The Berny Optimization Algorithm¶

Modern versions of Gaussian perform geometry optimization using the Berny Optimization Algorithm. Rather than relying solely on classical methods such as Steepest Descent or Newton–Raphson, the Berny algorithm combines several numerical optimization techniques to efficiently locate stationary points on the Potential Energy Surface (PES).

The Berny optimizer is specifically designed for molecular systems and has become one of the most widely used optimization methods in computational chemistry. It combines energy gradients, approximate Hessian matrices, internal coordinates, and Rational Function Optimization (RFO) to provide stable and rapid convergence for molecules ranging from small organic compounds to large biomolecules.

Why Does Gaussian Use the Berny Algorithm?¶

Geometry optimization presents several computational challenges.

Molecules often contain many atoms, leading to a large number of variables.
The exact Hessian matrix is computationally expensive to evaluate.
Large optimization steps may move the molecule away from the minimum.
Poor initial geometries can slow convergence or even cause optimization failure.

The Berny algorithm addresses these challenges by using an approximate description of the local Potential Energy Surface and continuously improving this approximation during the optimization.

Instead of searching the entire energy surface, it focuses only on the region surrounding the current molecular geometry.

Optimization in Internal Coordinates¶

One of the key features of the Berny algorithm is that it performs optimization using internal coordinates rather than Cartesian coordinates.

Internal coordinates include

bond lengths,
bond angles,
dihedral angles.

For example,

Bond Length

C — O

Bond Angle

H
 \
  C
 /
O

Dihedral Angle

H—C—C—H

These coordinates describe chemically meaningful structural changes and usually lead to faster convergence than moving atoms independently along the Cartesian x-, y-, and z-axes.

Approximate Hessian¶

At the beginning of the optimization, Gaussian constructs an initial approximate Hessian.

Instead of calculating every second derivative exactly, the program estimates the curvature of the Potential Energy Surface using empirical information.

After each optimization step,

new gradient information is obtained,
the Hessian estimate is refined,
the predicted shape of the energy surface becomes more accurate.

As the optimization progresses, the approximate Hessian gradually approaches the true Hessian.

Rational Function Optimization (RFO)¶

The Berny optimizer uses Rational Function Optimization (RFO) to determine the displacement of the molecular geometry.

Unlike simple Newton–Raphson optimization, RFO

prevents excessively large optimization steps,
improves numerical stability,
produces reliable convergence near stationary points.

This makes the optimization considerably more robust, especially for large molecular systems.

One Optimization Cycle¶

A single Berny optimization step consists of several stages.

Current Geometry
        │
        ▼
SCF Calculation
        │
        ▼
Electronic Energy
        │
        ▼
Energy Gradient
        │
        ▼
Update Hessian
        │
        ▼
RFO Step Calculation
        │
        ▼
New Geometry

The process is repeated until all convergence criteria are satisfied.

Updating the Hessian¶

One of the most important aspects of the Berny algorithm is that the Hessian is updated rather than recalculated.

Initial Hessian
       │
       ▼
Optimization Step
       │
       ▼
New Gradient
       │
       ▼
Improved Hessian
       │
       ▼
Next Optimization Step

This strategy dramatically reduces the computational cost while preserving excellent convergence.

Following the Potential Energy Surface¶

The optimization process may be viewed as repeatedly constructing a local model of the Potential Energy Surface.

Potential Energy Surface

          ●
         /
        ●
       /
      ●
     /
    ●
   /
 ●____________________

Minimum

Each optimization step moves the molecule closer to the minimum energy structure.

Because the local approximation is rebuilt after every step, the optimizer continually improves its prediction of the energy surface.

Advantages of the Berny Algorithm¶

Compared with classical optimization methods, the Berny algorithm offers several advantages.

Uses chemically meaningful internal coordinates.
Avoids repeated calculation of the exact Hessian.
Employs Rational Function Optimization for stable geometry updates.
Efficiently handles molecules containing many atoms.
Provides rapid convergence for most equilibrium geometry optimizations.
Forms the default optimization method in Gaussian.

These features explain why the Berny optimizer has become the standard algorithm for molecular geometry optimization in Gaussian.

Limitations¶

Although the Berny algorithm is highly robust, convergence is not always guaranteed.

Optimization difficulties may arise when

the initial geometry is far from the equilibrium structure,
the electronic wavefunction does not converge,
the Potential Energy Surface is unusually flat,
multiple stationary points lie close together,
the molecular system contains highly flexible regions.

In such situations, additional keywords or alternative optimization strategies may be required.

Berny Optimization Workflow¶

The complete workflow can be summarized as

Read Initial Geometry
          │
          ▼
Generate Initial Hessian
          │
          ▼
Perform SCF Calculation
          │
          ▼
Calculate Gradient
          │
          ▼
Update Hessian
          │
          ▼
Construct Local PES
          │
          ▼
Determine RFO Step
          │
          ▼
Move Atoms
          │
          ▼
Convergence Achieved?
      │              │
      │ No           │ Yes
      ▼              ▼
 Repeat        Optimized Geometry

This cycle continues until the energy gradient and atomic displacements satisfy all convergence criteria.

Summary¶

The Berny Optimization Algorithm is the default geometry optimization method implemented in Gaussian. It combines internal coordinate optimization, approximate Hessian updates, and Rational Function Optimization to efficiently locate stationary points on the Potential Energy Surface. By avoiding repeated evaluation of the exact Hessian while continually improving its approximation, the Berny algorithm achieves rapid and stable convergence for a wide variety of molecular systems. Its efficiency and robustness make it one of the most successful optimization algorithms in computational quantum chemistry.

Next Section¶

In the next chapter, we examine the Convergence Criteria used by Gaussian to determine when a geometry optimization has successfully reached a stationary point on the Potential Energy Surface.