Ionic Radii and Their Determination

Take a crystal of NaCl and measure, by X-ray diffraction, the distance between a sodium nucleus and its nearest chloride nucleus: 2.81 Å. That single number is unambiguous. But splitting it into "how big is Na⁺" and "how big is Cl⁻" is not. This is because X-rays see only the total electron density along the bond. They do not see a dividing line between two ions. Ionic radii are therefore not directly measured quantities. They are constructed quantities, built by choosing a partition rule. That rule is then propagated consistently across hundreds of compounds. That construction is one of the most successful pieces of applied crystal chemistry in existence. It underlies everything from predicting crystal structures to explaining why Mg²⁺ substitutes for Fe²⁺ in olivine but not for Ca²⁺ in feldspar. The problem matters because The Ionic Model treats ions as charged, incompressible spheres. These spheres are held together by Coulomb attraction and short-range repulsion. That model only becomes predictive once each ion is assigned a definite size. This is because the sum of two ionic radii predicts bond lengths in compounds never yet synthesized. The ratio of cation to anion radius is also used. It predicts which structure type (rock salt, zinc blende, fluorite, and so on) a given ionic solid will adopt. Get the radii wrong and the whole downstream machinery gives wrong answers. That machinery includes Radius Ratio Rules, lattice energy calculations, and even qualitative arguments about solubility and hydration. The big picture is this: an ion's "radius" is an operationally defined, transferable parameter. It is extracted from many crystal structures simultaneously. This extraction uses physically motivated but ultimately conventional assumptions. Those assumptions concern how electron density divides between a cation and an anion in contact. Different conventions (Goldschmidt, Pauling, and the internally most self-consistent modern set of Shannon and Prewitt) give different absolute numbers. These sometimes differ by tenths of an ångström for the same ion. Yet all of them reproduce experimental bond lengths. All of them also correctly predict which structures are stable. This is because what matters for structural prediction is not the absolute radius. What matters is the ratio of radii. That ratio is far more convention-independent than either radius alone. This article builds the ionic radius concept from the ground up. It covers how contact distances are measured. It explains why a single crystal structure cannot resolve two radii. It traces the historical progression of partition schemes. It covers the periodic trends radii display, and why those trends have the physical shapes they do. It also gives practical rules for using tabulated radii correctly. These include the coordination-number and spin-state corrections. Such corrections trip up most students the first time they open a radius table.

Key takeaways

Ionic radii are not measured directly; only the cation–anion contact distance $d = r_+ + r_-$ is measured, by X-ray or neutron diffraction.
Splitting $d$ into $r_+$ and $r_-$ requires an extra assumption — historically Landé's touching-anions method, then Goldschmidt/Pauling scaling, then Shannon's self-consistent global fit.
Radii increase down a group (more shells) and generally decrease left-to-right across a period at fixed charge (rising $Z_{\text{eff}}$ ).
Isoelectronic series (same electron count, e.g. O²⁻, F⁻, Na⁺, Mg²⁺, Al³⁺) show radius falling smoothly as nuclear charge rises with the same electron cloud.
Radii are not fixed numbers per ion: they depend on coordination number (CN) and spin state, and increase systematically with CN.
Key idea: an ionic radius is a self-consistent, transferable bookkeeping device fitted to reproduce thousands of measured bond lengths, not a directly observable boundary of an ion.

Where this lives today

Modern crystal chemistry, mineralogy, and materials design all lean on Shannon radii as a working currency. In geochemistry, the radius-and-charge matching between trace cations and major lattice sites (Goldschmidt's substitution rules, a direct descendant of this article's material) predicts which trace elements partition into olivine versus feldspar during magma crystallization, underpinning isotope geochronology and ore-deposit models. In battery materials science, the radii of Li⁺, Na⁺, and Mg²⁺ (and how those radii change with coordination in a spinel or layered oxide host) are used to screen candidate cathode and solid-electrolyte materials computationally before any synthesis is attempted, because ionic radius mismatch strongly predicts whether a substituting ion will fit into a given crystallographic site without destabilizing the structure. In pharmaceutical and biological chemistry, ionic radii of Na⁺, K⁺, Ca²⁺, and Mg²⁺ explain selectivity in ion channels and metalloenzyme active sites, where a protein pocket sized for one cation radius can discriminate against another differing by only a few hundredths of an ångström. In solid-state device engineering, radius-ratio arguments extended from this framework (leading directly into Radius Ratio Rules and Close-Packing of Spheres) guide the design of perovskite photovoltaic materials, where the Goldschmidt tolerance factor — built entirely from ionic radii — predicts which combinations of A-site, B-site, and halide/oxide ions will form the stable perovskite structure needed for high-efficiency solar cells.

