Minimal and Split-Valence Basis Sets

A single Gaussian function cannot mimic the sharp cusp of a Slater orbital at the nucleus. So quantum chemists glue several Gaussians together into one fixed bundle called a contracted Gaussian function. They then decide how many such bundles to spend on each atomic orbital. That single decision, more than almost any other choice in a calculation, determines whether a Hartree–Fock or density-functional run finishes in a second or an hour. It also determines whether the resulting bond length is embarrassingly wrong or usefully accurate. This article is about that decision: the difference between a minimal basis set, which allots exactly one contracted function per atomic orbital, and a split-valence basis set, which recognizes that the chemically active valence shell deserves more flexibility than the inert, tightly bound core. The story matters because basis sets are not free. Every added function multiplies the cost of the two-electron integral evaluation and the size of the Fock matrix. Yet skimping on them corrupts every energy difference downstream: bond dissociation energies, reaction barriers, vibrational frequencies, NMR shifts. Minimal sets like STO-3G were the first practical compromise. They were built explicitly to approximate Slater-type orbitals using three Gaussians each. They were cheap enough to run on 1970s hardware but rigid enough to badly mis-describe bond angles and polarizabilities. Split-valence sets like 3-21G and 6-31G fixed the worst of this. They gave each valence orbital two independent radial pieces — an "inner" and an "outer" contraction. This lets self-consistent-field optimization dial in how diffuse or contracted the electron cloud becomes as the molecular environment changes. The big picture is a trade-off triangle: accuracy, transferability, and cost. A minimal basis is fast and qualitatively sensible for closed-shell organic skeletons. But it fails badly for anions, hydrogen bonds, or anything where the electron density needs to breathe. A split-valence basis roughly doubles the valence functions. For a modest cost increase, it recovers most of the missing radial flexibility. It correctly captures how bonds elongate or contract, how charge polarizes toward electronegative atoms, and how orbital sizes respond to the SCF field. This article builds the contraction machinery from the ground up — how Gaussians are fit and glued. It then works through STO-3G, 3-21G, and 6-31G function by function. This way, the counting of primitives and contracted functions becomes second nature, rather than a black box borrowed from a software manual. Understanding this material presumes familiarity with the basic contest between Basis Sets: Slater vs Gaussian Functions. There, Slater-type orbitals get the correct nuclear cusp and long-range decay, but Gaussians win on integral tractability. Everything here builds directly on that trade-off. It also sets up the next natural refinement, adding polarization and diffuse functions, once the valence-splitting idea is secure.

Key takeaways

Gaussian primitives are contracted — fixed linear combinations — to mimic the shape of Slater orbitals cheaply; STO-3G uses exactly three primitive Gaussians per Slater-type orbital.
A minimal basis set has one contracted function per atomic orbital occupied in the free atom (1s for H/He; 1s,2s,2p for Li–Ne).
A split-valence basis doubles (or more) the valence functions only — core orbitals stay minimal/contracted, valence orbitals get inner + outer pieces.
The notation $6\text{-}31G$ reads: core = 6 primitives in 1 contraction; valence = inner shell of 3 primitives + outer shell of 1 primitive.
Splitting the valence lets SCF vary the radial extent of the density — essential for bond-length accuracy, dipole moments, and charge polarization.
Key idea: contraction trades variational flexibility for speed; minimal sets fix the radial shape of every orbital rigidly, while split-valence sets leave the valence shell's size as an SCF-optimizable degree of freedom.

