Potential energy surfaces

Result

Potential energy surfaces

T-113Home CU-401Threads quantum

Statement

Geometries, transition states and reaction paths.

Why it matters

hartree-fock and density-functional-theory, together with basis-sets' numerical representation of orbitals, give a way to compute the electronic energy for one single, fixed set of nuclear coordinates, via the Born-Oppenheimer separation. potential-energy-surfaces is the concept that assembles all of these single-point energies, computed across every possible nuclear geometry, into one continuous function — the object that actually determines molecular structure, vibrational frequencies, and reaction pathways, and the object computational chemistry is ultimately exploring whenever it optimises a geometry or searches for a mechanism.

Hypotheses

The Born-Oppenheimer approximation holds: nuclei move much more slowly than electrons, so electronic energy can be computed for each fixed nuclear arrangement and then treated as a potential under which the nuclei subsequently move.Without this separation, "a potential energy surface as a function of nuclear coordinates alone" would not even be a well-defined concept, since electronic and nuclear motion would remain fundamentally entangled. The surface is smooth (continuously differentiable) over the region of chemical interest.This is needed to define gradients (forces) and curvatures (force constants, vibrational frequencies) at any point, and to justify the gradient-based optimisation algorithms that locate stationary points. A single, well-defined electronic ground-state surface is non-degenerate and well-separated in energy from excited states at every geometry visited.Where two electronic states approach in energy or genuinely cross (a conical intersection, common in photochemistry), the single-surface picture breaks down entirely, and a proper treatment requires multiple, coupled surfaces.

Proof

E(\mathbf{q}) = E_{\text{elec}}(\mathbf{q}) \quad (\text{hartree-fock or density-functional-theory, for a chosen basis-sets representation})

For a molecule of \(N\) atoms, the potential energy surface is defined as the electronic energy evaluated as a function of the full set of \(3N{-}6\) independent internal nuclear coordinates \(\mathbf{q}\) (nonlinear molecule, after removing overall translation/rotation). A

\nabla E(\mathbf{q})=0

A stationary point on the surface has zero net force on every nucleus; the Hessian matrix of second derivatives at that point classifies it — all-positive eigenvalues indicate a genuine local minimum, exactly one negative eigenvalue indicates a first-order saddle point. B

\text{Minimum: all Hessian eigenvalues} > 0 \quad \Rightarrow \quad \text{stable geometry (reactant, product, intermediate)}

Geometry optimisation algorithms locate minima by following the local gradient downhill until \(\nabla E(\mathbf{q})\approx0\) to numerical tolerance. A

\text{Transition state: exactly one Hessian eigenvalue} < 0

Locating a transition state requires specifically seeking a first-order saddle point — uphill along exactly one direction, downhill along all others — a genuinely harder numerical search than simple minimisation, requiring specialised algorithms. B

\text{Frequency calculation: Hessian eigenvalues} \to \text{vibrational frequencies}

Evaluating the Hessian's eigenvalues at a stationary point and converting them, via the harmonic approximation, into vibrational frequencies confirms its character numerically — all-real, positive frequencies for a minimum, exactly one imaginary frequency for a transition state — and supplies the thermochemical corrections needed to convert a bare electronic energy into a genuine free energy. B

Result

\nabla E(\mathbf{q})=0,\quad \text{all-positive Hessian eigenvalues}\Rightarrow\text{minimum};\quad \text{exactly one negative eigenvalue}\Rightarrow\text{transition state}

Reading. The surface's stationary points are exactly the chemically meaningful geometries — stable species sit at minima, and the rate-determining transition states connecting them sit at first-order saddle points, distinguished purely by the sign pattern of local curvature.

Scope. Relies on the Born-Oppenheimer approximation and a smooth, well-separated electronic ground state; multi-step reactions correspond to paths crossing multiple minima and saddle points on the same surface; genuinely photochemical or near-degenerate-electronic-state problems require moving beyond a single surface.

Corollaries & converses

reaction-energy-profiles' qualitative energy diagrams are literally a one-dimensional slice, the minimum-energy path, through the full, multidimensional potential energy surface constructed here.
Geometry optimisation directly reuses hartree-fock or density-functional-theory to evaluate \(E(\mathbf{q})\) and its gradient at each trial geometry, tying this result concretely back to those two electronic-structure methods.
Frequency calculations' distinguishing of minima from transition states by counting imaginary frequencies gives a rigorous, purely numerical confirmation of a proposed mechanism's stationary points, in place of relying on chemical intuition alone.

