Variational Principle - Physical Chemistry 2

SECTION 1

Historical Context & Motivation

The early decades of quantum mechanics delivered spectacular successes for exactly solvable systems—the hydrogen atom, the harmonic oscillator, the particle in a box—yet these triumphs also exposed a profound limitation. The moment one considers a system with more than one electron, the Schrödinger equation becomes analytically intractable because the electron–electron repulsion term couples all coordinates and prevents exact separation of variables. Physicists and chemists therefore required systematic approximation methods that could guarantee reliable estimates of key observables—particularly the ground-state energy—without demanding a closed-form wavefunction. This need catalyzed the development of the variational principle, which became one of the two pillars of approximate quantum mechanics alongside perturbation theory.

1926 • Schrödinger's Wave Equation

Erwin Schrödinger publishes his wave equation, providing the mathematical framework for quantum mechanics. Exact solutions are quickly found for one-electron systems, but multi-electron atoms remain unsolved.

1928 • Ritz Variational Method

Walther Ritz formalizes the linear variational approach, expressing trial wavefunctions as linear combinations of basis functions and minimizing the energy functional. This becomes the backbone of modern computational chemistry.

1929 • Helium Ground-State Calculation

Egil Hylleraas applies the variational method to the helium atom using a trial wavefunction with explicit electron–electron distance dependence, achieving remarkable agreement with experiment and demonstrating the power of the principle.

1930 • Hartree–Fock Theory Emerges

Douglas Hartree and Vladimir Fock combine the variational principle with antisymmetrized Slater determinant trial functions, laying the foundation for self-consistent field methods used throughout quantum chemistry.

1950s–Present • Modern Computational Chemistry

The variational principle underpins density functional theory (DFT), configuration interaction, and essentially all modern electronic structure methods. Kohn and Pople share the 1998 Nobel Prize in Chemistry for variational-based computational advances.

The central question the variational principle addresses is deceptively simple: given a quantum system whose Hamiltonian is known but whose eigenstates are not, how can one obtain an upper bound on the ground-state energy using only a cleverly chosen guess for the wavefunction? The answer—elegant and powerful—is that any normalized trial function yields an expectation value of the Hamiltonian that is guaranteed to be greater than or equal to the true ground-state energy, and systematic optimization of that trial function drives the estimate toward the exact value from above.

SECTION 2

Core Principles & Definitions

The variational principle rests on a few interconnected ideas that together form a rigorous and intuitive framework. Understanding each principle individually, and how they relate, is essential before engaging with the mathematics. The key insight is that quantum mechanics provides a built-in quality metric for any approximate wavefunction: the expectation value of the Hamiltonian. No matter how poor the trial function, the energy it yields will never dip below the true ground-state energy, giving the method its characteristic one-sided error guarantee.

Trial Wavefunction

A parameterized guess ψ_trial(α, β, …) chosen based on physical intuition. It must satisfy the same boundary conditions as the true wavefunction and belong to the Hilbert space of the system.

Energy Functional

The expectation value E[ψ] = ⟨ψ|Ĥ|ψ⟩ / ⟨ψ|ψ⟩ maps any trial wavefunction to a single real number. This functional is the object we seek to minimize with respect to the variational parameters.

Variational Theorem

For any trial wavefunction, E[ψ_trial] ≥ E₀, where E₀ is the exact ground-state energy. Equality holds only when the trial function is identical to the true ground-state eigenfunction.

Optimization

Parameters in the trial function are varied to minimize E[ψ]. This is accomplished by setting ∂E/∂α = 0 for each parameter α. The minimum energy obtained is the best variational estimate.

Upper Bound Guarantee

Because E[ψ_trial] is always ≥ E₀, every improvement in the trial function lowers the energy estimate toward the true value. The method converges from above, providing a rigorous bound.