Where this lives today

Split-valence bases such as 6-31G and 3-21G remain the default teaching and screening tools in undergraduate and industrial computational chemistry, used to obtain quick geometries and vibrational frequencies before a more expensive polarized or correlated calculation is run. Pharmaceutical modeling pipelines routinely pre-optimize candidate drug conformers at HF or DFT/6-31G(d) level (the "(d)" denoting an added polarization function, the natural next step beyond plain split-valence) before submitting the most promising poses to costlier correlated or QM/MM treatments in a binding-site environment. Materials chemists use split-valence basis sets in periodic and cluster models of surfaces and defects when screening large candidate libraries, reserving triple-zeta or correlation-consistent bases for final, publication-quality numbers. In spectroscopy, split-valence-level vibrational frequency calculations — scaled by empirical factors to correct for basis-set and anharmonicity errors — are a routine first pass in assigning infrared and Raman spectra of small organic molecules. Even in modern machine-learned interatomic potentials, training data are frequently generated at a hierarchy of basis-set levels starting from split-valence quality, precisely because the qualitative valence-flexibility argument developed here still determines whether the reference data faithfully captures how bonds respond to stretching, bending, and polarization by neighboring atoms. The concepts here also connect directly to Polarization and Diffuse Functions, the natural next refinement once radial valence splitting is understood, and to broader questions of basis-set convergence explored whenever Basis Sets: Slater vs Gaussian Functions is extended toward correlation-consistent or complete-basis-set extrapolation schemes.

Ask the tutor: Why does adding polarization functions, rather than further splitting the valence shell, become the next necessary step for accurately predicting bond angles and hybridization effects?

The physical picture — thorough prose

Every atomic orbital in a Hartree–Fock or Kohn–Sham calculation is represented as a linear combination of fixed, pre-built basis functions, and each basis function is itself built from primitive Gaussian-type functions,

g_{\alpha}(\mathbf{r}) = N x^l y^m z^n e^{-\alpha r^2},

centered on a nucleus, where $\alpha$ is the orbital exponent controlling how sharply the Gaussian falls off, and $(l,m,n)$ set the angular shape (s, p, d, ...). A single Gaussian, however, has zero slope at $r=0$ and falls off as $e^{-\alpha r^2}$ — far too fast at long range and far too flat at the nucleus compared to the true Slater-type orbital's cusp, $e^{-\zeta r}$ . The fix, introduced by Pople and coworkers, is contraction: fit a fixed sum of several primitive Gaussians of different widths so the sum approximates a Slater orbital's shape,

\chi(\mathbf{r}) = \sum_{p=1}^{K} d_p\, g_{\alpha_p}(\mathbf{r}),

where the contraction coefficients $d_p$ and exponents $\alpha_p$ are determined once (by least-squares fitting to a Slater orbital, or by atomic Hartree–Fock optimization) and then frozen for every calculation that uses that basis set. The resulting object $\chi$ is called a contracted Gaussian-type function (CGTF), and it behaves computationally like a single basis function even though the electron-repulsion integrals are still evaluated over all $K$ primitives underneath. This is the central economy of modern basis sets: chemists pay the steep integral cost of many primitives only once, at basis-set-construction time, and the SCF procedure itself only ever has to build a Fock matrix over the small number of contracted functions. The question that separates a minimal basis from a split-valence basis is: how many contracted functions represent each atomic orbital? In a minimal basis, the answer is always one. Hydrogen and helium get one function for 1s; lithium through neon get one function each for 1s, 2s, 2p $_x$ , 2p $_y$ , 2p $_z$ — five contracted functions total, each built (in STO-3G) from three primitives, so 15 primitive Gaussians altogether for a second-row atom. This is compact but rigid: the shape of the 2s or 2p orbital is permanently fixed by the fitted exponents and coefficients, no matter what molecular environment the atom finds itself in. A carbon atom in a stiff triple bond and a carbon atom in a loose van der Waals complex get exactly the same 2p radial function. Split-valence sets attack precisely this rigidity, but only where chemistry actually happens: the valence shell. Core orbitals (1s for a second-row atom) stay as a single tight contraction, because core electrons barely respond to bonding — they are essentially atomic no matter the molecule. Valence orbitals, by contrast, are represented by two independent contracted functions of different radial extents: a tighter "inner" valence function built from several primitives, and a looser "outer" valence function, often just one primitive, that can expand or contract during the SCF variational optimization. Because both functions enter the same LCAO expansion with independently optimized coefficients, the effective valence orbital,

