Goldschmidt Tolerance Factor: Predicting Perovskite Stability

What the Tolerance Factor Is and Where It Applies

The Goldschmidt tolerance factor is a purely geometric stability index for the perovskite structure, the ABX₃ lattice named after the mineral CaTiO₃. In an ideal perovskite, a large A cation sits at the cube corners (12-coordinate), a smaller B cation sits at the body center inside a corner-sharing octahedron of X anions (6-coordinate), and the X anions (oxide O²⁻ or halide Cl⁻, Br⁻, I⁻) occupy the face centers.

• A site: large, low-charge cations — Ca²⁺, Sr²⁺, Ba²⁺, La³⁺, or the molecular methylammonium cation CH₃NH₃⁺ in hybrid solar-cell perovskites.
• B site: smaller transition-metal or main-group cations — Ti⁴⁺, Fe³⁺, Nb⁵⁺, Pb²⁺, Sn²⁺.
• X site: O²⁻ in oxide perovskites, or a halide in Pb-halide photovoltaic perovskites.

The factor answers a screening question before any synthesis: given the ionic radii, does the A cation actually fill its 12-fold cavity? If it is too small, the octahedra tilt to close the gap; if it is too large, the perovskite cage cannot form at all. It is the first filter materials chemists apply when hunting for new ferroelectrics, ionic conductors, and photovoltaic absorbers.

Deriving the Formula Step by Step

The derivation is pure Pythagorean geometry on the ideal cubic cell. Consider the cube edge as the B–X–B axis. Along the cell edge, a B cation and an X anion touch, so the edge length a satisfies:

a = 2·(r_B + r_X)

Along the face diagonal, the A cation touches the X anions. A face diagonal has length a√2 and equals four times the sum of the A–X contact radii along that line:

a√2 = 2·(r_A + r_X)

For a perfect fit, both relations must hold simultaneously. Setting them equal (substitute the first into the second) gives:

√2·(r_B + r_X) = (r_A + r_X)

Real ions rarely satisfy this exactly, so Goldschmidt defined the ratio of the two sides as a tolerance factor:

t = (r_A + r_X) / [√2·(r_B + r_X)]

Here r_A, r_B, and r_X are the Shannon ionic radii (in ångström) for the correct coordination numbers — 12-fold for A, 6-fold for B, 6-fold for X. When t = 1 the geometry is perfect. When t < 1 the B–X framework is oversized relative to the A cavity, so the corner-sharing octahedra rotate (tilt) to shrink the cavity, lowering the symmetry from cubic to orthorhombic or rhombohedral.

Key Numbers and a Worked Example

Take strontium titanate, SrTiO₃, the textbook cubic perovskite. Using Shannon radii: r(Sr²⁺, 12-coord) = 1.44 Å, r(Ti⁴⁺, 6-coord) = 0.605 Å, r(O²⁻, 6-coord) = 1.40 Å.

• Numerator: r_A + r_X = 1.44 + 1.40 = 2.84 Å
• Denominator: √2·(r_B + r_X) = 1.4142 × (0.605 + 1.40) = 1.4142 × 2.005 = 2.835 Å
• t = 2.84 / 2.835 ≈ 1.00

A value essentially at 1.0 correctly predicts that SrTiO₃ is cubic at room temperature. Contrast with CaTiO₃: replacing Sr²⁺ (1.44 Å) with the smaller Ca²⁺ (≈1.34 Å) drops t to about 0.97, and CaTiO₃ is indeed orthorhombic with tilted octahedra — the very compound perovskite is named for. The empirical map is: 0.90 ≤ t ≤ 1.00 → cubic; 0.71 ≤ t < 0.90 → distorted (tilted) perovskite; t > 1.00 or t < 0.71 → non-perovskite (hexagonal, ilmenite, or decomposition). A useful companion metric is the octahedral factor μ = r_B/r_X, which must exceed ≈0.414 for the BX₆ octahedron itself to be stable.

