Stationary Points on a Surface

Introduction

A potential energy surface has special geometries where its local slope vanishes. These stationary points organise much of a proposed reaction mechanism. A local minimum can describe a stable conformer or intermediate; a first-order saddle can describe an elementary bottleneck. The gradient identifies stationarity, while the Hessian tells whether the geometry is a minimum, saddle or still higher-index feature. Both checks are needed: a low computed energy alone is not a classification.

Core explanation

Let the surface be E(q₁,…,q m) in independent internal or mass-weighted coordinates. Its gradient has components ∂E/∂q i. At a true stationary point every independent component is zero. Numerically, optimisation stops when gradient components and energy changes fall below chosen tolerances; a program reporting “converged” should be checked for sensible thresholds and geometry, especially on a flat surface.

Near a stationary point q₀, a Taylor expansion is E(q₀+δq) ≈ E(q₀) + ½ δqᵀHδq , because the first-order gradient term vanishes. The symmetric Hessian matrix H contains second derivatives. Its eigenvectors give local directions, and eigenvalues give curvature along them. Positive eigenvalues in all independent internal directions indicate a local minimum. Exactly one negative eigenvalue indicates a first-order saddle. Two negative eigenvalues indicate a second-order saddle. A nearly zero eigenvalue can indicate a very soft torsion, numerical difficulty or an omitted symmetry consideration.

In raw Cartesian coordinates for an isolated nonlinear molecule, overall translation and rotation produce six zero modes in the exact ideal description; a linear molecule has five such modes. These are not internal reaction instabilities. Vibrational analysis projects or interprets them separately, often using a mass-weighted Hessian. A negative mass-weighted eigenvalue corresponds to an imaginary harmonic frequency in the usual computational output; it is a sign of unstable curvature, not a physically oscillating mode with an imaginary measurable rate.

Classification is local. A minimum may be metastable compared with a distant, lower minimum. A first-order saddle may connect two conformers rather than the intended chemical reactant and product. A stationary-point calculation also does not provide a finite-temperature free-energy barrier without vibrational, thermal and entropic treatment. Following paths and checking endpoints are separate steps.

At exact symmetry, special directions can make a high-symmetry structure stationary while not stable. A square-like or planar arrangement may have zero gradient because first-order forces cancel, but a symmetry-breaking displacement can lower energy. Hessian analysis reveals the negative curvature. Conversely, a small negative frequency may be numerical noise from an incomplete optimisation, inadequate integration grid or nearly free rotation. Reoptimising with tighter settings and inspecting the displacement helps decide.

Step-by-step reasoning

Choose the electronic state and independent geometry coordinates. Optimise until the gradient is small. Compute the Hessian at the same method and basis. Remove overall translation and rotation, then count negative internal eigenvalues. Inspect eigenvectors to see which atoms move. If the intended structure is a transition candidate, follow its single unstable direction both ways before assigning reactant and product endpoints.

Visual explanation

Draw three contour surfaces. A bowl has positive curvature in every direction and represents a minimum. A mountain pass bends downward along the pathway and upward across it, giving one negative direction. A hilltop bends downward in two drawn directions, representing a higher-order saddle in that two-dimensional slice. Label gradient zero at all three centres to show that slope alone cannot distinguish them.

Real-world analogy

A ball can rest at the bottom of a bowl, balance at a mountain pass or balance on a hilltop. At each exact point the immediate slope is zero, but small pushes produce different outcomes. This illustrates Hessian curvature. Molecular motion has many more coordinates, and a thermal molecule need not sit still at any one point.

Real-world example

A computational chemist optimises a proposed proton-transfer structure with the proton halfway between donor and acceptor. The geometry converges, but frequency analysis shows two imaginary modes: one proton motion and one ring puckering. It is a second-order saddle, not the desired elementary transition structure. Displacing along the puckering mode and reoptimising may locate one or more true first-order saddles.

Why?

Why is a zero gradient insufficient? Minima and saddles both have zero first derivative, differing in second-order curvature. Why count eigenvalues? Diagonalising the Hessian identifies independent local directions that lower or raise energy. Why inspect the eigenvector? A negative mode unrelated to the proposed bond change may reveal the wrong process or a numerical artefact.

Common misconception

A converged geometry is not necessarily a minimum, and one imaginary frequency is not automatically proof of the desired reaction mechanism. The mode could connect unexpected conformers or represent an artefact if the geometry was not accurately converged. Likewise, a lower total energy than reactants does not remove a possible barrier elsewhere on the surface.

Worked example

Question: A stationary geometry has internal Hessian eigenvalues proportional to [8, 4, 1, −2] in consistent units. How is it classified locally? What if the −2 were positive?

Reasoning: One negative eigenvalue means one downhill curvature direction from the stationary point, so it is a first-order saddle in the four-dimensional model. If all four eigenvalues were positive, it would be a local minimum. The magnitudes alone do not reveal endpoints or a reaction rate.

Answer: It is a first-order saddle; replacing the negative value with a positive one would make it a local minimum.

Quick check

1. What property do a true local minimum and a true first-order saddle share at their central geometry? Answer: Both have zero energy gradient in all independent internal coordinates.

Exam focus

Write the local quadratic expansion and use the number of negative internal Hessian eigenvalues to classify points. Exclude translation and rotation zero modes. Distinguish electronic-energy stationary-point classification from finite-temperature thermodynamic stability, and verify the negative eigenvector and endpoints.

Advanced insight

Hessian eigenvalues depend on the coordinate metric, so vibrational frequencies conventionally use mass-weighted Cartesian coordinates after proper treatment of overall motions. The qualitative count of unstable directions at a nondegenerate stationary point remains a robust topological property under smooth nonsingular coordinate changes. A near-zero mode can make classification sensitive to numerical details and signal anharmonic or large-amplitude motion that a harmonic approximation handles poorly.

Summary

Stationary points have zero gradient; the Hessian resolves their local shape. All positive internal curvatures indicate a minimum, one negative a first-order saddle and multiple negatives a higher-order saddle. Proper classification needs convergence, projection of overall motions, frequency or eigenvector inspection and path checks to establish chemical meaning.

Practice questions

1. A stationary point has two negative internal Hessian eigenvalues. What is its index? Answer: Index two, usually called a second-order saddle.

2. Why is a zero gradient not enough to call a structure stable? Answer: A saddle also has zero gradient but can lower energy along one or more directions.

3. What do six near-zero Cartesian modes normally represent for an isolated nonlinear molecule? Answer: Three overall translations and three overall rotations.

4. Why should a small negative frequency be inspected rather than accepted immediately? Answer: It could reflect a real soft instability or numerical noise from loose optimisation, integration or nearly free motion.

Sources: IUPAC, potential-energy reaction surface concepts; First-principles reaction dynamics and stationary points, Journal of Physical Chemistry A; Simons, normal modes of vibration.