Beyond dematerialization and inscription

1. Introduction

Historically, the development of classical structural theory in chemistry has been concomitant with the development of non-textual, graphical techniques to render the molecular structures that structural theories, hypothesis and concepts were ostensibly about. These techniques are now so entrenched that is difficult to imagine that we could talk, write or even think about molecular structures without recourse to them. Chemists have developed and adopted over time a number of techniques and conventions to depict – to visually represent – the three-dimensional structure of molecules on a two-dimensional graphic space.[1] These allow the production of a variety of more or less elaborate structural representations which literally pepper the documents of chemists; the polysemic, yet consistent, rendering of what constitutes for all practical purposes a molecule’s identity.[2]

Another related approach commonly adopted for the representation of molecular structures in chemistry has been that of physical molecular modeling, i.e., the representation, or rendering, of molecular structures through actual physical structures, commonly referred to as molecular models_m[3], or less commonly as physical models_m or structural models_m. This approach has been a constitutive practice of modern chemistry (Francoeur 1997) and its origin can be traced as far back as its graphical counterpart (Ramsay 1974a; Ramsay 1974b). The practice of physical molecular modeling_m is basically rather simple. It is the production and use of three-dimensional structures that render, more or less to scale, the spatial position of atoms (or groups of atoms) in a molecular structure, as well as the bonds between them. Most molecular models_m are assembled from commercially available modeling kits, which generally consist of modular elements (atomic units, or species, and connectors) that allow a vast number of different structures to be assembled. Although a seemingly simple task, the practice of molecular modeling_m has led to the proliferation of conventions and tools.[4]

Recently, Pierre Laszlo (1998), considering specifically the practice of chemical analysis, discussed the conversion of a specific chemical sample, through instrumentalized readings, into a "molecular object", an object which can be made ostensible using one of the conventions discussed above. He described this process of translation as one of dematerialization, concluding that the work of analytical chemists consists mostly of handling mental representations, and that chemistry is to a great extent a "science of mind".[5] This process described here by Laszlo, by which the chemists’ proximal concern becomes detached from the object of chemical analysis (the sample), falls generally within current science studies accounts of instrument-mediated observations in the experimental sciences. On the other hand, underlying his concept of dematerialization is the idea that the representational techniques and conventions of chemistry are no more than a poor reflection of the abstract mental objects which constitute the true and legitimate knowledge of chemistry, crude but necessary, if only for the purpose of communication. The material signs of the chemists are to be considered (following Laszlo’s metaphoric use of language, or rather parole) as no more than a mere supplement, "a storage medium to assist memory inessential to the meaning of language" (Lenoir 1998, p. 5). This position clashes somewhat with current theoretical movements within philosophy, literary studies and science studies which emphasize the materiality of literary and scientific inscriptions, upholding the idea that writing (and inscribing more generally) is "constitutive of meaning rather than a passive medium for restoring the presence of language to thought" (idem.). My aim here is not to resolve this tension. I simply want to suggest that what can be understood in the broadest sense as the materiality of the signs used by chemists in the course of their work does somehow, in some circumstances, matter.

The circumstances I have in mind here are more closely associated with the forefront of chemical research, the development of new concepts and theories, than with routine, entrenched practices such as chemical analysis. Already, Klein (1997; 1999) has made an interesting and convincing argument in the case of Berzelian chemical formulas and the development of the concept of substitution in the 19^th century. Developing the notions of epistemic techniques and paper-tools, she has shown how these Berzelian formulas, as used by the likes of Dumas, were not a simple repository for established knowledge. She argues, for example, that in relation to the development of the concept of substitution, these "chemical formulas were not only the direct referents of the new conception, but also the paper-tools by which it was produced" (Klein 1997, p. 43).

My own goal here is to illustrate and explore how, concretely, from the early days of stereochemistry, chemists have used molecular models_m in various research settings and projects; how, in a literal sense, the use of models_m was an integral part of the articulation of the chemists’ knowledge and experience of molecular structures, as conceptual entities. I make no claims of being exhaustive or systematic in my treatment of this topic, and my case is restricted to organic chemistry. Nevertheless, the examples I offer here will be sufficient, I believe, to show how in some sense the development and extension of what could be called "structural thinking" in chemistry has hinged at specific times upon the working out and the sorting out of the geometrical and mechanical properties of molecular models_m.[6]

Before going any further, let me go back briefly to the "materiality of text and inscription" approach which I discussed briefly above. The concept of inscription is at the heart of this approach in which scientific research is considered in terms of a quasi-literary practice and which focuses on the graphematic materiality of scientific representational devices; a trend which implies a move from representation in scientific practice to representation as scientific practice (Rheinberger 1997). Bruno Latour (1986, 1990, 1993a) has probably been one of the most influential authors in the development of this approach. The process of scientific knowledge production (or rather, the condition of possibility for such production)[7] presented in Latour’s work is invariably one of transition from ‘things-in-themselves’ to a semiological realm; a passage from raw materiality to a graphematic space which allows the full deployment of the "powerful functionalities"[8] of inscriptions[9]. In other words, Latour equates the epistemic dominion over the world with the graphematic two-dimensional space of inscriptions, while the tangibly three-dimensional is explicitly associated with perceptual and epistemic "anomie".[10]

The graphematic ‘condensation’ of chemists’ experience and knowledge of matter in the form of graphical representation of molecular structures falls neatly into the framework elaborated by Latour and other proponents of the ‘semiotic turn’.[11] But their notion of materiality, grounded in the concept of inscription, is of little help when considering the use of molecular models_m in chemical research. In the cases I will be discussing here, we will be looking at scientists involved in the articulation of ideas, hypotheses, traces, diagrams or theoretical statements, not through further graphematic translation, but through their translation into synthetic objects – structural models_m – which become the tool and the means of this articulation. From a Latourian perspective, we are facing a peculiar phenomenon, i.e. cases where physical objects become the privileged mode of access to ‘epistemic things’ at points far downstream from where we expect scientists to have traded ‘things-in-themselves’ for signs and symbols – for inscriptions.[12] Let me consider this point further.