✦

KEY TAKEAWAY

Think of the variational principle like adjusting the shape of a satellite dish to maximize signal reception. The 'signal strength' you measure (the energy functional) is always weaker than or equal to the theoretical optimum (the true ground-state energy, which is the lowest possible). By tweaking the dish's curvature (the variational parameters), you systematically improve reception, and you know you can never overshoot the perfect alignment—the feedback is inherently one-sided. In quantum mechanics, the energy you calculate with any guess wavefunction is always above or equal to the true ground-state energy, so lower is always better.

SECTION 3

Visual Explanation

The following diagram illustrates the central logic of the variational principle. The energy landscape is plotted as a function of a single variational parameter α. The horizontal dashed line represents the exact ground-state energy E₀, which the variational energy E(α) can approach but never cross. The optimal parameter value α_opt corresponds to the minimum of the curve, yielding the best possible estimate within the chosen functional form.

The parabolic-like curve shows how the variational energy E(α) varies with the parameter α. The minimum of the curve at α_opt gives the best upper bound on the true ground-state energy E₀ (green dashed line). The pink vertical gap ΔE = E(α_opt) − E₀ is always non-negative.

Several features of this diagram deserve emphasis. First, the variational curve lies entirely above the exact ground-state energy line—this is the graphical manifestation of the variational theorem. Second, the curve possesses a single minimum at α_opt for a one-parameter trial function; with additional parameters, the landscape becomes multidimensional, but the upper-bound guarantee remains. Third, the gap ΔE between the variational minimum and E₀ quantifies how well the chosen functional form captures the true physics—a smaller gap indicates a more physically motivated trial wavefunction.

SECTION 4

Mathematical Framework

The mathematical foundation of the variational principle follows from the completeness of the eigenstates of a Hermitian Hamiltonian. Let {ψₙ} be the complete set of orthonormal eigenstates of Ĥ with eigenvalues E₀ ≤ E₁ ≤ E₂ ≤ ⋯. Any trial function can be expanded in this basis, and the resulting energy expectation value is a weighted average of eigenvalues—a weighted average that is necessarily at least as large as the smallest eigenvalue.

Statement of the Variational Theorem

VARIATIONAL THEOREM

E[ψ_trial] = ⟨ψ_trial | Ĥ | ψ_trial⟩ / ⟨ψ_trial | ψ_trial⟩ ≥ E₀

Here Ĥ is the full Hamiltonian, ψ_trial is any normalizable function in the domain of Ĥ, and E₀ is the exact ground-state energy.

Proof Sketch

Expand the normalized trial function in the eigenbasis: ψ_trial = Σₙ cₙ ψₙ, where Σₙ |cₙ|² = 1. The expectation value becomes E[ψ_trial] = Σₙ |cₙ|² Eₙ. Because each Eₙ ≥ E₀ and the weights |cₙ|² are non-negative and sum to one, we obtain E[ψ_trial] = Σₙ |cₙ|² Eₙ ≥ Σₙ |cₙ|² E₀ = E₀. This completes the proof.

EIGENSTATE EXPANSION

E[ψ_trial] = Σₙ |cₙ|² Eₙ ≥ E₀ × Σₙ |cₙ|² = E₀

Each coefficient cₙ = ⟨ψₙ|ψ_trial⟩ quantifies the overlap of the trial function with the n-th eigenstate.

Optimization Condition

STATIONARY CONDITION

∂E / ∂αᵢ = 0 for all variational parameters αᵢ

This yields a set of coupled equations whose solution gives the optimal parameters and the lowest achievable energy within the chosen trial function space.

Linear Variational Method

A particularly important special case arises when the trial wavefunction is a linear combination of basis functions: ψ_trial = Σᵢ cᵢ φᵢ. Minimizing the energy with respect to the coefficients cᵢ transforms the problem into a generalized matrix eigenvalue equation: Hc = ESc, where H_ij = ⟨φᵢ|Ĥ|φⱼ⟩ is the Hamiltonian matrix, S_ij = ⟨φᵢ|φⱼ⟩ is the overlap matrix, and the lowest eigenvalue provides the best linear variational estimate of E₀. This is the Ritz method and forms the mathematical backbone of Hartree–Fock and density functional theory calculations.