\phi_{\text{val}} = c_{\text{inner}}\,\chi_{\text{inner}} + c_{\text{outer}}\,\chi_{\text{outer}},

can smoothly interpolate between compact (large $c_{\text{inner}}$ ) and diffuse (large $c_{\text{outer}}$ ) character depending on what the molecule demands — a bonding carbon in ethane, an anionic oxygen in hydroxide, an aromatic $\pi$ system in benzene, all draw different ratios of $c_{\text{inner}}:c_{\text{outer}}$ from the same fixed basis. This single extra variational handle per valence orbital is why split-valence sets so dramatically improve bond lengths, dipole moments, and vibrational frequencies relative to minimal sets, at a valence-only (not core) increase in function count.

Figure

Contraction: three primitive Gaussians summed into one 1s STO-3G function

Figure

Minimal vs split-valence: function count for a carbon atom

The core ideas — governing relationships

\chi_\mu(\mathbf{r}) = \sum_{p=1}^{K_\mu} d_{p\mu}\, g_{\alpha_{p\mu}}(\mathbf{r})

where: $\chi_\mu$ is contracted basis function $\mu$ ; $K_\mu$ is its contraction length (number of primitives glued together); $d_{p\mu}$ and $\alpha_{p\mu}$ are fixed contraction coefficients and exponents, determined once by fitting and never varied during a molecular SCF calculation. Reading: this equation says a basis function is a frozen recipe, not a variational object — only the coefficients multiplying entire contracted functions $\chi_\mu$ in the molecular orbital expansion are optimized by SCF, never the $d_{p\mu}$ or $\alpha_{p\mu}$ themselves.

N_{\text{basis}}^{\text{minimal}} = \sum_{\text{atoms}} n_{\text{occ. AOs in free atom}}

where: the sum runs over every atom in the molecule, and $n_{\text{occ. AOs}}$ counts the atomic orbitals occupied (or partially occupied, e.g. 2p on B) in the ground-state free atom — 1 for H/He, 5 for Li–Ne (1s,2s,2p $_{x,y,z}$ ), 9 for Na–Ar (adding 3s,3p). Reading: a minimal basis is entirely determined by simple atomic electron configuration counting — no bonding information enters at all.

N_{\text{basis}}^{\text{split-valence}} = \sum_{\text{atoms}} \Big( n_{\text{core AOs}} + 2\, n_{\text{valence AOs}} \Big)

where: $n_{\text{core AOs}}$ are the atomic orbitals not in the outermost occupied shell (e.g. 1s for carbon), left as a single contracted function each; $n_{\text{valence AOs}}$ are outermost-shell orbitals (2s, 2p for carbon), each duplicated into inner + outer contracted functions. Reading: the "2" appears only in front of valence orbitals — this is the algebraic definition of "split valence," and it is exactly what the $6\text{-}31G$ / $3\text{-}21G$ naming encodes.

6\text{-}31G:\quad \underbrace{6}_{\text{core, 1 contraction}} \; \text{-} \; \underbrace{3}_{\text{valence inner, 1 contraction}}\underbrace{1}_{\text{valence outer, 1 primitive}}\; G

where: the first number (before the dash) gives the number of primitive Gaussians contracted into the single core function; the two numbers after the dash give the primitive count in, respectively, the inner and outer valence contractions. Reading: reading the label directly off the periodic-table-style name tells you the exact primitive/contracted-function recipe without consulting any external table.

Derivation ACounting basis functions: STO-3G water vs 6-31G water

Goal. Determine the total number of contracted basis functions for a water molecule, H $_2$ O, under STO-3G and under 6-31G, and show explicitly why the difference is entirely in the valence shell.

