Self-Consistent Field (SCF) Algorithm¶
The previous chapter showed that the electronic structure problem can be written as the Roothaan–Hall matrix equation
At first glance, this appears to be an ordinary eigenvalue problem that can be solved directly.
However, there is one major complication.
The Fock matrix itself depends on the electron density, while the electron density depends on the molecular orbitals, and the molecular orbitals are obtained only after solving the Roothaan–Hall equation.
This circular dependence prevents a direct solution.
To overcome this difficulty, Gaussian uses an iterative procedure known as the Self-Consistent Field (SCF) algorithm.
Why is an Iterative Procedure Necessary?¶
The dependency between the various quantities can be summarized as
Molecular Orbitals
│
▼
Electron Density
│
▼
Fock Matrix
│
▼
Solve Roothaan–Hall Equation
│
▼
New Molecular Orbitals
Since each quantity depends on the previous one, no quantity is known initially.
Instead, Gaussian begins with an approximate solution and gradually improves it through repeated iterations until all quantities become self-consistent.
What Does "Self-Consistent" Mean?¶
A calculation is said to be self-consistent when
The electron density used to construct the Fock matrix is identical to the electron density obtained after solving the Fock equations.
In other words,
Input Density
│
▼
Build Fock Matrix
│
▼
Solve Matrix Equation
│
▼
New Density
If
New Density = Input Density
SCF has converged.
At this point, further iterations no longer change the electronic structure.
Step 1: Initial Guess¶
Every SCF calculation begins with an initial estimate of the electron density.
Gaussian automatically generates this guess using built-in procedures such as
- Superposition of Atomic Densities (SAD)
- Core Hamiltonian Guess
- Previous checkpoint files (when available)
Although the initial guess is only approximate, it is usually sufficiently close to allow rapid convergence.
Step 2: Construct the Fock Matrix¶
Using the initial electron density, Gaussian constructs the Fock matrix
where
- \(\mathbf{H}_{core}\) contains the one-electron terms,
- \(\mathbf{G}(P)\) contains the electron–electron interaction,
- \(P\) is the density matrix.
Since the density matrix changes during every iteration, the Fock matrix also changes.
Step 3: Solve the Roothaan–Hall Equation¶
The newly constructed Fock matrix is diagonalized,
to obtain
- molecular orbital coefficients,
- orbital energies.
These molecular orbitals represent the best solution corresponding to the current Fock matrix.
Step 4: Build a New Density Matrix¶
Using the occupied molecular orbitals,
Gaussian constructs an updated density matrix,
for a closed-shell system.
The density matrix describes how electrons are distributed throughout the molecule.
Step 5: Compare with the Previous Iteration¶
The newly constructed density matrix is compared with the previous density matrix.
Gaussian monitors quantities such as
- change in electronic energy,
- change in density matrix,
- RMS density error,
- maximum density error.
If these quantities are larger than the convergence thresholds,
another SCF iteration is performed.
Otherwise,
the calculation has converged.
Complete SCF Cycle¶
The entire procedure is illustrated below.
Read Molecular Geometry
│
▼
Generate Initial Guess
│
▼
Construct Density Matrix
│
▼
Build Fock Matrix
│
▼
Solve Roothaan–Hall Equation
│
▼
Obtain Molecular Orbitals
│
▼
Construct New Density Matrix
│
▼
Convergence Test
│ │
│No │Yes
▼ ▼
Repeat SCF Converged
This loop represents the heart of every Hartree–Fock and Kohn–Sham DFT calculation.
SCF Convergence¶
As the iterations proceed,
- the electron density stabilizes,
- the molecular orbital energies change less,
- the total electronic energy approaches a constant value.
A typical convergence pattern looks like
Iteration Energy (Hartree)
1 -378.512341
2 -379.018672
3 -379.112418
4 -379.118974
5 -379.119241
6 -379.119248
Notice that the energy changes become progressively smaller until convergence is achieved.
SCF in the Gaussian Output¶
During execution, Gaussian prints the progress of the SCF procedure.
A typical output contains
Each line corresponds to one SCF iteration.
Once convergence has been achieved, Gaussian prints
This is the final electronic energy obtained from the converged electron density.
Why SCF Sometimes Fails¶
Not every calculation converges easily.
SCF convergence problems may occur when
- the initial guess is poor,
- the molecular geometry is unrealistic,
- near-degenerate orbitals are present,
- transition-metal systems are studied,
- highly charged molecules are considered.
Gaussian provides several techniques for improving convergence, including
- DIIS (Direct Inversion in the Iterative Subspace),
- level shifting,
- damping,
- quadratic convergence,
- improved initial guesses.
These methods are discussed in more advanced Gaussian calculations.
Relationship to Density Functional Theory¶
Although this chapter has described Hartree–Fock terminology, the same iterative procedure is used in Density Functional Theory.
The only difference is that
- Hartree–Fock constructs a Fock matrix,
- Kohn–Sham DFT constructs a Kohn–Sham matrix.
The iterative SCF procedure itself remains essentially unchanged.
Summary¶
The Self-Consistent Field (SCF) algorithm is the numerical procedure used to solve the Hartree–Fock and Kohn–Sham equations. Starting from an initial guess for the electron density, Gaussian repeatedly constructs the Fock matrix, solves the Roothaan–Hall equations, updates the density matrix, and checks for convergence. The cycle continues until the input and output electron densities become identical within predefined numerical thresholds. The final converged solution provides the molecular orbitals, electron density, and electronic energy reported in the Gaussian output.
Next Section¶
The previous chapters developed the Hartree–Fock framework used to solve the electronic Schrödinger equation. The next chapter introduces Density Functional Theory (DFT), explaining why Hartree–Fock alone is often insufficient and how DFT improves the description of electron correlation while retaining the same computational framework.