Taylor Series Expansion of the Potential Energy Surface¶

In the previous chapters, we introduced the three fundamental quantities used during geometry optimization:

the electronic energy,
the energy gradient,
the Hessian matrix.

The next question is:

How does Gaussian use these quantities to predict a better molecular geometry?

Instead of calculating the complete Potential Energy Surface (PES), which is computationally impossible for most molecules, Gaussian approximates only the small region surrounding the current molecular geometry.

This approximation is performed using the Taylor Series Expansion.

The Taylor series provides a local mathematical model of the Potential Energy Surface and allows Gaussian to estimate how the energy changes when the atoms are displaced.

Why Is an Approximation Needed?¶

The exact Potential Energy Surface of a molecule containing many atoms is enormously complicated.

For example,

a molecule containing 20 atoms possesses

\[ 3N-6=54 \]

independent internal coordinates.

Constructing the complete energy surface in this 54-dimensional space would require an astronomical number of electronic structure calculations.

Instead,

Gaussian approximates only the small region surrounding the current geometry.

This local approximation is sufficiently accurate for determining the next optimization step.

Taylor Series Expansion¶

Suppose the molecule is currently located at the geometry

\[ \mathbf{R}. \]

If the nuclei are displaced by a small amount

\[ \Delta\mathbf{R}, \]

the electronic energy can be approximated using the Taylor series,

\[ \boxed{ E(\mathbf{R}+\Delta\mathbf{R}) = E(\mathbf{R}) + \mathbf g^T\Delta\mathbf R + \frac12 \Delta\mathbf R^T \mathbf H \Delta\mathbf R +\cdots } \]

where

\(E(\mathbf R)\) is the current electronic energy,
\(\mathbf g\) is the energy gradient,
\(\mathbf H\) is the Hessian matrix.

Higher-order terms are usually neglected because the displacement between optimization steps is small.

Meaning of Each Term¶

The Taylor expansion contains three important contributions.

Current Energy¶

\[ E(\mathbf R) \]

This is the electronic energy of the current molecular geometry obtained from the SCF calculation.

It serves as the starting point for the local approximation.

Gradient Term¶

\[ \mathbf g^T\Delta\mathbf R \]

The gradient term predicts how the energy changes in the immediate neighborhood of the current geometry.

It determines

the downhill direction,
whether the energy increases or decreases,
the initial tendency of atomic motion.

Hessian Term¶

\[ \frac12 \Delta\mathbf R^T \mathbf H \Delta\mathbf R \]

The Hessian term accounts for the curvature of the Potential Energy Surface.

It improves the approximation by describing how rapidly the gradient itself changes.

This term is responsible for determining the appropriate optimization step size.

Linear Approximation¶

If only the gradient term is retained,

the Taylor expansion becomes

\[ E \approx E_0 + \mathbf g^T \Delta\mathbf R. \]

This corresponds to approximating the Potential Energy Surface by a straight line.

Energy

  ^
  |
  |       /
  |     /
  |   /
  | /
  +-------------------->

Although simple,

this approximation is generally insufficient for efficient geometry optimization.

Quadratic Approximation¶

Including the Hessian produces

\[ E \approx E_0 + \mathbf g^T \Delta\mathbf R + \frac12 \Delta\mathbf R^T \mathbf H \Delta\mathbf R. \]

The energy surface is now represented by a parabola.

Energy

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

The quadratic approximation provides a much more realistic description of the local Potential Energy Surface.

Nearly all modern optimization algorithms are based on this expression.

Why Does Gaussian Use Only the Local Surface?¶

During optimization,

Gaussian moves the atoms only a small distance at each iteration.

Therefore,

only the neighborhood surrounding the current geometry needs to be described accurately.

After the geometry has been updated,

a new SCF calculation is performed,

a new gradient is computed,

the Hessian is updated,

and a new Taylor expansion is constructed.

Thus,

Gaussian continuously rebuilds a local approximation of the Potential Energy Surface throughout the optimization.

Local Approximation During Optimization¶

The optimization procedure may be visualized as

Current Geometry
        │
        ▼
SCF Calculation
        │
        ▼
Electronic Energy
        │
        ▼
Gradient
        │
        ▼
Hessian
        │
        ▼
Taylor Series Approximation
        │
        ▼
Predict New Geometry
        │
        ▼
Repeat

Each optimization step creates a new local approximation around the updated geometry.

Importance in Geometry Optimization¶

The Taylor series forms the mathematical foundation of nearly all optimization algorithms.

It is used in

Newton–Raphson optimization,
Quasi-Newton methods,
Rational Function Optimization (RFO),
Berny optimization.

Although these algorithms differ in how they estimate the Hessian and determine the optimization step, they all begin with the same local Taylor expansion of the Potential Energy Surface.

Limitations¶

The Taylor series is only accurate for small displacements.

If the optimization step becomes too large,

the local approximation no longer represents the true Potential Energy Surface.

For this reason,

Gaussian carefully controls the optimization step length and continually updates the gradient and Hessian after each iteration.

This ensures that the approximation remains valid throughout the optimization.

Summary¶

The Taylor Series Expansion provides a local mathematical representation of the Potential Energy Surface around the current molecular geometry. By combining the electronic energy, the energy gradient, and the Hessian matrix, Gaussian predicts how the molecular energy will change for small atomic displacements. This quadratic approximation forms the theoretical basis of modern geometry optimization methods and enables efficient prediction of improved molecular geometries without constructing the complete Potential Energy Surface.

Next Section¶

The next chapter introduces the Optimization Algorithms used in computational chemistry. We compare Steepest Descent, Newton–Raphson, Quasi-Newton, and Rational Function Optimization before examining the Berny optimization algorithm employed by Gaussian.