Fails without

Violate the Born-Oppenheimer separability near a conical intersection (Hypotheses, third assumption): the single-surface picture itself breaks down, since nuclear motion can efficiently drive transitions between nearly degenerate electronic states, and a non-adiabatic, multi-surface treatment is required instead.
Assume smoothness and differentiability everywhere (Hypotheses, second assumption) near a genuine surface crossing or cusp: gradient-based optimisation algorithms fail to converge, or converge to a spurious point, since the local gradient information they rely on is no longer well-defined there.

Common errors

Assuming any stationary point found by an optimisation algorithm is automatically the desired minimum, without checking the Hessian to confirm it is not actually a saddle point.
Confusing a transition state (first-order saddle point, one imaginary vibrational frequency) with a reactive intermediate (a genuine local minimum, all-real frequencies, however shallow or short-lived).
Forgetting that optimisation algorithms generally find only the nearest local minimum to the starting structure, not necessarily the global minimum across the whole surface.
Treating the potential energy surface as literally, physically three-dimensional; beyond a diatomic or triatomic, the true surface is a function of many more coordinates than can be directly plotted, and visualisations show only a low-dimensional slice.

Discussion

The potential energy surface concept dates to the earliest days of quantum chemistry following the Born-Oppenheimer approximation's 1927 formulation, and became a practically computable object only with the growth of computational methods powerful enough to evaluate \(E(\mathbf{q})\) at the very large number of geometries a genuine surface exploration requires.

Near a conical intersection, the Born-Oppenheimer separation itself becomes a poor approximation, since nuclear motion can efficiently drive transitions between two nearly degenerate electronic states — a regime requiring non-adiabatic (surface-hopping or fully coupled multi-surface) treatments well beyond the single-surface picture developed here, and central to much of modern excited-state photochemistry.

Common misconception: that a computed transition state is a real, observable, isolable chemical species like a reactant or product. A transition state is instead a fleeting, formally zero-lifetime stationary configuration — a genuine energy maximum along the reaction-coordinate direction, unlike a true intermediate, which is a real local minimum with at least transient stability.

Worked examples

\text{HCN} \rightleftharpoons \text{HNC}

For this small unimolecular isomerisation, the relevant potential energy surface can be described by essentially a single internal coordinate (the H–C–N bond angle); the minimum corresponds to the stable, linear HCN structure, and a maximum along that one coordinate corresponds to the transition state connecting HCN to the less stable HNC isomer. A

\text{Frequency output: one imaginary frequency}

Confirming the candidate transition-state structure computationally requires exactly one imaginary vibrational frequency (Step 5), corresponding to motion along the reaction coordinate itself; any additional imaginary frequencies would indicate the structure is not a genuine first-order saddle point at all. A

\text{Minimum: all-real frequencies}; \qquad \text{Transition state: exactly one imaginary frequency}

Reading. A single numerical check — counting imaginary vibrational frequencies — distinguishes a genuine intermediate from a genuine transition state on the computed surface.

Scope. This diagnostic applies to any stationary point located on any computed potential energy surface, regardless of the specific electronic-structure method used to generate it.

Problems

A computed stationary point has Hessian eigenvalues \(\{+0.3, +0.5, +1.2\}\) (arbitrary units). Classify it.
Solution
All eigenvalues are positive, so this is a local minimum — a stable geometry.
Explain why finding a transition state computationally is generally harder than finding a minimum.
Solution
Finding a minimum only requires following the gradient downhill in every direction, a well-conditioned search converging from a wide range of starting points. Finding a transition state instead requires locating a point that is simultaneously a maximum along exactly one direction and a minimum along every other — a much more delicate condition, typically requiring specialised algorithms (e.g. following the lowest-curvature direction upward, or interpolation between reactant and product geometries) rather than simple downhill-only optimisation.
Using the Born-Oppenheimer approximation, explain why "potential energy surface" is meaningfully defined as a function of nuclear coordinates alone, without reference to electron coordinates.
Solution
The Born-Oppenheimer approximation (Hypotheses) allows the electronic Schrödinger equation to be solved separately for each fixed set of nuclear coordinates, since electrons respond essentially instantaneously to nuclear motion; the resulting electronic energy, once electron coordinates have been integrated out of the problem, depends only on the nuclear coordinates that were held fixed — exactly the function \(E(\mathbf{q})\) defined in Step 1, with no remaining electronic degrees of freedom appearing explicitly.