Step 1 — Free-atom electron configurations. Oxygen: $1s^2\,2s^2\,2p^4$ , so occupied AOs are 1s (core), 2s, 2p $_x$ , 2p $_y$ , 2p $_z$ (valence) — 5 AOs total, 1 core + 4 valence. Hydrogen: $1s^1$ , one AO, which for hydrogen is conventionally treated as its only shell (no separate "core"); under split-valence conventions this whole 1s counts as "valence-like" and gets duplicated too, since H has no closed inner shell to exclude.

Step 2 — STO-3G count (minimal: one function per AO). Oxygen contributes 5 contracted functions (1s, 2s, 2p $_x$ , 2p $_y$ , 2p $_z$ ), each built from 3 primitives. Each hydrogen contributes 1 contracted function (1s), also 3 primitives.

N^{\text{STO-3G}}_{\text{H}_2\text{O}} = 5_{\text{O}} + 1_{\text{H}} + 1_{\text{H}} = 7 \ \text{contracted functions.}

Total primitives underneath: $5\times3 + 1\times3 + 1\times3 = 21$ primitive Gaussians.

Step 3 — 6-31G count (split-valence: double the valence, keep core singular). Oxygen's 1s core stays a single function (6 primitives, per the "6" in 6-31G). Oxygen's four valence AOs (2s, 2p $_x$ , 2p $_y$ , 2p $_z$ ) each split into an inner (3-primitive) and outer (1-primitive) piece — that is $4\times2 = 8$ functions.

N^{\text{O}}_{\text{6-31G}} = 1_{\text{core}} + 8_{\text{valence(split)}} = 9.

Hydrogen has no core shell to exclude, so its single 1s is treated purely as valence and is split into inner (3 primitives) + outer (1 primitive):

N^{\text{H}}_{\text{6-31G}} = 2 \ \text{per hydrogen atom}.

Step 4 — Assemble the molecular total.

N^{\text{6-31G}}_{\text{H}_2\text{O}} = 9_{\text{O}} + 2_{\text{H}} + 2_{\text{H}} = 13 \ \text{contracted functions.}

Step 5 — Compare and interpret.

\frac{N^{\text{6-31G}}}{N^{\text{STO-3G}}} = \frac{13}{7} \approx 1.86,

nearly double the basis size, but the increase is concentrated entirely in the valence functions that actually participate in O–H bonding and lone-pair polarization — the physically inert core 1s of oxygen stayed a single function in both bases. This is the precise sense in which split-valence spends its extra cost where chemistry happens.

Go deeper · Derivation B · 2 minWhy splitting the valence shell changes the variational bond length

Derivation BWhy splitting the valence shell changes the variational bond length

Goal. Show algebraically that a single fixed-shape valence orbital cannot adjust its radial extent, while two independently-weighted contracted functions can — and that this directly affects the energy-minimizing bond length obtained from the variational principle.

Step 1 — Minimal-basis valence orbital is rigid. In a minimal basis, the 2p valence orbital of, say, a carbon atom in a C–H bond is

\phi_{2p} = 1\cdot \chi_{2p}^{\text{STO-3G}},

where $\chi_{2p}^{\text{STO-3G}}$ is a fixed sum of three primitives with exponents chosen to fit a free-atom Slater orbital of exponent $\zeta \approx 1.6$ . There is no coefficient available to reshape $\chi_{2p}$ — its radial extent (its "spread," roughly $\langle r^2\rangle^{1/2}$ ) is fixed at whatever the free-atom fit produced, regardless of whether the carbon is in methane, ethylene, or a stretched transition state.

Step 2 — Split-valence orbital has one extra variational parameter. With split-valence functions $\chi_{\text{in}}$ (tighter, larger effective exponent) and $\chi_{\text{out}}$ (more diffuse, smaller effective exponent),

\phi_{2p} = c_{\text{in}}\,\chi_{\text{in}} + c_{\text{out}}\,\chi_{\text{out}},

and both $c_{\text{in}}$ and $c_{\text{out}}$ are determined by the Roothaan–Hall SCF equations, i.e. by minimizing the total electronic energy variationally. The effective radial spread of $\phi_{2p}$ ,

\langle r^2 \rangle_{\phi} = c_{\text{in}}^2 \langle r^2\rangle_{\text{in}} + c_{\text{out}}^2\langle r^2\rangle_{\text{out}} + 2\,c_{\text{in}}c_{\text{out}}\langle \chi_{\text{in}}|r^2|\chi_{\text{out}}\rangle,

now depends on the ratio $c_{\text{out}}/c_{\text{in}}$ , which SCF is free to adjust independently for every molecule and every internuclear distance.

