Structure Field Map¶

An empirical materials map that places compositionally comparable crystalline compounds in descriptor coordinates and uses the resulting structure-type fields as an interpretable classification and candidate-structure prior.

Version: v2 · 2026-09-06 · History
Domain-specific #: 2879
Origin domain: crystallography
Subdomain: crystal chemistry
Aliases: Structure-field map, SFM

Core Idea¶

A Structure Field Map (SFM) is an empirical crystal-chemical representation that places compositionally comparable materials in a low-dimensional coordinate system defined by constituent descriptors, labels the resulting points by observed crystal-structure type, and interprets neighborhoods or regions occupied by the same type as structure fields. Its characteristic question is not merely “what is this compound's structure?” but “how do known structure types distribute across a declared chemical-descriptor space, and what candidate structures does the local pattern suggest for an unmeasured composition?” Early ionic-size studies, systematic ternary compilations, Pettifor's chemical-scale maps, and database-generated maps retain this same role structure while changing descriptors and implementation.^[1]^[2]^[3]^[4]

The map's input is a bounded comparison class: for example, oxides with a common stoichiometry and charge pairing, or binary alloys at one stoichiometric ratio. Each constituent is converted to one or more declared descriptors, such as effective ionic radius or a chemical scale. A compound becomes a point whose coordinates are derived from those descriptors. Its experimentally established structure type, prototype, or consistently defined structural class supplies the label. If chemically neighboring points repeatedly share labels, their occupied region becomes an empirical field. A new compound placed in or near that field acquires a candidate-structure prior, not a theorem about its equilibrium structure.

The locked identity is fixed material family and stoichiometric/valence slice + documented constituent-to-descriptor convention + curated compounds under stated conditions + consistent structure-type labels + coordinate placement + observed same-type concentration or field + exception and sparsity accounting + bounded classification or candidate-ranking use → an interpretable descriptor-to-crystal-structure map. A plotted collection without the comparison-class contract, structure labels, or field/neighborhood interpretation is only a scatterplot. A colored region without the compound evidence and descriptor rule is an illustration, not an SFM.

This abstraction survives independently because its method recurs across distinct crystal-chemical practices. Keith and Roy related double-oxide structures to ionic-radius ratio and charge; Muller and Roy systematized large ternary structural families using ionic-radius and crystal-chemical organization; Pettifor replaced radius-like coordinates with a chemical scale and showed structural separation for hundreds of binary compounds; Morgan, Rodgers, and Ceder formalized database cleaning, neighbor lookup, cross-validation, and ranked candidate lists.^[1]^[2]^[3]^[4] Those are not one software product or one diagram. They instantiate a repeatable way of organizing known structures and obtaining auditable, limited predictions from their geometry.

Structural Signature¶

the comparison class — a declared family within which points are intended to be comparable, normally constrained by chemistry, stoichiometry, oxidation-state pairing, and observation conditions;
the material instances — experimentally characterized or otherwise explicitly qualified compounds whose compositions and structure assignments are traceable to records;
the descriptor scheme — one or more reproducible functions that turn constituents or compositions into coordinate values, such as effective ionic radii or Pettifor chemical-scale numbers;
the descriptor convention — charge, coordination, spin state, ordering, axis orientation, missing-value, and tie-handling rules required to make coordinate assignment reproducible;
the coordinate space — typically two-dimensional, though higher-dimensional mappings are possible, in which each admissible compound receives a point;
the structure vocabulary — consistently defined crystal-structure types, prototypes, space groups, or another declared structural labeling grain;
the labeled reference set — points paired with known structure labels, rather than unlabeled compositions alone;
the field or neighborhood relation — the empirical concentration, separation, overlap, or local frequency of structure labels in descriptor space;
the exceptions and ambiguities — polymorphs, condition-dependent phases, uncertain assignments, duplicate records, overlapping fields, sparse regions, and structures represented only once;
the inference rule — a stated procedure, informal or algorithmic, that uses a new point's field or nearest labeled neighbors to organize, classify, or rank candidate structures;
the validation frame — holdout tests, cross-validation, retrospective checks, experimental confirmation, first-principles calculations, or another method for estimating where the map's inference succeeds;
the applicability claim — a bounded statement about which stoichiometries, conditions, chemical systems, and label definitions the observed field structure supports.