SECULAR EQUATION

det(H − ES) = 0

The roots E of this determinantal equation are the variational energies. The lowest root is the best approximation to the ground-state energy obtainable within the chosen basis set.

SECTION 5

Applications & Trial Function Design

Choosing a good trial wavefunction is both an art and a science. The variational principle guarantees convergence from above regardless of the trial form, but a physically motivated choice converges far more rapidly and yields deeper insight into the system's electronic structure. The diagram below compares two common trial strategies for the helium atom: a simple product of hydrogen-like 1s orbitals with an adjustable effective nuclear charge Z_eff, and a more sophisticated correlated trial function that includes explicit dependence on the electron–electron distance r₁₂.

Comparison of two trial wavefunction strategies for helium. Strategy A (screened hydrogenic) uses a single parameter and achieves ~2% error. Strategy B (Hylleraas-type) incorporates electron correlation explicitly and achieves sub-0.01% accuracy, illustrating the dramatic improvement possible with physically motivated trial functions.

The contrast between the two strategies underscores a fundamental lesson: the quality of the variational estimate depends critically on the physics encoded in the trial function. Strategy A captures the main effect of nuclear screening but completely neglects electron–electron correlation. Strategy B, by including r₁₂ explicitly, builds in the physical reality that electrons dynamically avoid each other. Modern computational chemistry extends this logic by systematically enlarging basis sets and incorporating correlation through post-Hartree–Fock methods—all within the variational framework.

ℹ

Basis Set Completeness

In the linear variational method, increasing the number of basis functions always improves (or at worst maintains) the variational energy. This is a direct consequence of the fact that adding basis functions enlarges the subspace in which the minimization is performed, so the minimum can only decrease or stay the same—never increase. This property is sometimes called the variational collapse theorem and is the theoretical justification for using ever-larger basis sets in modern electronic structure calculations.

SECTION 6

Worked Example: Helium Ground-State Energy

We now apply the variational principle to estimate the ground-state energy of helium using the simplest possible trial function: a product of two hydrogen-like 1s orbitals with an adjustable effective nuclear charge Z_eff. The true nuclear charge of helium is Z = 2, but each electron partially screens the other, so we expect Z_eff < 2.

Variational Estimate of He Ground-State Energy

1
Step 1 — Define the Trial Wavefunction

Choose ψ_trial(r₁, r₂) = φ_1s(Z_eff, r₁) × φ_1s(Z_eff, r₂), where φ_1s(Z_eff, r) = (Z_eff³ / π a₀³)^1/2 exp(−Z_eff r / a₀). The single variational parameter is Z_eff.
2
Step 2 — Write the Hamiltonian

The helium Hamiltonian (in atomic units, ℏ = mₑ = e = 1) is: Ĥ = −½∇₁² − ½∇₂² − Z/r₁ − Z/r₂ + 1/r₁₂, with Z = 2 for helium. We add and subtract Z_eff/r₁ and Z_eff/r₂ to separate the Hamiltonian into a solvable hydrogen-like part and a perturbative remainder.
3
Step 3 — Evaluate the Energy Expectation Value

Using known hydrogen-like integrals, the kinetic and nuclear attraction terms give E_H-like = Z_eff² − 2Z × Z_eff (per electron in Hartree units, accounting for both electrons). The electron–electron repulsion integral evaluates to ⟨1/r₁₂⟩ = (5/8)Z_eff. Collecting all terms: E(Z_eff) = Z_eff² − 2Z × Z_eff + (5/8)Z_eff. With Z = 2: E(Z_eff) = Z_eff² − 4Z_eff + (5/8)Z_eff = Z_eff² − (27/8)Z_eff.
4
Step 4 — Minimize with Respect to Z_eff

Take the derivative and set it to zero: dE/dZ_eff = 2Z_eff − 27/8 = 0, which gives Z_eff = 27/16 ≈ 1.6875.