Ask the tutor: How does the radius-ratio rule actually use these tabulated radii to predict whether a compound adopts rock-salt, zinc-blende, or fluorite structure?

The physical picture

When two ions of opposite charge are packed together in a crystal, their electron clouds overlap slightly and then a repulsive wall — dominated by Pauli exclusion between filled shells — stops them from approaching further. Diffraction experiments locate the nuclei, so what comes out of a crystal structure refinement is the internuclear distance between a cation and its nearest-neighbor anion. This distance is real, precise (often to ±0.001 Å for good single-crystal data), and completely unambiguous. The trouble starts when you try to say where Na⁺ "ends" and Cl⁻ "begins" along that line, because the electron density does not drop to zero at some clean boundary — it has a broad minimum somewhere between the two nuclei, and different theoretical conventions place the dividing point at different locations within that minimum. The earliest resolution, due to Landé (1920), used a physical trick rather than a theoretical partition: find a crystal in which the anions touch each other while the cation, sitting in the octahedral or tetrahedral hole, is too small to keep them apart. In such a structure the anion–anion distance measured across the shared face of the hole gives $2r_-$ directly, with no ambiguity at all, because it is anion touching anion, not cation touching anion. Landé identified compounds close to this limiting geometry (small cations like Li⁺ in a large anion lattice) and used the anion radius extracted this way as an anchor. Once $r_-$ is fixed for one anion, every cation radius follows by subtracting $r_-$ from measured cation–anion distances throughout the periodic table, and every other anion radius follows the same way once a few cations are cross-calibrated. Goldschmidt (1926) and later Pauling (1927) refined this idea using a physical model for how much of the internuclear distance "belongs" to each ion. Pauling's argument was electrostatic: for ions with the same electron configuration (isoelectronic ions like Na⁺ and F⁻, both with the neon configuration), the radius should scale inversely with the effective nuclear charge felt by the outermost electrons, because a larger effective charge pulls the same electron cloud in more tightly. This gives a theoretically motivated way to split any measured $d = r_+ + r_-$ between the two ions, self-consistently across the whole periodic table. Pauling's radii became the standard classroom set for decades. The modern standard, due to Shannon and Prewitt (1969, revised by Shannon in 1976), abandons any single physical model and instead performs a global least-squares fit: take thousands of accurately measured bond lengths across oxides and fluorides of known structure, and simultaneously solve for a self-consistent set of radii (one per ion, per coordination number, per spin state where relevant) that best reproduces all the observed distances at once. Shannon radii are anchored so that $r(\text{O}^{2-}, \text{CN 6}) = 1.40$ Å, close to the traditional Pauling value, which is why the two scales resemble each other numerically even though the underlying logic (physical model vs. global fit) is completely different. Shannon radii are what appear in modern data tables and are the ones used throughout crystal chemistry today.

Figure

Splitting a measured bond length: the ambiguity and its resolution

Figure

Ionic radius trends: isoelectronic series and group/period behavior

The core ideas

The measured quantity is always a sum:

d_{+-} = r_{+} + r_{-}

where: $d_{+-}$ is the crystallographically measured cation–anion internuclear distance; $r_+$ , $r_-$ are the (as-yet unassigned) cation and anion radii.

Reading: one diffraction experiment gives one number, $d_{+-}$ ; it constrains a line in $(r_+, r_-)$ space but cannot, by itself, fix a point on that line. Resolving the ambiguity requires an external assumption — anion–anion contact, an electrostatic model, or a global multi-compound fit. Pauling's electrostatic partition, for isoelectronic ions, uses:

r \propto \frac{1}{Z_{\text{eff}}}

where: $r$ is the ionic radius; $Z_{\text{eff}} = Z - \sigma$ is the effective nuclear charge felt by the outermost electron shell, with $Z$ the atomic number and $\sigma$ a screening constant (approximated, e.g., by Slater's rules) that is the same for all members of an isoelectronic series since they share the same electron configuration.

