An evolutionary system of mineralogy: Proposal for a classification of planetary materials based on natural kind clustering

Natural kinds that have been lumped together by the IMA classification

The IMA mineral species criteria of idealized major element composition and crystal structure in some instances lump together demonstrably distinct natural kinds of minerals. Diamond offers a dramatic example (Table 1). The first mineral in the cosmos was nanocrystalline diamond that condensed from incandescent, expanding, and cooling gases ejected from the first generation of large stars more than 13 billion years ago (Hazen et al. 2008; Ott 2009). These vapor-deposited nanodiamonds, which still must form in vast numbers around energetic stars across the universe today and that are distinct from all other diamond populations in their origins, morphologies, isotopic

compositions, and other properties [for example, anomalous microwave emissions (Greaves et al. 2018)], fall to Earth in chondrite meteorites and pre-solar grains.

Contrast these well-traveled nanodiamonds with two populations of mantle-derived macroscopic crystals, some formed from deep, high-pressure, carbon-bearing aqueous solutions, including “Type I” diamonds with IR- and UV-absorbing nitrogen impurities (Davis 1984; Shirey et al. 2013; Sverjensky and Huang 2015), and others from deep, high-pressure, carbon-saturated Fe-Ni melts (including nitrogen-poor, IR-transparent “Type II” diamonds; Smith et al. 2016). Equally distinct is the enigmatic “carbonado” form of diamond, which typically consists of porous, black, superhard sintered masses (Heaney et al. 2005; Garai et al. 2006). In addition, microscopic diamond with a disordered structure (equivalent to the now discredited “lonsdaleite”) can form from carbonaceous material subject to the shock of large impacts (Vishnevsky and Raitala 2000; Németh et al. 2014). Thus, from the standpoint of cosmic evolution, the mineral species “diamond” can be viewed as at least five different natural kinds of minerals, each with a distinct set of properties, mode of origin, and age range of formation (Table 1).

Similar arguments can be presented for many common minerals. Microscopic hydroxylapatite grains in chondrite meteorites (Papike 1998) differ in several respects from the coarser euhedral hydroxylapatite crystals in granitic pegmatite, which are in turn distinct from the hydroxylapatite/biopolymer composites of teeth and bones (Harlov and Rakovan 2015). Likewise, quartz displays multiple kinds: in granite and granite pegmatite, in hydrothermal veins, in quartzite, and in the biosilica of diatom skeletons (Heaney et al. 1994; Wysokowski et al. 2018). Pyrite differs in polymetallic veins, in black shales, and pyritized brachiopods (Rickard 2015); while calcite occurs in abiotic precipitates, biomineralized shells, and a dozen other contexts (Reeder 1983; Dove 2010).

Mineral solid solutions offer additional examples of the lumping of distinctive natural kinds. Of special note are the plagioclase feldspars, the most abundant minerals in Earth’s crust (Rudnick and Gao 2005) and probably on Mars (Milam et al. 2010). The official division of plagioclase into two idealized end-member species, albite (NaAlSi₃O₈) and anorthite (CaAl₂Si₂O₈), does not reflect the varied paragenetic modes of intermediate compositions (Klein and Hurlbut 1980; Wenk and Bulakh 2004). Andesine, oligoclase, labradorite, and bytownite, though admittedly defined by somewhat arbitrary and overlapping criteria, long served petrologists as useful natural mineral categories (e.g., Tilley et al. 1964).

Therefore, in the planetary context of understanding mineral diversity and distribution through deep time, especially in enumerating the evolving numbers of mineral kinds and their global distributions, many IMA-approved mineral species should be split into two or more distinctive natural kinds, each with a quantifiable combination (i.e., clustering) of physical and chemical properties.

Natural kinds that have been artificially split by the IMA classification

In many instances, especially in mineral structure groups with great compositional plasticity, the division of minerals into numerous species according to ideal end-member compositions potentially results in the artificial splitting of natural kinds. By imposing compositional boundaries between species of minerals that occur in a continuous solid solution, the present mineral classification scheme does not adequately reflect natural kinds with wide ranges of solid solution.

The tourmaline supergroup is a case in point. The IMA now recognizes at least 32 approved species of tourmaline (with 8 more species pending), each with a different distribution of major elements among six of its crystallographic sites (Henry et al. 2011; Grew et al. 2017; http://rruff.info/ima, accessed October 30, 2018). Given natural compositional variations, this classification means that an individual thin section may hold two or more different tourmaline species (e.g., Henry et al. 2011; Grew et al. 2015), even if all tourmaline grains were formed in the same petrogenetic event. Indeed, individual grains that grow during a single magmatic or metamorphic event can display major element zoning that modulates among more than one tourmaline species. (Note that it is also possible for an individual mineral grain to represent two natural kinds: a core of igneous tourmaline or zircon, for example, with a hydrothermally deposited rim.) A division of tourmaline into natural kinds might thus provide a more parsimonious description of the supergroup and would facilitate a more accurate understanding of boron mineral diversity, distribution, and evolution (Grew et al. 2017).