Z_eff,opt = 27/16 = 1.6875
5
Step 5 — Calculate the Optimal Energy

Substitute Z_eff = 27/16 back into E(Z_eff): E_opt = (27/16)² − (27/8)(27/16) = −(27/16)² = −729/256 ≈ −2.848 Hartree. Converting to eV: E_opt ≈ −2.848 × 27.211 eV ≈ −77.5 eV.

E_opt ≈ −2.848 Hartree ≈ −77.5 eV
6
Step 6 — Compare with Experiment

The experimental ground-state energy of helium is −79.005 eV (equivalently, −2.9037 Hartree). Our variational estimate of −77.5 eV is above this value, consistent with the variational theorem. The difference of ~1.5 eV (~2%) arises primarily from neglecting electron–electron correlation in the simple product trial function.

Error ≈ 1.5 eV (≈ 2%) — consistent with E_var ≥ E₀

SECTION 7

Strengths, Limitations & Comparison with Perturbation Theory

The variational principle and perturbation theory are the two dominant approximation strategies in quantum mechanics, and understanding their respective domains of applicability is crucial for any practicing physical chemist. The following table summarizes the key distinctions between these complementary approaches.

Feature	Variational Principle	Perturbation Theory
Error bound	Provides a rigorous upper bound on E₀; the error is always non-negative.	No guaranteed bound direction; corrections may overshoot or undershoot.
Requires exactly solvable reference?	No — only a reasonable trial function and the ability to evaluate integrals.	Yes — requires a known zeroth-order Hamiltonian with known eigenstates.
Convergence	Monotonic from above as the trial space is enlarged.	Not guaranteed; series may diverge for strong perturbations.
Excited states	Accessible via the linear variational method (higher eigenvalues of the secular equation) but requires orthogonality constraints for nonlinear methods.	Straightforward for any state, provided the perturbation is small.
Physical insight	Trial function design encodes and tests physical hypotheses about the system.	Correction terms reveal which interactions are most important.
Computational cost	Can be expensive for large basis sets (matrix diagonalization scales as N³).	Low order corrections are inexpensive; high order corrections become costly.

Comparison of the variational principle and perturbation theory as approximation methods in quantum mechanics.

⚖️

WHEN TO USE WHICH

In practice, the two methods are often used together. Hartree–Fock theory, for instance, is a variational method that produces an optimized reference state, and then perturbation theory (such as MP2) is applied on top to capture the remaining electron correlation. Think of the variational method as building the best possible foundation for a house—it gets the shape and structure right—while perturbation theory adds the fine details like plumbing and wiring. Both are essential for a complete description, and the variational principle's upper-bound guarantee ensures you always know which direction your errors lie.

SECTION 8

Connection to Advanced Theory

The variational principle is not merely a computational tool—it connects deeply to the formal structure of quantum mechanics and extends into nearly every branch of modern theoretical chemistry and physics. The Hohenberg–Kohn theorems that underpin density functional theory (DFT) are, at their core, variational statements: the ground-state energy is a functional of the electron density, and the true ground-state density minimizes this functional. Similarly, configuration interaction (CI) is the linear variational method applied in a many-electron determinant basis, and the coupled-cluster method, though not strictly variational, can be reformulated in variational terms.

Concept	Variational Foundation	Key Extension
Hartree–Fock	Minimizes energy over all single Slater determinants. The Fock operator arises from the variational condition.	Self-consistent field (SCF) iteration; exchange interaction emerges naturally.
Configuration Interaction	Linear variational method in a basis of excited determinants. Full CI is exact within the basis set.	Truncation (CISD, CISDT) trades accuracy for computational feasibility.
DFT (Kohn–Sham)	Hohenberg–Kohn theorem: the ground-state energy is a variational functional of the electron density ρ(r).	Exchange-correlation functional approximations (LDA, GGA, hybrid) are active research areas.
Quantum Monte Carlo	Variational Monte Carlo (VMC) evaluates the energy functional stochastically with flexible trial functions.	Diffusion Monte Carlo (DMC) projects out the exact ground state from a variational guess.