Step 3 — Consequence for the potential energy curve. Consider the electronic energy as a function of bond length $R$ , $E(R)$ , obtained by re-solving SCF at each $R$ . In a minimal basis, at every $R$ the valence orbital shape is frozen to the free-atom value; only the LCAO mixing coefficients between atoms can respond to $R$ , not the intrinsic radial profile of each atomic contribution. This under-relaxation makes the true minimum of $E(R)$ harder to reach, because the basis cannot simultaneously satisfy "orbital is compact near equilibrium" and "orbital expands as bond stretches."

Step 4 — With splitting, the minimum is reachable by ordinary variational optimization. Because $c_{\text{in}}(R)$ and $c_{\text{out}}(R)$ are re-optimized at every geometry (they are ordinary SCF coefficients, not fixed parameters), the effective orbital radius $\langle r^2\rangle_\phi^{1/2}$ becomes an implicit function of $R$ that the variational principle adjusts automatically, driving $E(R)$ closer to the true (complete-basis) curve. The variational theorem guarantees

E^{\text{split-valence}}(R) \le E^{\text{minimal}}(R) \quad \text{at every } R,

since the split-valence basis strictly contains more variational freedom (setting $c_{\text{out}}=0$ recovers something close to, though not identical to, the minimal-basis limit). Because this energy lowering is not uniform in $R$ — it is larger near equilibrium, where orbital reshaping matters most, and smaller at long range — the location of the minimum, $R_e$ , shifts, generally toward the experimentally correct value.

Step 5 — Limiting check. If the outer coefficient is forced to zero ( $c_{\text{out}} \to 0$ ) at every geometry, the split-valence basis degenerates algebraically to a single scaled inner function, reproducing (up to the fixed exponent ratio) the rigidity of the minimal-basis case; this confirms the derivation reduces correctly to the minimal-basis limit when the extra variational freedom is artificially switched off.

Worked examples

Example 1 (Foundational): Counting primitives and contracted functions for STO-3G methane. Methane, CH $_4$ , has one carbon (1s²2s²2p²... configuration, ground state $1s^2 2s^2 2p^2$ , four valence electrons in 2s and 2p) and four hydrogens. Carbon's occupied AOs in STO-3G: 1s, 2s, 2p $_x$ , 2p $_y$ , 2p $_z$ — five contracted functions, each a sum of 3 primitives, so $5 \times 3 = 15$ primitives on carbon. Each hydrogen: one contracted 1s function from 3 primitives.

N_{\text{basis}}(\text{CH}_4,\ \text{STO-3G}) = 5_{\text{C}} + 4\times 1_{\text{H}} = 9 \ \text{contracted functions.}

N_{\text{primitives}}(\text{CH}_4,\ \text{STO-3G}) = 15_{\text{C}} + 4\times 3_{\text{H}} = 27 \ \text{primitive Gaussians.}

Cross-check by a different route: total occupied AOs summed over free atoms is $5 + 4(1) = 9$ , matching directly the minimal-basis formula $N^{\text{minimal}} = \sum n_{\text{occ AOs}}$ — consistent since STO-3G is, by definition, minimal.

Example 2 (Applied): Comparing STO-3G and 6-31G basis sizes for formaldehyde, and estimating the integral cost ratio. Formaldehyde is H $_2$ C=O: one carbon, one oxygen, two hydrogens. STO-3G (minimal): Carbon: 5 functions. Oxygen: 5 functions. Hydrogens: $2\times 1 = 2$ functions.