Reading: within one isoelectronic series, all the messy quantum-mechanical detail cancels because $\sigma$ is common to every member; the radius ordering is then governed purely by $Z$ , giving a clean, testable prediction: radius must fall monotonically as $Z$ rises across the series. The dependence of radius on coordination number is captured empirically (not from first principles) as a smooth increase:

r(\text{CN}) = r(\text{CN}=6) + \Delta(\text{CN})

where: $r(\text{CN})$ is the radius at coordination number CN; $\Delta(\text{CN})$ is a tabulated correction, positive for CN > 6 (ion appears larger when surrounded by more neighbors) and negative for CN < 6 (ion appears smaller in tighter coordination, e.g. tetrahedral CN 4).

Reading: radius is not a single number per ion; it is a function of local environment, because more neighbors around an ion spreads the same charge cloud over a larger contact perimeter, effectively inflating the apparent radius. Typical shifts are of order 0.1–0.2 Å between CN 4 and CN 8 for a given ion — comparable to differences between whole rows of the periodic table, so ignoring CN can produce serious errors.

Derivation AWhy Pauling's isoelectronic scaling gives a definite splitting of d

Goal: starting only from the measured sum $d = r_+ + r_-$ for one compound, show how an isoelectronic assumption converts it into two separate numbers, and see why this only works two ions at a time unless anchored externally.

Step 1 — Set up the unknowns. Consider NaF, where Na⁺ and F⁻ are both isoelectronic with neon (10 electrons: $1s^2 2s^2 2p^6$ ). The measured Na–F distance is $d = 2.31$ Å. We want $r_{\text{Na}^+}$ and $r_{\text{F}^-}$ separately, but we have one equation, $r_{\text{Na}^+} + r_{\text{F}^-} = 2.31$ , and two unknowns.

Step 2 — Introduce the electrostatic scaling law. Since both ions have the same electron configuration, Pauling assumed each ion's outer electron density is governed by a common formula, differing only in $Z_{\text{eff}}$ :

r_{\text{Na}^+} = \frac{C}{Z_{\text{eff}}(\text{Na}^+)}, \qquad r_{\text{F}^-} = \frac{C}{Z_{\text{eff}}(\text{F}^-)}

where $C$ is a constant (same for both, since both are Ne-like) and $Z_{\text{eff}} = Z - \sigma$ with $\sigma$ the same screening constant for both ions (11 electrons removed to 10, or 9 electrons added to 10 — same shell, same shielding pattern).

Step 3 — Compute the effective nuclear charges. For the neon configuration, Slater's rules give a screening constant $\sigma \approx 4.15$ from the other n = 2 electrons and the n = 1 core (this is the standard Pauling value used historically; the exact number is less important than the ratio it produces). Then:

Z_{\text{eff}}(\text{Na}^+) = 11 - 4.15 = 6.85, \qquad Z_{\text{eff}}(\text{F}^-) = 9 - 4.15 = 4.85

Step 4 — Use the ratio to split the measured distance. Since $r_{\text{Na}^+}/r_{\text{F}^-} = Z_{\text{eff}}(\text{F}^-)/Z_{\text{eff}}(\text{Na}^+)$ , we get:

\frac{r_{\text{Na}^+}}{r_{\text{F}^-}} = \frac{4.85}{6.85} = 0.708

Combining with $r_{\text{Na}^+} + r_{\text{F}^-} = 2.31$ Å:

r_{\text{F}^-}(1 + 0.708) = 2.31 \ \text{Å} \implies r_{\text{F}^-} = \frac{2.31}{1.708} = 1.353 \ \text{Å}

r_{\text{Na}^+} = 2.31 - 1.353 = 0.957 \ \text{Å}

Step 5 — Recognize the limitation. This split depends entirely on the assumed value of $\sigma$ ; a different screening model gives a different ratio and hence different individual radii, even though the sum $d = 2.31$ Å is fixed by experiment and not in dispute. This is why Pauling radii and Shannon radii disagree ion-by-ion (Pauling $r_{\text{Na}^+} \approx 0.95$ Å, Shannon $r_{\text{Na}^+}(\text{CN6}) = 1.02$ Å) while both correctly reproduce measured bond lengths when the anion radius from the same scale is added back on. Radii from different scales must never be mixed.