Recognition test. A method is an SFM when comparable crystalline compounds are assigned reproducible descriptor coordinates, known structure labels organize those coordinates into interpretable fields or neighborhoods, and that organization is used to compare structures or generate bounded candidate-structure expectations. If the axes are thermodynamic control variables and the regions assert equilibrium phases, it is a phase diagram. If the output is only an optimized atomic configuration or energy ranking, it is crystal-structure prediction. If it merely classifies without retaining a descriptor-coordinate map and local field relation, it is a broader classifier.

What It Is Not¶

Not a crystal lattice or crystal structure. A crystal lattice describes periodic order in one structure. An SFM compares many compounds and relates chemical descriptors to their structure labels.
Not a phase diagram. A phase diagram represents stable phases and coexistence boundaries as functions of thermodynamic or compositional controls. SFM boundaries summarize an empirical distribution of labeled structures in descriptor space; crossing one does not by itself establish a phase transition, free-energy equality, or coexistence line.
Not unrestricted crystal-structure prediction. Modern prediction searches atomic configurations and evaluates energetic or statistical models. An SFM narrows attention to structure types suggested by mapped neighbors; it neither supplies complete atomic coordinates nor proves the global free-energy minimum.
Not the Goldschmidt tolerance factor. A tolerance factor is a scalar geometric relation among ionic sizes for a declared structure family. It may supply a coordinate or explanatory variable within a map, but it is not the map's labeled reference population and field geometry.
Not an arbitrary scatterplot. A scatterplot can show any two quantities. The SFM requires a crystal-chemical comparison class, structure labels, reproducible descriptor conventions, and a field or neighborhood interpretation.
Not generic clustering. Same-type concentrations can look like clusters, but many SFMs begin with experimentally assigned labels and ask whether those labels separate. An unsupervised clustering algorithm need not know a crystal structure type; an SFM does.
Not every supervised classifier. A black-box model that maps a high-dimensional composition vector to a class can perform a related task without being an SFM. The map preserves interpretable coordinate placement and neighborhood/field relations as the medium of reasoning.
Not a structure–property map in general. Property maps color or partition a materials space by conductivity, hardness, magnetism, or another response. An SFM's defining response is crystal-structure class, even though property overlays can be added.
Not a periodic table or chemical taxonomy. A chemical scale may order elements, but the SFM is the compound-level relation between descriptor positions and observed structure types.
Not a certainty surface. Empirical structure fields can overlap; databases can be incomplete or inconsistent; polymorphism and experimental conditions can assign more than one structure to a composition. The name “field” does not guarantee a sharp physical boundary.

Scope of Application¶

The most literal scope is inorganic crystallography, crystal chemistry, mineralogy, alloy science, and related materials informatics. Ionic-radius maps are natural for families in which charge balance, coordination, and cation size strongly organize feasible structure types. The 1954 Keith–Roy study examined structural relations among double oxides of trivalent elements and demonstrated ionic-radius ratio and cation charge as controlling variables in the studied systems.^[1] Muller and Roy's 1974 treatment assembled major ternary structural families and explicitly organized ionic radii, formula families, polymorphs, and structure-field maps within a broad crystal-chemical reference work.^[2]

Pettifor maps broaden the descriptor design. Pettifor introduced a chemical scale that placed binary compounds of a fixed stoichiometry in a two-dimensional constituent-coordinate map. His reported map of 574 AB compounds exhibited strong separation among structure types, including complete separation within the stated subset of 75 octet sp–sp compounds.^[3] That result supports the abstraction's empirical-field role, but it is not a universal accuracy guarantee for arbitrary materials, stoichiometries, or later database states.