N^{\text{STO-3G}} = 5 + 5 + 2 = 12.

6-31G (split-valence): Carbon: core 1s (1 function) + valence 2s,2p $_x$ ,2p $_y$ ,2p $_z$ split into inner+outer ( $4\times2 = 8$ ) = 9 functions. Oxygen: same pattern, 9 functions. Hydrogen (no core to exclude, 1s treated as valence, split into 2): $2\times 2 = 4$ functions.

N^{\text{6-31G}} = 9_{\text{C}} + 9_{\text{O}} + 4_{\text{H}} = 22.

Ratio of basis sizes:

\frac{N^{\text{6-31G}}}{N^{\text{STO-3G}}} = \frac{22}{12} = 1.83.

Two-electron repulsion integrals scale formally as $O(N^4)$ (before screening), so a crude estimate of the raw integral-count ratio is

\left(\frac{22}{12}\right)^4 \approx 11.3,

meaning the naive (unscreened) number of two-electron integrals needed for 6-31G formaldehyde is roughly an order of magnitude larger than for STO-3G — the price paid for a valence shell that can now polarize toward the electronegative oxygen and correctly lengthen or shorten the C=O bond as the molecular field demands. In practice modern integral screening reduces this scaling well below the formal $N^4$ , but the qualitative message — split-valence costs several times more than minimal, even though it only doubles the valence count — stands.

Example 3 (Exam/competition level): Predicting and verifying the basis-function count for the azide ion, N $_3^-$ , in 3-21G, and identifying why the ion (not merely the count) demands more than a minimal treatment. Azide is a linear, symmetric triatomic anion, N=N=N with overall charge $-1$ ; the charge does not change the number of basis functions (basis sets are built per atom-type, independent of the molecule's total charge), but it strongly affects why split-valence flexibility is essential here. Step A — atomic composition. Three nitrogen atoms, each with free-atom configuration $1s^2 2s^2 2p^3$ : one core AO (1s) and four valence AOs (2s, 2p $_x$ , 2p $_y$ , 2p $_z$ ). Step B — 3-21G recipe per nitrogen. The label $3\text{-}21G$ parses as: core = 3 primitives in a single contraction; valence = inner (2 primitives) + outer (1 primitive). So each nitrogen contributes

1_{\text{core}} + 2\times 4_{\text{valence, split}} = 1 + 8 = 9 \ \text{contracted functions.}

Step C — total for N $_3^-$ .

N^{\text{3-21G}}_{\text{N}_3^-} = 3 \times 9 = 27 \ \text{contracted functions.}

Step D — compare to the minimal (STO-3G) count for the same ion, as a sanity check on the multiplier: each nitrogen in STO-3G contributes 5 functions (1s,2s,2p $_x$ ,2p $_y$ ,2p $_z$ ), so

N^{\text{STO-3G}}_{\text{N}_3^-} = 3\times 5 = 15,

giving a ratio $27/15 = 1.8$ , consistent with the general pattern (core untouched, valence doubled) seen for oxygen and carbon above — confirming the counting rule is atom-independent and depends only on core/valence AO bookkeeping, not on molecular charge or symmetry. Step E — why the extra flexibility specifically matters for an anion. The N $_3^-$ ion carries a full extra electron relative to the neutral, delocalized over three nitrogen 2p $_\pi$ systems and the in-plane lone pairs of the terminal nitrogens; an anion's electron density is characteristically more diffuse than a neutral atom's, because the added electron is bound in a shallower effective potential (weaker net nuclear attraction per electron). A minimal basis, whose 2p exponents were fit to neutral-atom Hartree–Fock densities, has valence functions that are systematically too tight for this anion. Splitting the valence into inner + outer contracted functions at least gives the SCF procedure the freedom to weight the more diffuse "outer" primitive more heavily ( $c_{\text{outer}}$ larger than it would be for a neutral atom's cation-like case), partially compensating for the missing genuinely diffuse functions. This is real progress over STO-3G, though — as the derivation below foreshadows — even split-valence sets remain systematically too contracted for anions until still more diffuse functions are added explicitly, which is exactly the next refinement beyond this one.