Go deeper · Derivation B · 2 minWhy radius increases down a group and (mostly) decreases across a period

Derivation BWhy radius increases down a group and (mostly) decreases across a period

Goal: account for the two dominant periodic trends in ionic radius using effective nuclear charge and principal quantum number, and identify the ionic-model regime in which each trend applies cleanly.

Step 1 — Down a group, at fixed charge. Compare the Group 1 cations Na⁺ (CN6: 1.02 Å) and K⁺ (CN6: 1.38 Å). Both have a noble-gas-like closed shell, but K⁺'s outermost electrons occupy $n = 3$ orbitals while Na⁺'s occupy $n = 2$ . The radial extent of a hydrogen-like orbital scales roughly as:

\langle r \rangle \sim \frac{n^2 a_0}{Z_{\text{eff}}}

where $a_0$ is the Bohr radius. Even though $Z_{\text{eff}}$ increases somewhat down the group (more protons), the $n^2$ growth dominates overwhelmingly — going from $n=3$ (Na⁺ core... careful: Na⁺ outer shell is $n=2$ , K⁺ outer shell is $n=3$ ) the ratio $n^2$ jumps from 4 to 9, more than compensating any change in effective charge. Hence radius increases down a group whenever the ions being compared are isovalent (same charge).

Step 2 — Across a period, for isoelectronic species. Compare O²⁻ (1.40 Å), F⁻ (1.33 Å), Na⁺ (1.02 Å), Mg²⁺ (0.72 Å), Al³⁺ (0.535 Å) — all ten-electron, neon-configuration ions, $Z$ running from 8 to 13. Here $n$ is fixed (all outer electrons in $n=2$ ), so the radius formula reduces to:

r \propto \frac{1}{Z_{\text{eff}}} = \frac{1}{Z - \sigma}

with $\sigma$ essentially constant across the series (same electron configuration, same shielding pattern). As $Z$ rises from 8 (oxygen) to 13 (aluminum), $Z_{\text{eff}}$ rises monotonically and $r$ falls monotonically — exactly the trend plotted in the second figure.

Step 3 — Why this is not simply "radius decreases across a period" for neutral atoms extended naively to ions of mixed charge. The clean monotonic decrease only holds when comparing ions with the same charge and same electron count (isoelectronic, isovalent). Across a real period of neutral atoms turning into their typical ions, charge changes too (metals lose electrons, forming small cations; nonmetals gain electrons, forming large anions), so the overall left-to-right trend in a table of "the ions each element normally forms" shows a large discontinuous jump upward in radius at the metal/nonmetal boundary — small cations on the left, large anions on the right — superimposed on the smooth within-a-charge-type decrease.

Step 4 — Consistency check with the general shell-filling argument. For a fixed number of electrons, adding a proton to the nucleus (increasing $Z$ ) pulls every occupied orbital inward without changing the number of nodes or the principal quantum numbers involved; this must shrink the ion, with no exception, which is exactly the isoelectronic-series behavior derived in Step 2. There is no competing effect in this comparison because $n$ does not change — this is the cleanest, least-ambiguous periodic trend in the entire ionic radius data set.

Worked examples

Example 1 (foundational): resolving Na⁺ and Cl⁻ radii using Landé's anion-contact method. Suppose LiCl has measured Li–Cl distance $d = 2.57$ Å, and independent analysis of the LiCl rock-salt structure shows the chloride ions nearly touch each other along the face diagonal of the shared octahedral edge, with a Cl···Cl contact distance of $3.63$ Å (this is a real near-limiting case because Li⁺ is unusually small for the octahedral hole). In a rock-salt structure, the Cl⁻ ions form a face-centered cubic array with nearest-neighbor anion–anion distance equal to $a/\sqrt{2}$ , where $a$ is the cubic cell edge; if these anions are touching, that distance equals $2r_{\text{Cl}^-}$ directly:

2r_{\text{Cl}^-} = 3.63 \ \text{Å} \implies r_{\text{Cl}^-} = 1.815 \ \text{Å}

Then the Li⁺ radius follows immediately from the measured Li–Cl distance:

r_{\text{Li}^+} = d - r_{\text{Cl}^-} = 2.57 - 1.815 = 0.755 \ \text{Å}

This chloride radius, once fixed, can now be subtracted from any other measured M–Cl distance to obtain the radius of cation M, propagating the scale across the whole periodic table. (Modern Shannon values, $r_{\text{Cl}^-}$ (CN6) = 1.81 Å and $r_{\text{Li}^+}$ (CN6) = 0.76 Å, agree with this construction to within 0.005 Å — reassuring, since Shannon's fit used a completely different method, a global least-squares fit rather than a single limiting structure.)

Example 2 (applied): predicting a bond length and checking against a radius-ratio structural prediction. Predict the Mg–O distance in MgO (rock-salt structure, both ions CN 6) using Shannon radii $r_{\text{Mg}^{2+}}(\text{CN6}) = 0.72$ Å and $r_{\text{O}^{2-}}(\text{CN6}) = 1.40$ Å:

d_{\text{Mg–O}} = r_{\text{Mg}^{2+}} + r_{\text{O}^{2-}} = 0.72 + 1.40 = 2.12 \ \text{Å}

The experimental Mg–O distance in periclase (MgO) is $2.106$ Å — agreement to about 0.6%, well within the expected accuracy of additive radii for a simple binary oxide. Now check the coordination number this radius pair predicts, foreshadowing Radius Ratio Rules: the ratio

\rho = \frac{r_{\text{Mg}^{2+}}}{r_{\text{O}^{2-}}} = \frac{0.72}{1.40} = 0.514

falls in the range 0.414–0.732, which the radius-ratio rule assigns to octahedral (CN 6) coordination — exactly the coordination MgO actually adopts. This is the payoff of the whole radius-tabulation exercise: radii built purely from bond-length data, without reference to any particular structure type, nonetheless correctly predict which structure type a compound will choose.

Example 3 (exam/competition level): coordination-number correction and a trap with mismatched scales. A student is given that CaO adopts the rock-salt structure (CN 6 for both ions) with a measured Ca–O distance of 2.405 Å, and is asked to predict the Ca–O distance in a hypothetical CN 8 polymorph (CsCl-type), given the following Shannon data: $r_{\text{Ca}^{2+}}(\text{CN6}) = 1.00$ Å, $r_{\text{Ca}^{2+}}(\text{CN8}) = 1.12$ Å, $r_{\text{O}^{2-}}(\text{CN6}) = 1.40$ Å, $r_{\text{O}^{2-}}(\text{CN4}) = 1.38$ Å (oxide in a CsCl-type AO structure sits in a cubic, CN 8, hole too, but suppose for this problem we are told the oxide's effective coordination number stays formally paired at CN 6 due to local distortion — the trap is to blindly add CN8 + CN8). First, verify the CN6 baseline: $r_{\text{Ca}^{2+}}(\text{CN6}) + r_{\text{O}^{2-}}(\text{CN6}) = 1.00 + 1.40 = 2.40$ Å, matching the measured 2.405 Å essentially exactly — good, the scale is internally consistent for this pair. For the CN 8 polymorph, the naive (and wrong, if the oxide truly stayed CN 6-like) approach is to add two CN8 radii. Suppose the correct physical situation is CN 8 for both ions (a proper CsCl-type AO structure, cation and anion both eight-coordinate) — the problem statement above about "oxide stays CN6" was a deliberately inserted red herring; a real CsCl-type MO structure has both ions in cubic CN 8 holes symmetric to each other. Using the correct CN8 oxide radius (Shannon: $r_{\text{O}^{2-}}(\text{CN8}) = 1.42$ Å, not the CN4 value of 1.38 Å mistakenly quoted above):

d_{\text{CN8}} = r_{\text{Ca}^{2+}}(\text{CN8}) + r_{\text{O}^{2-}}(\text{CN8}) = 1.12 + 1.42 = 2.54 \ \text{Å}

The lesson embedded in the trap: coordination numbers must be matched to the correct table entry for both ions, and radii from different coordination environments must never be combined (using the CN4 oxide value of 1.38 Å instead of the correct CN8 value of 1.42 Å would have given a spuriously short 2.50 Å, a 1.5% error easily missed on an exam). Always confirm the coordination number of each ion in the actual structure before pulling numbers from a table.