Database-mediated practice makes the method reproducible and testable. CRYSTMET stored critically evaluated structural and bibliographic records for metals, alloys, intermetallics, and minerals, and its NIST account discusses chemical-element maps and structure/property-map applications.^[5] Morgan and colleagues then specified algorithms for cleaning database records, generating Pettifor maps, ranking nearby candidate structures, and testing the procedure by cross-validation for AB and A3B alloys.^[4] The method can therefore be hand-drawn, handbook-based, or automated; software is optional, while the comparison, coordinate, label, and field roles are not.

Applicability does not automatically cross stoichiometry, temperature, pressure, valence slice, or descriptor convention. Morgan and colleagues treated AB and A3B maps as independent and removed nonstandard-condition records for their particular assessment.^[4] Ionic radii themselves depend on oxidation state, coordination, spin state, distortion, occupancy, and bonding environment; Shannon's revised tables make these dependencies explicit.^[6] A map that silently mixes them can create apparent fields from inconsistent inputs.

Clarity¶

Consider an oxide family ABO3. The author first fixes one charge pairing, such as A2+B4+O3, rather than mixing it with A3+B3+O3. For each known compound, the chosen radius table supplies an A-site effective ionic radius on the horizontal axis and a B-site radius on the vertical axis, using declared coordination and spin assumptions. The compound's experimentally observed room-condition structure prototype supplies its label. Points labeled perovskite-like may occupy one region, while other prototypes occupy neighboring regions; overlaps and anomalous points remain visible rather than being erased.

A proposed A'B'O3 composition then receives coordinates under exactly the same convention. If nearby reference points overwhelmingly carry one prototype, that prototype becomes a reasonable first candidate for synthesis planning or computation. The inference is: “under this map's comparison and condition frame, local precedents favor this structure type.” It is not: “ionic radii mathematically force this structure,” “this phase is globally stable at every temperature and pressure,” or “the map determines the atomic coordinates.” Experimental structure determination or energetic calculation still adjudicates the candidate.

The same recognition test works for a Pettifor map. Replace radii with the declared chemical-scale values of constituents A and B; restrict the dataset to one stoichiometry; label known compounds by a consistent structure-type and space-group convention; then use nearest neighbors to rank candidate structures. The materials and descriptor names change, but the roles do not.

Manages Complexity¶

The space of chemical compositions is combinatorial, while a structure assignment contains detailed symmetry, atomic positions, coordination, and condition information. An SFM compresses that problem in three stages. First, it substitutes a small set of chemically motivated descriptors for a full constituent description. Second, it reduces detailed structures to a consistent type or prototype vocabulary. Third, it uses the geometry of many labeled examples to make repeated structure–chemistry regularities visible at once.

That compression supports browsing and triage. A crystallographer can see where a structure family is dense, where rival structures overlap, where the data are sparse, and which records appear inconsistent with their neighbors. A synthesis program can use a new composition's neighborhood to choose a short list of prototypes rather than searching every known structure. A database curator can identify duplicate, contradictory, or poorly conditioned records that disrupt otherwise coherent fields. Morgan and colleagues' automated study shows how the familiar visual intuition can become a defined candidate-ranking and cross-validation workflow rather than an unexamined eyeballing exercise.^[4]

The map manages complexity only by discarding information. Coordination, bonding, electronic structure, kinetics, pressure, temperature, defects, and polymorph stability may not fit into the selected axes. The proper response is not to treat every exception as noise but to ask whether an omitted variable, incorrect record, or genuinely nonlocal structural mechanism explains it.

