How To Find A Coordination Number

You’re staring at a crystal structure diagram — maybe a perovskite, maybe a simple metal lattice — and the question hits: How many neighbors does that central atom actually have?Worth adding: in practice, it’s the detail that decides whether your catalyst works, your battery degrades, or your protein folds correctly. * It sounds like a textbook definition. The coordination number (CN) isn't just a number. It’s the local geometry fingerprint of an atom. And finding it? That depends entirely on what* you’re looking at and how you’re looking.

What Is a Coordination Number

At its core, the coordination number counts the nearest neighbors surrounding a central atom or ion. In a perfect crystal, it’s an integer: 4 for tetrahedral, 6 for octahedral, 8 for cubic, 12 for close-packed metals. But the moment you step away from idealized textbook structures — defects, surfaces, amorphous materials, solutions, proteins — the definition gets slippery.

There are two main flavors. You pick a central atom, draw a cutoff sphere, and count what falls inside. The chemical coordination number is messier. Some don't. It cares about bonding. On the flip side, a hydrogen bond? Some count it. Even so, the literature is inconsistent. A long, weak interaction might count crystallographically but not chemically. The crystallographic coordination number comes from a static structure file (CIF, PDB, POSCAR). You have to decide which definition serves your question.

The cutoff problem

This is where everyone gets stuck. There is no universal cutoff radius. Ionic radii tables (Shannon, Pauling) give you a starting point. Sum the radii of the central ion and the ligand, add a tolerance — usually 0.Here's the thing — 2–0. 5 Å — and that’s your sphere. But covalent bonds? Also, metallic bonds? Think about it: van der Waals contacts? Each demands a different logic. That said, in a metallic glass, the first minimum of the radial distribution function (RDF) is the only defensible cutoff. In a MOF, you might use the pore-limiting diameter. Also, the cutoff is the analysis. Choose it consciously, not by default.

Why It Matters

Coordination number dictates properties. Full stop.

• Catalysis: A Pt nanoparticle’s edge sites (CN=7) bind CO differently than terrace sites (CN=9). That difference changes selectivity.
• Ion conduction: In solid electrolytes, Li+ hopping barriers correlate with the CN of the migration pathway. Lower CN often means lower barrier — but not always.
• Protein folding: The coordination number of a residue (Cα or side-chain heavy atoms) predicts burial, stability, and folding rates. It’s a coarse-grained order parameter that works.
• Phase transitions: In perovskites, the B-site CN drops from 6 to 5 at the tetragonal-to-cubic transition. That’s the structural signature of the ferroelectric switch.

If you get the CN wrong, you misassign the oxidation state, you misinterpret the XAS pre-edge, you build the wrong defect model. It propagates.

How to Find It — By Method

The method depends on your input. Here’s the practical breakdown.

From a CIF or POSCAR (single crystal, periodic)

You have a structure file. You want the CN of every symmetry-unique atom.

1. Load it into a real tool. pymatgen, ASE, CrystalToolkit, VESTA, Mercury. Don’t write your own neighbor finder unless you enjoy debugging periodic boundary conditions at 2 AM.
2. Pick a cutoff strategy.
   • Fixed radius*: Fast, dumb. Use 3.0 Å for oxides, 2.8 Å for metals, adjust.
   • Sum of ionic radii + tolerance*: Better. pymatgen’s CrystalNN or BrunnerNN_real implement this with published radii sets.
   • Adaptive (Voronoi/weighted)*: CrystalNN uses a weighted Voronoi approach. It handles distorted polyhedra better than a sphere. It’s the current gold standard for inorganic crystals.
3. Run the neighbor search. Get the list of indices, distances, and angles.
4. Inspect the histogram of distances. Plot the first coordination shell. Is there a clear gap? If distances run 2.05, 2.07, 2.10, 2.45, 2.48… the gap at 2.10→2.45 is your natural cutoff. The sphere method might catch the 2.45 ones. The Voronoi method might not. Neither is "wrong" — but they answer different questions.
5. Report the method. "CN=6 determined by CrystalNN (weighted Voronoi) on the ICSD structure #12345." That’s reproducible. "CN=6" alone is not.

From a trajectory (MD, AIMD, Monte Carlo)

Now you have time. And thermal noise. The CN fluctuates.

