approximations can compromise a naive fixed-node upper-bound statement. Variationally motivated treatments such as appropriate T-moves need their own assumptions and convergence analysis.
A variational inequality does not estimate the remaining distance from the exact energy. Without an independent reference or a controlled family of improvements, nodal bias may be unknown. Small sampling error only establishes precision within the constraint. Compare trial families, symmetry sectors, and matched energy differences, and report unresolved nodal uncertainty explicitly rather than inventing a numerical error bar.
5. Exercises and solutions
Exercise: If the average sign decreases from \(0.1\) to \(0.01\), how does rough work for the same relative precision change?
Solution
With comparable finite variance and independent equivalents, relative noise scales roughly as \(1/(\sqrt M\langle s\rangle)\). Work increases by about a factor 100. Actual covariance and correlations also matter.
Exercise: Can a lower noisy DMC energy alone prove improved nodes?
Solution
No. Compare statistical uncertainty and separately control timestep, population, finite size, and pseudopotential approximations. A significant difference under matched conditions is evidence, not an isolated noisy number.
Original teaching schematic of the mathematics or algorithm; it is not simulation or experimental data.
6. Sources and connections
- Ceperley (1991), Fermion nodes
- Casula (2006), Beyond the locality approximation in diffusion Monte Carlo
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