Limiting cases, scope & common traps

Situation	What happens	Why
Neutral, closed-shell hydrocarbon skeleton, qualitative geometry only	STO-3G often "good enough"	Rigid valence shapes still capture gross $\sigma$ -bonding topology
Anions, radicals, hydrogen-bonded complexes	STO-3G and even 6-31G under-bind / mis-predict structure	Valence orbitals fit to neutral-atom densities are too contracted; no diffuse functions present
Rotational barriers, small energy differences (< 5 kcal/mol)	Minimal basis unreliable; split-valence markedly better, still imperfect	Energy differences are exquisitely sensitive to valence radial flexibility
Bond lengths near equilibrium for main-group molecules	6-31G/3-21G typically within a few pm of experiment; STO-3G systematically too short	Split valence lets SCF relax orbital extent with $R$ (see derivation above)
Third-row and heavier atoms, or transition metals	Both minimal and split-valence (without polarization/relativistic treatment) are qualitatively poor	Core–valence orbital energy gaps and relativistic contraction not captured by simple s/p splitting
Angular flexibility (bond angles, hybridization changes)	Neither STO-3G nor 6-31G alone captures this well	Splitting only touches radial extent, not angular shape — that requires polarization functions

The chemist behind it

John Pople was a mathematician-turned-quantum-chemist whose defining contribution was making molecular electronic structure calculations programmable and systematic rather than bespoke. Trained at Cambridge in mathematics, Pople moved into theoretical chemistry via early work on liquid-water structure and NMR chemical shift theory before turning, in the late 1960s and 1970s at Carnegie Mellon University, to building a hierarchy of standardized Gaussian basis sets and the Gaussian software that implemented them. His group's papers defining STO-nG (minimal, Slater-fit) and then the split-valence families 4-31G, 6-31G, and (with collaborators including Warren Hehre, Robert Ditchfield, and Krishnan Raghavachari) 3-21G, turned basis-set choice from an idiosyncratic art into a reproducible, catalog-like decision — any two research groups anywhere could run "HF/6-31G" on the same molecule and get comparable numbers. This standardization, as much as any single formula, is why ab initio quantum chemistry became a routine tool across organic, inorganic, and biological chemistry rather than remaining a specialist's craft.

Origins · 1 minJohn Anthony Pople and how this was discovered

John Anthony Pople

1925–2004 · Mid-to-late 20th century development of computational quantum chemistry, Carnegie Mellon University and Northwestern University

Pople shared the 1998 Nobel Prize in Chemistry (with Walter Kohn) "for his development of computational methods in quantum chemistry." His Gaussian series of basis sets and the associated Gaussian program made routine ab initio calculations accessible to non-specialist chemists worldwide, transforming quantum chemistry from a purely theoretical pursuit into a standard bench tool for interpreting and predicting molecular structure and reactivity.

History

- 1951 — Boys shows Gaussian-type functions permit closed-form evaluation of molecular integrals, launching Gaussian-basis quantum chemistry.

- 1969 — Hehre, Stewart, and Pople introduce STO-nG minimal basis sets, fitting contracted Gaussians to Slater-orbital shapes.

- 1971 — Ditchfield, Hehre, and Pople publish 4-31G, an early split-valence basis separating inner and outer valence contractions.

- 1980 — Pople and coworkers introduce 6-31G, refining the split-valence recipe used across the Gaussian program family.

- 1982 — Binkley, Pople, and Gordon introduce 3-21G, a smaller, faster split-valence set widely used for larger systems and geometry surveys.

Why is STO-3G called "minimal" if it uses three Gaussian primitives per orbital?