The black mica known as “biotite,” which is distinguished in hand specimen and thin section based on optical properties and morphology, has now been subdivided into several mica group mineral species, including annite, fluorannite, siderophyllite, and tetraferriphlogopite (Rieder et al. 1998). Similar splitting of natural kinds into many species occurs in the complex amphibole group with more than 100 approved species (Hawthorne et al. 2012; http://rruff.info/ima), the garnet group (Grew et al. 2013), oxide and sulfide spinels (Biagioni and Pasero 2014), and other rock-forming mineral groups.

Binary solid solutions, including ferromagnesian olivine, orthopyroxene, and many other examples, underscore the potentially artificial nature of classification based only on major element composition and crystal structure. Any rock that incorporates olivine grains with close to a 50:50 ratio of Mg:Fe will likely technically contain two species, both forsterite (Mg₂SiO₄) and fayalite (Fe₂SiO₄), even though all olivine grains arose from a single petrogenetic event.

Current conventions for recognizing rare earth element (REE) mineral species provide additional potential cases of splitting natural kinds. Under the current system, parisite-(Ce) and parasite-(La) are distinct species, based on differences in the most abundant REE, even though both species contain the full suite of REE in solid solution and are represented by the general formula Ca(REE)₂(CO₃)₃F₂ (Table 2). Similar arguments could be made for some of the multiplicity of species of almost 40 other REE mineral structures (including Ce-, La-, Nd-, and Sm-dominant variants of florencite, a REE aluminum phosphate; Table 2). In some cases, these split species represent a single natural kind with one stability field that incorporates yttrium and a range of light and heavy REE and thus one paragenetic mode.

In each of the above examples, the end-member compositions of diverse mineral structural groups are useful idealized thermodynamic constructs; the present IMA-approved mineral species should be retained as the primary systematic means to identify minerals. However, in studies of mineral evolution and mineral ecology, which are based on the diversity and distribution of minerals through time and space, this splitting of natural mineral kinds into multiple species obscures relationships that determine mineral co-occurrence in varied paragenetic environments. As in the case of splitting a single mineral species into two or more natural kinds, the lumping of several species into a single natural kind affects any estimates of total mineral diversity and distribution.

Natural kinds of non-crystalline planetary materials

The definition of a mineral as a naturally occurring crystalline material is deeply ingrained, yet it arbitrarily excludes significant volumes of condensed planetary materials from formal mineralogical consideration. Volcanic glass, solidified silica gel, shungite, amber, composite materials such as coal and mixed-layer clays, and natural nanomaterials such as carbon nanotubes and buckeyballs are among the many potentially important condensed solid phases that may play key roles in our understanding of planetary evolution, yet which lie outside the purview of modern mineralogy (Rogers 1917; Povarennykh 1986; Povarennykh et al. 2018).

This mineralogical requirement of crystallinity may lead to biases when attempting to understand deposits rich in amorphous and nanoscale materials. For example, recent data from NASA’s Mars Science Laboratory reveals that some martian soils contain greater than 50 wt% amorphous material (e.g., Morrison et al. 2018). By expanding the classification of mineral kinds to consider a broader range of characteristic physical and chemical properties—i.e., beyond materials with a strictly periodic atomic structure—the potential exists to enrich mineralogy while better representing the actual materials that make up planets.

Creating data resources

Quantitative identification of natural mineral kinds (e.g., the several kinds of diamond in Table 1) will emerge from analysis of extensive, reliable, and open-access data resources that are not yet generally available for most mineral species and groups. Consequently, the mineralogical community needs to create an accurate and comprehensive tabulation of varied mineral attributes. The defining attributes will vary for different mineral species and groups, but will always include trace and minor elements. Additional parameters of interest might include isotopes of major, minor, and trace elements; crystal size, morphology, twinning, and defects; optical, electrical, magnetic, and elastic properties; the mineral’s biological context, including local microbiota, fauna, and flora; age of formation; associated minerals, petrological context, and tectonic setting; and climate zone, local aqueous chemistry, and other environmental parameters. Each of these variables adds information that collectively has the potential to differentiate natural kinds through multi-dimensional analysis, and which might elucidate the origins and subsequent history of a mineral specimen in the context of planetary evolution.

Building robust data resources is thus the next step in implementing this proposed evolutionary system of mineralogy, both through “brute-force” compilations (i.e., transcription of literature data by hand) and through emerging machine learning methods that can rapidly scan pre-existing literature (e.g., https://geodeepdive.org; accessed October 31, 2018). A major ongoing challenge to the international mineralogy and petrology community is to promote a culture of data sharing, including the retrieval of “dark data” and the implementation of FAIR data policies by publications and societies (Downs 2006; Lehnert et al. 2007; Hazen 2014; Wilkinson et al. 2016).