How the variational principle connects to advanced computational methods in quantum chemistry.

Looking forward, the variational principle continues to find new applications in quantum computing, where the Variational Quantum Eigensolver (VQE) algorithm uses a parameterized quantum circuit as the trial function and a classical optimizer to minimize the energy expectation value. This hybrid classical-quantum approach is one of the most promising near-term applications of quantum computers to chemistry, and it relies directly on the same variational theorem proven decades ago. Understanding the variational principle therefore provides foundational knowledge that spans from pencil-and-paper estimates to cutting-edge quantum technologies.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A student uses a Gaussian trial function ψ(x) = A exp(−αx²) to estimate the ground-state energy of the one-dimensional harmonic oscillator. She obtains E(α_opt) = ½ℏω, which equals the exact ground-state energy. Does this violate the variational principle? Explain why or why not.

Show answer

No, this does not violate the variational principle. The variational theorem states E[ψ_trial] ≥ E₀, where equality holds if and only if the trial function is the exact ground-state eigenfunction. The true ground-state wavefunction of the harmonic oscillator is itself a Gaussian, ψ₀(x) ∝ exp(−mωx²/2ℏ). Therefore, when α is optimized, the trial function coincides exactly with the true eigenfunction, and equality in the variational inequality is achieved. This is a case where the trial function space happens to contain the exact solution, making the variational method exact.

Question 2: BASIC CALCULATION

For a one-dimensional particle in a box of length L, use the normalized trial function ψ(x) = (30/L⁵)^1/2 x(L − x) to compute the variational energy estimate. The Hamiltonian is Ĥ = −(ℏ²/2m) d²/dx². Compare your result to the exact ground-state energy E₁ = π²ℏ²/(2mL²).

Show answer

The trial function ψ(x) =

(30/L⁵)^{1/2}

x(L − x) satisfies the boundary conditions ψ(0) = ψ(L) = 0 and is already normalized. Applying the Hamiltonian: −(ℏ²/2m) d²/dx² [x(L − x)] = −(ℏ²/2m)(−2) = ℏ²/m. The energy expectation value is E = (30/L⁵) ∫₀ᴸ x(L−x)(ℏ²/m) dx = (30ℏ²)/(mL⁵) × L³/6 = 5ℏ²/(mL²). Numerically, E = 10ℏ²/(2mL²), while the exact value is π²ℏ²/(2mL²) ≈ 9.8696ℏ²/(2mL²). The ratio E/E₁ = 10/π² ≈ 1.013, so the variational estimate is only about 1.3% above the exact value—a remarkably good result for such a simple trial function, and it lies above E₁ as the theorem requires.

Question 3: INTERMEDIATE

Consider the hydrogen atom Hamiltonian and the trial function ψ(r) = A exp(−αr²), where α is a variational parameter (note: this is a Gaussian, not an exponential). Show that the kinetic energy expectation value is ⟨T⟩ = 3ℏ²α/(2m) and the potential energy expectation value is ⟨V⟩ = −e²(2α/π)^1/2 (in Gaussian units). Find the optimal α and the resulting energy. How does it compare to the exact hydrogen ground-state energy?

Show answer

With the Gaussian trial function, the known integrals yield ⟨T⟩ = 3ℏ²α/(2mₑ) and ⟨V⟩ = −e²

(2α/π)^{1/2}

. The total energy is E(α) = 3ℏ²α/(2mₑ) − e²

(2α/π)^{1/2}

. Setting dE/dα = 0: 3ℏ²/(2mₑ) = e²/

(2πα)^{1/2}

, solving gives α_opt = 8mₑ²e⁴/(9π ℏ⁴). Substituting back yields E_opt = −4mₑe⁴/(3πℏ²) = −(8/3π) × 13.6 eV ≈ −11.5 eV. The exact hydrogen ground-state energy is −13.6 eV, so the Gaussian trial function recovers about 85% of the binding energy. The 15% error reflects the fact that a Gaussian poorly represents the cusp at r = 0 that the true exponential wavefunction possesses. This illustrates how the functional form of the trial function—not just the optimization—critically determines accuracy.