"Minimal" refers to the number of contracted functions per atomic orbital, not the number of underlying primitives. STO-3G assigns exactly one contracted function (built from three primitives glued together) to each atomic orbital occupied in the free atom — one function for hydrogen's 1s, five for carbon's 1s/2s/2p set. The "3" in the name only describes how many primitives were fit together to approximate each Slater-type orbital's shape; it says nothing about how many independent contracted functions represent each orbital, which is what the minimal/split-valence distinction is actually about.

Does adding split-valence flexibility ever make results worse than minimal basis, or is it strictly an improvement?

For the electronic energy at a fixed geometry, the variational principle guarantees split-valence energies are never higher than minimal-basis energies, since split-valence strictly contains more flexibility. However, for derived properties like geometry-optimized bond angles or specific spectroscopic constants, going from STO-3G to 6-31G can occasionally shift a property further from experiment for an individual molecule, because both bases are missing polarization functions and error cancellation in the minimal basis was fortuitous. The energy comparison is rigorous; individual property comparisons are not, since real accuracy also depends on angular flexibility that neither basis provides.

Why don't we just always use a huge basis set instead of worrying about minimal vs split-valence?

Cost. Two-electron repulsion integral evaluation formally scales as roughly the fourth power of the number of basis functions, so doubling the basis size can raise raw computational cost by an order of magnitude before integral screening kicks in. For large molecules — proteins, materials slabs, reaction networks with dozens of stationary points — minimal or split-valence bases remain the only tractable choice for initial screening, geometry surveys, or generating training data, with more expensive polarized or correlation-consistent bases reserved for final, publication-quality single-point energies on the most important structures.

Does splitting the valence shell fix the wrong asymptotic behavior of Gaussian functions at long range?

No. Even an "outer" valence contracted function in 6-31G is still built from one or more Gaussian primitives, which decay as $e^{-\alpha r^2}$ , always faster than the exact $e^{-\zeta r}$ Slater-type decay at long range. Splitting gives the SCF procedure a diffuseness knob by mixing a more spread-out Gaussian with a tighter one, which helps approximate a slower decay in an effective sense over the chemically relevant region, but it does not repair the fundamentally wrong Gaussian tail. Genuinely diffuse functions, with very small fixed exponents added explicitly for this purpose, are needed for anions and long-range properties.

How is the number "9" for oxygen in 6-31G actually built from primitives, step by step?

Oxygen's core 1s is one contracted function built from 6 primitive Gaussians (the "6" before the dash). Its four valence orbitals — 2s, 2p_x, 2p_y, 2p_z — are each split into an inner contraction of 3 primitives and an outer contraction of 1 primitive (the "31" after the dash), giving $4 \times 2 = 8$ valence contracted functions. Adding the single core function gives $1 + 8 = 9$ total contracted functions for oxygen, built from $6 + 4\times(3+1) = 22$ primitive Gaussians underneath.

Is 3-21G simply a smaller, cheaper version of 6-31G with no other differences?

Broadly yes in spirit — both are split-valence sets from the Pople group with a single core contraction and a two-piece valence split — but the specific exponents and contraction coefficients were independently optimized for each set, and the core contraction lengths differ (3 primitives for 3-21G's core versus 6 for 6-31G's). Because 3-21G's core is described by fewer primitives, it is noticeably cheaper and often used for quick geometry pre-optimization on larger systems, while 6-31G's more flexible core description and generally better-tested valence exponents make it the more common choice for production-quality single-point energies at the split-valence level.

Why does a minimal basis particularly struggle with hydrogen bonding and dipole moments?

Hydrogen bonding and molecular dipole moments both depend sensitively on how far electron density can polarize away from a nucleus toward an electronegative partner or away from a positive one. A minimal basis fixes each valence orbital's radial shape to a free-atom fit, with no mechanism for the SCF density to become more diffuse toward a hydrogen-bond acceptor or more contracted near an electronegative donor. Split-valence bases introduce exactly this missing degree of freedom via the outer contracted function's independently optimizable coefficient, which is why 6-31G-level dipole moments and hydrogen-bond geometries are systematically closer to experiment than STO-3G's.