Limiting cases, scope & common traps

Situation	What happens to the radius concept	Correct handling
Isoelectronic series (fixed electron count, rising $Z$ )	Radius decreases smoothly and monotonically	Compare only within the series; do not mix charges from different series
Same ion, increasing coordination number	Radius increases (larger apparent size)	Always match CN between the radius value and the real structure
High-spin vs. low-spin transition-metal ion (e.g., Fe²⁺)	Low-spin radius noticeably smaller (fewer electrons in $e_g^{}$ -type antibonding-like orbitals pointing at ligands)	Use spin-state-specific Shannon entries; never use a generic "Fe²⁺ radius"
Very small, highly charged cations (e.g., Si⁴⁺, high field strength)	Ionic model itself weakens; bonding gains covalent/polarization character	Radii still tabulated and useful, but treat bond lengths as approximate; consult The Ionic Model limits
Comparing radii from different scales (Pauling vs. Shannon vs. Goldschmidt)	Absolute values differ by up to ~0.1–0.2 Å per ion	Never mix scales in one calculation; always add two radii from the same table
Using radii to predict structure type	Works well because the ratio $r_+/r_-$ is far more scale-independent than either radius alone	Proceed to Radius Ratio Rules using ratios, not absolute values

The chemist behind it

Linus Pauling (1901–1994) is the figure most associated with codifying ionic radii into the form generations of chemists learned them, though the underlying groundwork — recognizing that anion-contact structures could anchor an absolute radius scale — was laid by Wasastjerna and especially Victor Goldschmidt in the 1920s, working from early X-ray crystallographic data on alkali halides. Pauling, already deep into building the electrostatic and quantum-mechanical foundations that would become The Nature of the Chemical Bond (1939), applied Slater-type screening arguments to isoelectronic ion series to derive a fully self-consistent radius scale from first principles rather than from empirical anchoring alone. His scale, though later shown to disagree in detail with more accurate measurements, became the pedagogical standard because it connected a tabulated number directly to an orbital-based physical argument that students could follow and verify — exactly the kind of transparent reasoning this article has walked through.

Origins · 1 minLinus Carl Pauling and how this was discovered

Linus Carl Pauling

1901–1994 · American chemist, X-ray crystallographer, and structural theorist

Pauling combined quantum mechanical arguments (electronegativity, hybridization, effective nuclear charge) with meticulous crystallographic measurement to build quantitative, testable rules for chemical bonding and structure — of which the ionic radius scale was one of the earliest and most widely used. He received the Nobel Prize in Chemistry (1954) for work on the nature of the chemical bond, and the Nobel Peace Prize (1962) for anti-nuclear-testing activism, making him the only person to receive two unshared Nobel Prizes.

History

- 1920 — Landé proposes using anion–anion contact in small-cation halide structures to fix an absolute anion radius.

- 1926 — Goldschmidt compiles an extensive empirical table of ionic radii from X-ray data on hundreds of compounds.

- 1927 — Pauling derives a self-consistent radius scale from effective nuclear charge arguments applied to isoelectronic ion series.

- 1969 — Shannon and Prewitt perform the first large-scale, self-consistent least-squares fit of radii across oxides and fluorides.

- 1976 — Shannon publishes the revised, coordination-number- and spin-state-resolved radius tables still used as the modern standard.

Why don't we just calculate ionic radii from quantum mechanics instead of fitting them to crystal data?

Quantum mechanical calculations do give an electron density around an isolated ion or an ion embedded in a crystal, but "radius" requires choosing where within that continuously varying density to draw a boundary — there's no natural cutoff. Different reasonable choices (e.g., the radius at which density falls to some threshold, or the location of a minimum in density along the bond) give different numbers, and none of them automatically reproduce the additive relationship $d = r_+ + r_-$ across hundreds of compounds simultaneously as well as an empirical fit does. Empirical radius scales like Shannon's remain more practically useful precisely because they are calibrated against the actual measured quantity chemists care about — bond lengths in real crystals — rather than against a theoretically defined but experimentally inaccessible boundary.