Abstract Reasoning¶

If two maps use different ionic-radius tables, coordination numbers, or spin-state assumptions, their point geometry is not directly comparable even when the axes share the same labels.
If a compound changes structure with pressure or temperature, one unqualified point cannot represent every polymorph. The observation condition must be attached to the label or the map restricted to one condition frame.
If oxidation-state pairings are mixed in one ABO3 map, apparent proximity can compare chemically different charge-balance regimes. Separate slices or encode valence explicitly.
If a structure type forms a compact field with a few isolated exceptions, first audit the exception records and descriptor conventions; if they survive, treat them as evidence that the chosen descriptors omit a relevant mechanism.
If two structure fields overlap, a hard boundary exaggerates the evidence. Report local frequencies or a ranked candidate list rather than a deterministic class.
If a new point lies far from every labeled point, nearest-neighbor assignment is extrapolation disguised as interpolation. The correct output may be “outside applicability domain.”
If a structure type occurs only once, conventional cross-validation cannot reliably estimate its recoverability; rarity and uniqueness must be reported separately.
If removing duplicate or ambiguous database records sharply changes accuracy, data curation is part of the method rather than neutral preprocessing.
If an alternative descriptor yields cleaner held-out separation, it may be a better mapping for that family, but cleanliness alone does not establish a causal physical variable.
If a map predicts the correct prototype among five candidates, that is useful search-space reduction but not equivalent to exact top-one prediction or atomic structure solution.
If a phase-diagram boundary is overlaid, the reader must know which regions are empirical structure fields and which are thermodynamic equilibrium claims.
If an SFM is updated as new structures enter the database, movement of a field boundary can reflect improved evidence rather than a physical change in old compounds.

Knowledge Transfer¶

Within crystalline-materials research, exact transfer occurs when the same roles are rebuilt for a new family. Oxide crystal chemistry may use effective ionic radii; alloy maps may use a chemical scale; a higher-dimensional informatics implementation may use several constituent descriptors before projecting the result for inspection. Each remains an SFM only if comparison scope, coordinate rule, structure labels, field/neighborhood relation, and bounded inference survive.

The method also transfers between practices. A mineralogist uses it to organize naturally occurring structure families; a solid-state chemist uses it to choose synthesis targets; a database curator uses it to expose contradictory assignments; and a computational materials scientist uses its candidate list to seed first-principles calculations. The outcome and evidence standard change, but the structural relation is literal: known labeled structures occupy descriptor space and local geometry guides the next judgment.

Outside crystallography, “map labeled examples into a descriptor space and reason from neighborhoods” belongs to broader Representation, Classification, Clustering, and Projection. A biological phenotype map or marketing segmentation chart can share that skeleton without becoming a Structure Field Map. The domain-specific name should not travel when crystal structures, composition families, and chemical descriptors have disappeared. That bounded transfer is why the candidate is a domain-specific abstraction rather than a prime.

Examples¶

Ionic-size organization of double oxides. Keith and Roy reported structures among equimolar and 3:5 combinations of trivalent-element sesquioxides and examined ionic-radius ratio and cation charge as structural controls.^[1] In SFM terms, the composition family and charge regime define the comparison slice, ionic dimensions provide descriptors, observed structures supply labels, and the distribution relates size compatibility to structural prototype. The study also illustrates why an early map is not a timeless universal rule: its claims are limited to investigated chemistries and then-available structure determinations.

Pettifor chemical-scale map. Pettifor assigned each element a chemical-scale value and plotted binary compounds of a given stoichiometry in (χA, χB) coordinates. The 574-compound AB example showed strong structural separation.^[3] Here the constituent scale replaces ionic radius, but the SFM signature remains intact: bounded family, descriptor coordinates, known structure labels, and fields whose separation supports comparison and expectation.

Automated candidate ranking from CRYSTMET. Morgan, Rodgers, and Ceder cleaned AB and A3B database records, generated maps, formalized neighbor lookup as a ranked candidate list, and evaluated predictions through cross-validation.^[4] Under their stated dataset conditions, a five-structure candidate list contained the correct structure about 86% of the time for the tested unknown-alloy setting. That number belongs to the particular data, cleaning, stoichiometries, and evaluation rule; the enduring abstraction is the conversion of an interpretable map into a testable candidate-prior workflow.

Hypothetical synthesis triage. A researcher proposes a previously unreported ternary oxide. They select the map matching its stoichiometry and valence pairing, calculate coordinates with the declared Shannon-radius conventions, inspect nearby compounds, and retain the two locally dominant prototypes plus one chemically motivated outlier. Those candidates seed density-functional calculations and experimental diffraction comparisons. The SFM has reduced the search, while the later methods decide stability and realized structure.

