Convergence Criteria¶

During a geometry optimization, Gaussian repeatedly modifies the molecular geometry to locate a stationary point on the Potential Energy Surface (PES). After each optimization step, the program must determine whether the structure is sufficiently close to the minimum or whether additional optimization cycles are required.

This decision is based on a set of convergence criteria. These criteria measure the magnitude of the forces acting on the atoms and the amount by which the atoms move between successive optimization steps.

Only when all convergence criteria satisfy their prescribed thresholds does Gaussian terminate the optimization successfully.

Why Are Convergence Criteria Necessary?¶

Suppose the optimization stopped after only a few iterations.

Energy
  ^
  |
  |             *
  |          *
  |       *
  |    *
  | *
  +---------------------------->

Although the energy has decreased, the molecule may still be far from the true minimum.

On the other hand, if the optimization continued indefinitely,

Energy
  ^
  |
  |        *
  |      *   *
  |    *       *
  |___*_________*________>

the energy would change only by extremely small amounts while consuming unnecessary computational time.

Therefore, Gaussian uses objective mathematical criteria to determine when the structure has effectively reached a stationary point.

The Four Convergence Criteria¶

Gaussian monitors four quantities during every optimization cycle.

Maximum Force
RMS Force
Maximum Displacement
RMS Displacement

These values appear near the end of every optimization step.

A typical output is

 Item                 Value      Threshold   Converged?

 Maximum Force       0.000018    0.000450      YES
 RMS Force           0.000010    0.000300      YES
 Maximum Displacement0.000042    0.001800      YES
 RMS Displacement    0.000021    0.001200      YES

The optimization is complete only when every criterion is marked YES.

Maximum Force¶

The Maximum Force is the largest force acting on any atom in the molecule.

Mathematically,

\[ \boxed{ F_{\max} = \max_i \left| \mathbf F_i \right| } \]

where

\(\mathbf F_i\)

is the force acting on atom \(i\).

A large maximum force indicates that at least one atom is still far from its equilibrium position.

As the optimization proceeds,

the maximum force approaches zero.

RMS Force¶

The Root Mean Square (RMS) Force measures the average force acting on all atoms.

It is defined as

\[ \boxed{ F_{\rm RMS} = \sqrt{ \frac1N \sum_i F_i^2 } } \]

where

\(N\)

is the number of atomic coordinates.

Unlike the maximum force,

the RMS force reflects the overall quality of the optimization.

Maximum Displacement¶

The Maximum Displacement measures the largest movement of any atom during the current optimization step.

Before Step

C ---- O

After Step

C ----- O

If an atom is still moving significantly,

the molecular geometry has not yet stabilized.

Near convergence,

the maximum displacement becomes extremely small.

RMS Displacement¶

The Root Mean Square Displacement measures the average movement of all atoms between two optimization steps.

It is given by

\[ \boxed{ D_{\rm RMS} = \sqrt{ \frac1N \sum_i (\Delta R_i)^2 } } \]

where

\(\Delta R_i\)

is the displacement of coordinate \(i\).

Small RMS displacement indicates that the molecular structure is no longer changing appreciably.

Why Are Four Criteria Used?¶

No single quantity is sufficient to determine convergence.

For example,

a single atom may still experience a large force even though the average force is small,
the forces may be small while some atoms continue to move significantly,
the geometry may appear stationary although the electronic structure has not completely stabilized.

Using four independent criteria ensures that

the forces are small,
the geometry changes are small,
the entire molecule has reached a stationary point.

Typical Convergence Behaviour¶

During optimization,

the convergence quantities gradually decrease.

Optimization Step

Force

│\
│ \
│  \
│   \
│    \____

──────────────► Step Number

Similarly,

Displacement

│\
│ \
│  \
│   \
│    \____

──────────────► Step Number

Eventually,

all quantities fall below their convergence thresholds.

Tight and Loose Convergence¶

Gaussian provides several convergence options.

Default Convergence¶

Suitable for most molecular optimizations.

Tight Convergence¶

Opt=Tight

Uses smaller convergence thresholds.

Advantages

More accurate geometries.
Better vibrational frequencies.
Improved thermochemical properties.

Disadvantages

More optimization cycles.
Increased computational time.

Very Tight Convergence¶

Opt=VeryTight

Used primarily for

benchmark calculations,
high-accuracy thermochemistry,
potential energy surface studies.

Successful Optimization¶

When every convergence criterion is satisfied,

Gaussian prints

Optimization completed.

This indicates that the optimizer has located a stationary point on the Potential Energy Surface.

However,

a successful optimization does not necessarily guarantee a true minimum.

A subsequent Frequency Calculation is required to determine whether

the structure is a minimum,
or a transition state.

Relationship with Frequency Analysis¶

The optimization convergence criteria determine only whether

\[ \nabla E = 0 \]

within numerical accuracy.

They do not determine the curvature of the Potential Energy Surface.

The Hessian matrix obtained during a Frequency Calculation distinguishes

local minima,
transition states,
higher-order saddle points.

Therefore,

Geometry Optimization is almost always followed by a Frequency Calculation.

Summary¶

The convergence criteria determine when Gaussian should terminate a geometry optimization. During every optimization cycle, the program evaluates the Maximum Force, RMS Force, Maximum Displacement, and RMS Displacement. These quantities measure how close the molecular geometry is to a stationary point on the Potential Energy Surface. When all four values fall below their respective thresholds, Gaussian reports "Optimization completed." Although this confirms that a stationary point has been located, a subsequent Frequency Calculation is required to verify whether the optimized structure corresponds to a true minimum or a transition state.

Next Section¶

In the final chapter, we combine all the mathematical concepts discussed so far into the Complete Geometry Optimization Algorithm, illustrating how Gaussian proceeds from the initial molecular geometry to the final optimized structure using the Berny optimization method.