Question 4: APPLIED

In a quantum chemistry computation on the lithium hydride molecule (LiH), a researcher performs variational calculations using STO-3G, 6-31G, and cc-pVTZ basis sets. The computed total energies are −7.862, −7.981, and −8.052 Hartree, respectively. The estimated complete-basis-set (CBS) limit is −8.070 Hartree. Explain the trend in these energies using the variational principle. Can any of these values be lower than the CBS limit?

Show answer

The variational principle guarantees that E[ψ_trial] ≥ E₀ for any trial function, and the linear variational method (used in all Hartree–Fock and CI calculations) ensures that enlarging the basis set can only lower or maintain the computed energy. The trend STO-3G (−7.862) > 6-31G (−7.981) > cc-pVTZ (−8.052) is consistent: each successively larger basis set expands the variational space, allowing a better approximation. The CBS limit (−8.070 Hartree) represents the energy at infinite basis set size within the given level of theory (e.g., Hartree–Fock or a specific CI truncation). None of these finite-basis results can be lower than the CBS limit because the CBS limit is the infimum of the variational energy over all possible basis functions at that level of theory. However, none of these—including the CBS limit—can be lower than the exact non-relativistic ground-state energy, since the Hartree–Fock CBS limit still neglects dynamical electron correlation.

Question 5: CRITICAL THINKING

The variational principle guarantees an upper bound on the ground-state energy. Prove that for excited states, the standard variational inequality E[ψ] ≥ E₀ is not sufficient to identify E₁ (the first excited-state energy). Then describe how the linear variational method addresses this limitation using the Hylleraas–Undheim–MacDonald theorem. Under what conditions does the second-lowest eigenvalue of the secular equation provide an upper bound on E₁?

Show answer

The standard variational theorem only guarantees E[ψ] ≥ E₀, which provides no information about excited states—a trial function that gives E > E₀ might approximate the ground state poorly rather than approximate any excited state well. For the first excited state, one would need ψ_trial to be orthogonal to ψ₀, but ψ₀ is generally unknown. The Hylleraas–Undheim–MacDonald (HUM) theorem resolves this for the linear variational method: if one diagonalizes an N×N Hamiltonian matrix in a basis of N functions, the k-th eigenvalue Eₖ^(N) provides an upper bound on the exact k-th eigenvalue Eₖ. Moreover, adding basis functions (going from N to N+1) can only lower each eigenvalue: Eₖ^(N+1) ≤ Eₖ^(N). The second-lowest eigenvalue of the secular equation thus provides an upper bound on E₁ provided the basis set spans a subspace that has nonzero overlap with the true first excited state. This theorem is essential for practical excited-state calculations and underpins the CI method's ability to target multiple states simultaneously.

SUMMARY

Summary & Key Concepts

Summary

The variational principle states that the energy expectation value of any normalized trial wavefunction is always greater than or equal to the true ground-state energy E₀. This upper-bound guarantee provides a rigorous one-sided error metric: any improvement in the trial function lowers the energy estimate toward the exact value. The proof follows from expanding the trial function in the Hamiltonian's eigenbasis and recognizing that the resulting weighted average of eigenvalues cannot be smaller than the smallest eigenvalue.

In practice, one constructs a parameterized trial function guided by physical intuition, evaluates the energy functional E[ψ] = ⟨ψ|Ĥ|ψ⟩/⟨ψ|ψ⟩, and minimizes with respect to the variational parameters. The linear variational method (Ritz method) casts this optimization as a matrix eigenvalue problem via the secular equation det(H − ES) = 0, forming the mathematical foundation of Hartree–Fock theory, configuration interaction, density functional theory, and modern variational quantum eigensolvers.