Structural Tensions¶

Interpretability versus omitted physics. Two axes make chemical trends visible and auditable, but they cannot encode every bonding, electronic, kinetic, and condition-dependent influence. Diagnostic: do exceptions cluster around a plausible omitted variable, or are they scattered as record errors?

Separation versus honest overlap. Clean fields support compact rules, while real families can contain competing polymorphs and boundary compounds. Drawing crisp regions improves readability but can convert a probabilistic neighborhood into false certainty. Diagnostic: do held-out examples support a hard field, or only ranked local candidates?

Stable convention versus context-sensitive descriptors. Reusing one descriptor table makes maps comparable; ionic radius, however, varies with oxidation state, coordination, spin, distortion, occupancy, and bonding.^[6] Diagnostic: is the selected value a reproducible convention appropriate to each mapped site, or a convenient number detached from its coordination context?

Database scale versus database inconsistency. Large crystallographic databases enable automation and rare-pattern discovery, yet duplicate, ambiguous, or condition-mixed records can manufacture conflicts. Diagnostic: does the claimed field survive deduplication, label harmonization, and condition filtering?

Useful prior versus self-confirming search. A short candidate list makes computation and experiment tractable, but repeatedly testing only familiar prototypes can miss a novel structure outside existing fields. Diagnostic: does the workflow retain an exploration path for out-of-domain compositions and poorly represented structure types?

Structural–Framed Character¶

Structure Field Map is mixed-framed (0.58, boundary true). Its geometric skeleton is structural: points, coordinates, labels, neighborhoods, separation, and overlap can be recognized in any mapped dataset. It is evaluatively light; no structure field is intrinsically good or bad. Yet the named abstraction does not exist without significant scientific-practice framing. Researchers decide the admissible chemical family, stoichiometry and charge slice, descriptor convention, structure-label grain, treatment of polymorphs, database cleaning, neighborhood algorithm, and evidential threshold.

The natural structures are not created by the map, but the fields are purpose-relative summaries of which observations were included and how they were encoded. A different radius convention, condition filter, or prototype equivalence rule can move points and boundaries without any material changing. The term therefore travels exactly within crystal chemistry and materials informatics, while transfer to arbitrary labeled coordinate plots becomes import of a metaphor or reduction to broader primes.

Structural Core vs. Domain Accent¶

The portable core is labeled cases mapped into a descriptor space, with local geometric regularity used to organize classes and form bounded predictions. That core is already covered by Representation, Classification, Clustering, Comparison, and Projection. It can appear in ecology, medicine, linguistics, or commerce without crystal structures.

The domain accent is constitutive rather than decorative: chemical constituents and stoichiometry define the instances; ionic radii, chemical scales, valence, and site roles define coordinates; crystallographic prototypes or structure types define labels; experimental condition and polymorphism qualify assignments; and synthesis, diffraction, database curation, or first-principles calculation supplies validation. Remove those roles and “structure field map” no longer denotes this materials method.

This split defeats prime classification. The same coordinate-and-neighborhood skeleton recurs broadly, but its substrate-independent identity already has catalog homes. What is novel here is precisely the crystalline-material specialization: a stable empirical apparatus for organizing and predicting structure types from chemically meaningful descriptors. The cross-domain generality belongs to its parents and related primes, not to the named SFM.

Representation. The SFM maps a target population of materials into a coordinate-and-label medium, preserving selected chemical similarities and observed structural classes while explicitly dropping much microscopic detail. This is the proposed strict DAG parent.
Classification. Known structures supply classes, and new compositions may receive candidate labels. Classification captures the assignment role but not the spatial descriptor medium or neighborhood evidence.
Clustering. Same-type points may concentrate into fields. Clustering describes that distributional pattern, but labels can be supplied before the field is observed and need not be generated unsupervised.
Projection. A high-dimensional chemical description is reduced to a small coordinate display. Projection supports the map but does not provide its crystalline labels or inference rule.
Comparison. Placing many compounds in one convention-governed space makes similarities, boundary cases, and exceptions comparable.
Phase Diagram. This is an important contrast relation, not a parent: both use regions on a map, but their axes and region semantics differ.