How It Is Measured and Used in Practice

The tolerance factor is computed, not measured, but every input and prediction is checked against real data. In practice the workflow is:

• Pick radii from a consistent table. Shannon–Prewitt effective ionic radii (1976) are the standard source; using the correct coordination number matters — the 12-coordinate A radius is larger than its 6-coordinate value, and mixing them corrupts t.
• Compute t (and μ). A t in 0.8–1.0 with μ > 0.414 flags a likely perovskite worth synthesizing.
• Validate structure experimentally by X-ray or neutron powder diffraction. The Glazer octahedral-tilt notation (e.g., a⁻a⁻c⁺ for the common orthorhombic GdFeO₃ tilt) classifies the distortion that a sub-unity t predicts.

The factor is a first-pass filter in high-throughput screening: to design a lead-free halide photovoltaic absorber, a chemist sweeps A/B/X combinations, keeps only those with 0.8 ≤ t ≤ 1.0, then hands survivors to DFT total-energy calculations. It also guides tuning — mixing cations on the A site (e.g., formamidinium/cesium blends in FAPbI₃-based cells) nudges an effective average t back toward 1 to stabilize the desired black photoactive phase.

Tolerance Factor vs. Its Close Cousins

The Goldschmidt t is one of a small family of geometric descriptors, and knowing what it does not capture is essential.

• Octahedral factor (μ = r_B/r_X): t alone can be misleading because it says nothing about whether the small BX₆ octahedron is even stable. μ fixes that lower bound (≈0.414 from radius-ratio rules). Modern screens use the (t, μ) pair together.
• Radius ratio rules: the classic r₊/r₋ cutoffs (0.414 for octahedral, 0.732 for cubic coordination) are the ancestor of the octahedral factor; the tolerance factor extends that single-site logic to a two-cation framework.
• Bartel τ (2019): Christopher Bartel and coworkers used machine learning (the SISSO method) on 576 experimental ABX₃ compounds to build a new one-dimensional descriptor, τ = (r_X/r_B) − n_A·[n_A − (r_A/r_B)/ln(r_A/r_B)], where n_A is the oxidation state of the A cation. A compound is predicted to be a perovskite when τ < 4.18. It reaches 92% accuracy versus 74% for classic t on the same set.

The takeaway: Goldschmidt t is unmatched for intuition and speed, while τ and μ patch its two biggest blind spots — cation oxidation state and octahedral instability.

Exceptions, Limits, and Why It Still Matters

For all its utility, t is an approximation, and its failures are instructive:

• Radii are model-dependent. Ionic radii are not physical constants; different tabulations shift t by ±0.02–0.05, enough to flip a borderline prediction. Covalent bonding (common in Pb- and Sn-halide perovskites) further breaks the hard-sphere assumption.
• False positives near the edges. Some compounds with 0.8 ≤ t ≤ 1.0 still do not form perovskites (e.g., certain t-in-range oxides adopt hexagonal polytypes), and some with t just outside the window do — this is exactly why Goldschmidt t scores only ~74% and Bartel τ was developed.
• Molecular A cations. For hybrid perovskites like methylammonium lead iodide (MAPbI₃), the non-spherical CH₃NH₃⁺ cation needs an effective radius, adding uncertainty.

Yet the factor endures because it is right far more often than a coin flip, requires no computation beyond arithmetic, and gives a physical picture — tilting, cavity size, close packing — that a black-box model does not. Nearly a century after 1926, it remains the first thing a solid-state chemist writes down when a new ABX₃ composition is proposed, from oxide fuel-cell cathodes to the halide perovskites driving 25%-efficient solar cells.

Tolerance factor ranges and the structures they predict for ABX3 perovskites, with representative examples

t range	Predicted structure	Symmetry / behavior	Example
t > 1.0	Hexagonal or non-perovskite	A cation too large; layered stacking	BaNiO3, CsCuCl3
0.90 – 1.00	Cubic perovskite	Ideal, undistorted BX6 octahedra	SrTiO3 (t ≈ 0.99)
0.80 – 0.90	Distorted perovskite	Octahedral tilting, orthorhombic/rhombohedral	CaTiO3 (t ≈ 0.97 by some radii; GdFeO3 tilt)
0.71 – 0.80	Heavily tilted perovskite	Strong tilt, low symmetry, often orthorhombic	GdFeO3, CaSnO3
t < 0.71	Non-perovskite (ilmenite/corundum)	B cation too large; different packing	FeTiO3 (ilmenite)