1. Compute the RDF (g(r)) for the pair of interest. The first peak position = typical bond distance. The first minimum = natural cutoff.
2. Define a time-resolved CN. For each frame, count neighbors within that cutoff. You get a distribution: CN=4 (5%), CN=5 (60%), CN=6 (35%).
3. Don’t just average. The average CN=5.3 is meaningless. The distribution* tells you the mechanism. Is it a two-state jump? A continuous distortion?
4. Watch for artifacts. In AIMD with a small cell, periodic images can sneak into the first shell. Check the minimum image convention. If your cutoff exceeds half the box length, you’re double-counting.

From a PDF (Pair Distribution Function) — experimental

You have neutron or X-ray total scattering. Even so, no long-range order. The CN is hidden in the peak area.

Continue exploring with our guides on parallel lines bisected by a transversal and what does the word velocity mean.

1. Model the first peak. Fit a Gaussian (or Voigt, or asymmetric peak shape) to the first coordination shell in G(r).
2. Integrate the peak area. The coordination number is: $N = 4\pi\rho_0 \int_{r_1}^{r_2} r^2 g(r) dr$ where $\rho_0$ is the average number density. You need* the density. If you don’t know the sample density, you cannot get an absolute CN from PDF alone. Relative changes? Yes. Absolute? No.
3. Correlated motion kills you. In liquids and glasses, the peak width contains both thermal and static disorder. The area still gives CN — but only if the peak is fully separated from the next shell. Overlapping peaks? You need a structural model (Reverse Monte Carlo, EPSR, or a crystalline approximate) to deconvolute.

From EXAFS (Extended X-ray Absorption Fine Structure)

This is element-specific. You tune to the Fe K-edge, you see Fe neighbors.

1. Fit the first shell in R-space (Fourier filtered). The amplitude gives $N \cdot S_0^2$. $S_0^2$ is the amplitude reduction factor (usually 0.8–0.95, calibrated on a known standard).
2. Degeneracy warning. EXAFS fits $N \times S_0^2$. If you don’t know $S_0^2$, you don’t know $N$. You must* measure a reference compound with known CN (e.g., Fe foil for CN=12, FeO for CN=6) under identical conditions

Reference and calibration
Continuing the EXAFS discussion, the sentence above should read: “…under identical conditions, and then extract the amplitude reduction factor $S_0^2$ from the known standard.” Once $S_0^2$ is known, you can reliably convert the fitted amplitude into an absolute coordination number for the unknown sample.

Cross‑validation with complementary techniques
No single method can capture the full complexity of a coordination environment, especially when disorder, dynamics, or overlapping shells are present. A pragmatic workflow often combines several approaches:

• Crystal structures (sphere/Voronoi) give a static, idealized picture. Use them as a baseline and compare to experimental or simulation data.
• Molecular dynamics or Monte‑Carlo trajectories provide time‑resolved distributions. The RDF‑derived cutoff and the neighbor‑counting statistics can be fed back into the static analysis to test whether the “average” structure is representative.
• PDF analysis supplies the total‑pair distribution function, integrating over all atoms in the sample. It is excellent for checking the overall mass balance but must be anchored to a known density.
• EXAFS offers element specificity and sensitivity to short‑range order, but it is limited to the absorbing atom’s immediate environment.

When these methods converge on a consistent picture, confidence in the derived coordination number is greatly increased. Discrepancies, on the other hand, are valuable: they point to phenomena such as mixed‑site occupancy, dynamic coordination changes, or systematic artifacts (e.g., periodic‑image contamination in simulations).

Practical checklist for a reproducible CN determination

Final thoughts
Coordination numbers are not immutable constants; they are emergent properties that depend on the structural model, the experimental probe, and the temporal or thermal window of observation. By rigorously reporting the methodology, explicitly stating assumptions (e.g., density for PDF, $S_0^2$ for EXAFS), and leveraging complementary techniques, researchers can move beyond “CN=6” as a vague shorthand to a well‑defined, reproducible descriptor of local order.

In practice, the most solid conclusions arise when the static picture from crystallography, the dynamic distribution from simulations, and the element‑specific signature from spectroscopy all point to the same coordination environment—or when their differences themselves reveal the underlying physics of the material. This holistic, transparent approach is the cornerstone of modern structural science.