Relationships to Other Abstractions¶

Current abstraction Structure Field Map Domain-specific

Parents (1) — more general patterns this builds on

Structure Field Map is a kind of Representation Prime

Representation. The SFM maps a target population of materials into a coordinate-and-label medium, preserving selected chemical similarities and observed structural classes while explicitly dropping much microscopic detail.

Hierarchy path (1) — routes to 1 parentless root

Structure Field Map → Representation → Abstraction

Neighborhood in Abstraction Space¶

Structure Field Map sits in a sparse region of the domain-specific corpus (90^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Crystallographic Coordinates & Symmetry (5 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Crystal Lattice. One periodic arrangement of a basis over translations, versus a comparative map whose labeled points refer to many structure types.
Crystal-structure database. A database stores records and may feed a map. The SFM is a derived coordinate-and-field representation, not the repository itself.
Phase Diagram. Equilibrium phase stability over thermodynamic/compositional controls, versus empirical structure-type occupancy over selected descriptors.
Goldschmidt Tolerance Factor. A scalar geometrical indicator for particular structural families, versus the full labeled population and field relation. It can be one input to an SFM.
Pettifor Map. A major subtype of SFM using Pettifor's chemical scale, not a synonym for every radius-, electronegativity-, or multi-descriptor structure map.
Structure–Property Map. A broader map whose response may be a material property. SFM identity specifically requires crystal-structure class as the organized outcome.
Crystal-Structure Prediction. An umbrella problem and family of energetic, evolutionary, statistical, or other methods. SFM is one empirical candidate-prior approach and does not solve atomic coordinates or equilibrium stability alone.
Scatterplot or dimensionality-reduction visualization. These are representational forms. They become an SFM only with crystalline comparison scope, structure labels, descriptor conventions, and field-based reasoning.
Topographic Map. Both use spatialized fields, but an SFM's geometry is abstract descriptor proximity rather than geographical position or terrain elevation.

References¶

[1] M. L. Keith and Rustum Roy, “Structural Relations Among Double Oxides of Trivalent Elements,” American Mineralogist 39 (1954): 1–23. Primary article PDF. registry ↩a ↩b ↩c ↩d

[2] Olaf Muller and Rustum Roy, The Major Ternary Structural Families (Springer-Verlag, 1974), 487 pp., ISBN 978-0-387-06430-7. Bibliographic and contents record. registry ↩a ↩b ↩c

[3] D. G. Pettifor, “A Chemical Scale for Crystal-Structure Maps,” Solid State Communications 51, no. 1 (1984): 31–34. DOI: 10.1016/0038-1098(84)90765-8. registry ↩a ↩b ↩c ↩d

[4] Dane Morgan, John Rodgers, and Gerbrand Ceder, “Automatic Construction, Implementation and Assessment of Pettifor Maps,” Journal of Physics: Condensed Matter 15, no. 25 (2003): 4361–4369. DOI: 10.1088/0953-8984/15/25/307; author-hosted full text. registry ↩a ↩b ↩c ↩d ↩e ↩f

[5] Gordon H. Wood, John R. Rodgers, S. Roger Gough, and Pierre Villars, “CRYSTMET—The NRCC Metals Crystallographic Data File,” Journal of Research of the National Institute of Standards and Technology 101, no. 3 (1996): 205–215. DOI: 10.6028/jres.101.021; NIST/PMC full text. registry ↩

[6] R. D. Shannon, “Revised Effective Ionic Radii and Systematic Studies of Interatomic Distances in Halides and Chalcogenides,” Acta Crystallographica Section A 32 (1976): 751–767. DOI: 10.1107/S0567739476001551. registry ↩a ↩b

[7] D. G. Pettifor, “Structure Maps Revisited,” Journal of Physics: Condensed Matter 15, no. 25 (2003): V13–V16. DOI: 10.1088/0953-8984/15/25/402. registry