Applying natural kind cluster analysis

“Cluster analysis” is a constellation of analytical methods that group objects into subsets (clusters) whose members are more alike than objects outside the cluster (e.g., Bailey 1994; Everitt 2011). Expanded mineral data resources on numerous specimens with many attributes are well suited for varied analytical approaches under the cluster analysis umbrella. Though there exists no single definition of a “cluster,” two groups of minerals could be represented as distinct “kinds” if cluster analysis of any combination of attributes results in two non-overlapping groups in n-dimensional space.

In this endeavor, mineralogists can learn important lessons from other scientific disciplines that have embraced the classifications system based on natural kind clusters (Boyd 1991, 1999; Millikan 1999). Paleontologists and biologists have long employed the data-driven, cluster approach of morphometric analysis, for example, to distinguish superficially similar species from ontonogenic sequences (Lohmann 1983; Bookstein 1991; Ashraf 2004; Turvey et al. 2018). In this formalism, the key to recognizing distinct natural kinds is the collection and analysis of numerous individual specimens to obtain statistically meaningful distributions of traits related to size and shape.

A revealing recent example is the application of cluster analysis to discern four major human personality types (Gerlach et al. 2018), based on analysis of more than 1.5 million individual results from the Five-Factor model of human personality traits (e.g., Widiger 2015). Their study reveals both the promise and potential pitfalls of cluster analysis. They initially employed an unsupervised Gaussian mixture model, which led to an unreasonably large number of discrete clusters—a difficulty that can arise in cluster analysis if groups differ significantly in size (Burnham and Anderson 2002; Lancichinetti et al. 2015). Gerlach et al. employed varied statistical tests to reveal that four major clusters provide an optimal fit to this large data set. The authors thus emphasize “limitiations of unsupervised machine learning methods to the analysis of big data.”

A similar strategy can be applied to the implementation of an evolutionary system of mineralogy, which must be predicated on the analysis and comparison of numerous specimens from different mineral-forming environments. As with classifications in other domains, the recognition of mineral natural kinds will ultimately depend on the analytical comparison of large numbers of individual mineral descriptions, with richly varied information, data analysis, and visualization.

A significant challenge lies in determining what constitutes a mineral cluster. Ambiguity arises because no one algorithm or set of criteria can be universally applied to discern optimal clustering (Jain 2010). Experts in specific mineral species or groups must therefore supervise the application of cluster analysis, for example by selecting the most important attributes and constraining the total number of clusters. Each natural kind should be defined by continuous ranges of multiple attributes that arise from a well-defined paragenetic process—ideally a combination of characteristics that do not overlap with those of any other mineral kind. For example, vapor-deposited nanodiamonds from the expanding and cooling gaseous envelopes of energetic stars possess physical and chemical characteristics that are inherently different from those of high-pressure/temperature diamonds formed in Earth’s mantle and thus will form a separate cluster.

Identifications of natural kinds, including both the lumping and the splitting of existing IMA species, is context dependent and must be considered on a case-by-case basis. Depending on one’s focus, the resulting number of clusters may differ significantly. For example, one researcher studying hydroxylapatite from the broad-brush perspective of billions of years of Earth history might choose to divide all specimens into three clusters: meteoritic, crustal, and biomineralized. By contrast, a biologist examining hydroxylapatite from a different evolutionary viewpoint might recognize multiple types of biomineralization with different kinds of bio-apatite in brachiopods, fish scales, cartilage, teeth, and bones (Roy 1974; Onozato 1979; Ohirta 1986; Sherman 2008).

The flexible, context-dependent nature of cluster analysis renders this approach ill-suited to be the primary classification system for minerals. Nevertheless, recognition of distinct mineral kinds is essential if we are to understand how the mineralogy of planets evolves through a succession of physical, chemical, and biological processes.

Developing a nomenclature for mineral natural kinds

Unambiguous, standardized mineralogical nomenclature is essential. In spite of attempts to introduce systematic and rational approaches to mineralogical nomenclature (e.g., Nickel and Grice 1998), relatively few of the more than 5300 approved mineral names provide useful clues regarding chemical or physical attributes of species, much less their varied modes of occurrence. The community of Earth scientists would be ill served by any system that adds layers of taxonomic obscurity on the existing scheme.

A logical solution in the case of an evolutionary system is to employ descriptive paragenetic modifiers to the existing IMA mineral names. In the case of diamond, for example, names for several natural kinds already exist. “Presolar diamond” and “impact diamond” are self-explanatory, whereas “type I,” “type II,” and “carbonado” are well-established varietal names for diamond. In each case the name “diamond” is retained, but with a familiar modifier. In the context of planetary evolution, recognizing and naming mineral natural kinds in this manner will enhance our ability to communicate stages of mineral evolution with clarity.

Finally, it is important to recognize that in the majority of instances, especially rare minerals with only one known mode of formation (Hazen and Ausubel 2016), natural mineral kinds will be exactly equivalent to IMA mineral species. Therefore, in most instances mineralogical nomenclature will require no modification.