Do ionic radii include the effects of covalency, or are they purely for "ideal" ionic bonds?

Ionic radius tables are built mostly from oxide and fluoride structures, chosen because O²⁻ and F⁻ form some of the most nearly ionic bonds available, minimizing covalent contamination of the fit. However, no real bond is perfectly ionic, so some covalent character is baked into every radius, especially for smaller, higher-charge, or more polarizing cations such as Si⁴⁺ or Ti⁴⁺. For strongly covalent compounds (sulfides, many bonds to carbon or nitrogen), additive ionic radii systematically fail to reproduce bond lengths, signaling that the ionic model itself — not just the radius values — is breaking down for that bond type.

Why do transition-metal ions have different radii in the high-spin and low-spin states?

In an octahedral crystal field, the $d$ orbitals split into a lower-energy $t_{2g}$ set and a higher-energy $e_g$ set that points directly at the ligands. High-spin configurations place more electrons in the $e_g$ orbitals, which increases electron density along the metal–ligand bond direction and pushes ligands slightly further away, inflating the apparent ionic radius. Low-spin configurations pack electrons preferentially into $t_{2g}$ , which points between ligands rather than at them, allowing ligands to approach more closely and giving a measurably smaller radius for the identical ion and identical charge. Fe²⁺, for instance, is noticeably larger in its high-spin state than in its low-spin state at the same coordination number.

How much do Goldschmidt, Pauling, and Shannon radii actually differ numerically for a typical ion?

For Na⁺, Goldschmidt's original value was about 0.98 Å, Pauling's electrostatic-scaling value was about 0.95 Å, and Shannon's globally fitted CN6 value is 1.02 Å — differences of a few hundredths of an ångström, small compared to the ion's size but not negligible for precision bond-length prediction. For some ions the discrepancies are larger, particularly where the original Goldschmidt/Pauling anchoring compounds were not ideal (highly ionic, well-characterized structures). The practical rule is always to take both radii in an additive bond-length calculation from the same table, since the scales are each internally consistent but not cross-compatible.

Why does the radius-ratio rule work at all if ionic radii themselves are somewhat arbitrary conventions?

The radius-ratio rule depends only on $r_+/r_-$ , and this ratio is much less sensitive to the choice of partition convention than either radius individually, because whatever assumption shifts $r_+$ up tends to shift $r_-$ down by a correlated amount (since both are extracted from the same fixed sum $d = r_+ + r_-$ across many compounds). Structural predictions based on radius ratios turn out to be robust across the Goldschmidt, Pauling, and Shannon scales even though the individual radii differ, which is strong indirect evidence that the radius concept, while conventional, captures something physically real about relative ionic sizes.

Can an anion be smaller than a cation, contradicting the usual picture of "big anions, small cations"?

Yes — this is common for anions of small, highly electronegative atoms combined with cations of large, low-charge metals. For example, F⁻ (1.33 Å) is smaller than K⁺ (1.38 Å, CN6) or Rb⁺ (1.52 Å), even though the general classroom heuristic is "anions are bigger than cations." The heuristic is a reasonable first approximation across a typical period (compare Na⁺ to Cl⁻, where the anion is indeed much larger) but fails once you compare across very different rows of the periodic table, which is exactly why quantitative radius tables, not qualitative heuristics, are needed for real structural predictions.

What role do ionic radii play in predicting whether two ions will substitute for each other in a mineral or a doped material?

Goldschmidt's substitution rules state that an ion can readily replace another in a crystal lattice if their radii differ by less than about 15% and their charges are compatible (equal charge, or a coupled substitution that maintains overall charge balance). This radius-matching criterion, built entirely from tabulated ionic radii, explains why Fe²⁺ and Mg²⁺ substitute freely in olivine (radii close, both +2) but Ca²⁺ does not substitute easily into the same site (radius too different), and it is the starting point for designing doped materials in solid-state chemistry, from doped semiconductors to substituted battery cathodes.