First of all, as I have argued elsewhere (Francoeur 1997), physical models_m in chemistry do not constitute a sudden about-face on the Latourian path that leads from complex three-dimensional objects to simple, less confusing two-dimensional ones, a return to the ‘objects’ scientists left behind by translating them into graphs, diagrams and similar inscriptions. Relative to events and phenomena recorded in the lab, physical and graphical representations of molecular structures are equally ‘eidetic’. Molecular structures, and a fortiori any means of representing them, are arguably not about what molecules ‘look like’.[13] To use a distinction made by Galison (1997), any representation of molecular structure does not stand in a homomorphic relation to a referent ‘out-there’, but only to other structural representations. They should rather be considered as ‘synthetic’ representations, or interpretations, that stand in a homologic relation to empirical events and experimental phenomena – events and phenomena which they can be construed as either explaining or predicting. In other words, and to paraphrase Michael Lynch (1991, p. 208), there is no such thing as comparing the structural representation of a molecule to the ‘real’ thing, since it is through representational work that a molecular structure becomes coherently visible. The realm of molecular structures is thus essentially cultural, i.e. coextensive with the means chemists have given themselves to show, talk about and work with these structures – means which are, ceteris paribus, epistemically equivalent[14] while phenomenologically distinct.

Yet, it seems difficult to shake the idea that three-dimensional physical, models_m are somehow more ‘real’ than their graphical counterparts. For example, in his discussion of graphical methods of representation in engineering, Ferguson claims that the advantage of models_m is that they "can take an observer one step closer to reality than can a drawing [...]". This position appears, at first glance, to contradict the argument of epistemical equivalence. On the other hand, Ferguson’s notion of "reality" is in fact more phenomenological than epistemological:

Let me now move on and illustrate these points by discussing specific uses of models_m in chemical research, using examples taken from the development of stereochemistry and conformational analysis.
 
 

2. Molecular models_m in early stereochemistry: The case of saturated cyclic compounds

In 1885, Adolf von Baeyer reported his ‘strain theory’ of atomic bonds in cyclic compounds (1885). His argument ran roughly like this: given the model of the tetrahedral carbon atom proposed by van’t Hoff, any cyclic compound composed of fewer or more than 5 atoms will see its valency forces deviate from its ideal orientation. Carbon bonds in these compounds are thus ‘bent’, or ‘strained’. On this basis, von Baeyer proposed a structure for cyclohexane that was both planar and strained. This proposition has of course puzzled later chemists for whom molecular models_mc self-evidently imply the non-planarity of cyclohexane (Eliel 1975). O.B. Ramsay (1981) has suggested in particular that since von Baeyer had demonstrated experimentally the relationship between benzene and cyclohexane, he must have felt compelled to extend the planar, hexagonal shape of benzene to cyclohexane. Furthermore, von Baeyer indicated in his article that the sense of his proposition could be made evident by using "Kekulé ball models" (Kekulé’sche Kugelmodelle), an early type of ball-and-stick models_m. These had been designed by Kekulé in the 1860s, thus before the advent of stereochemistry. Their tetrahedral arrangement provided a convenient way of assembling models of double and triple bonded structures and held no further signification for Kekulé. A particularity of these models_m, as Ramsay has pointed out, is that the bond between two carbon atom models is made up of two wires (representing valency forces), each coming out of its respective carbon model, joined in their middle by a flexible joint.[16] This arrangement was quite different from later standard models where the shared covalent bond is represented by a single wire between two atoms, and it does make the strained planar structure consistent with the assumptions built in the models_m that von Baeyer used (Figure 1). What exact role these models played in the initial formulation of the concept of strained, planar, cyclic molecules is not clear, but, as Ramsay concluded, its rationale can be better understood by careful consideration of both von Baeyer’s writings on the topic and the models_m he used (Ramsay 1975).

Five years later, another German chemist, Hermann Sachse, suggested there was no need to postulate a strained structure for cyclohexane, stating instead that two non-planar, strain free conformations[17] were possible (1890). These conformations, which he called unsymmetric and symmetric, were respectively similar to the boat and chair forms familiar to later chemists. No illustrations of these conformations were given in his article. Sachse did however instruct his readers how to assemble models_m of these two forms, using cardboard tetrahedra attached to octahedral shapes. This particular choice of models seems to have been dictated by the way they related to his subsequently published mathematical argument (Sachse 1892). We thus see in both the cases of von Bayer and Sachse a reliance on models_m, rather than simple images, to illustrate and demonstrate their respective (and antithetical) structural concepts. Sachse goes further in discussing the role of models_m. In the course of his argument, Sachse suggests that one of the conformations he proposed, the asymmetric (or boat) conformation has some flexibility and that it could change its shape without any deviation (Abweichung) of the bond angles. He explains that this phenomenon was suggested by his calculations but confirmed through experiments (Versuche) with Kekulé ball models_m. We thus see here clearly how models_m come to define and embody a phenomenon (today referred to ‘twist conformers’). That phenomenon is, regardless of whether it is/was true of actual molecules, witnessably and demonstrably true of certain types of models_m; an unintended emerging property stemming out of intentional or unintentional conditions (constraints and degrees of freedom) built into these models_m.

It was to be decades before the conformations proposed by Sachse were empirically verified and accepted. Von Baeyer’s planar cyclohexane remained dominant in organic chemistry well into the 20^th century (Russell 1975). Sachse had recognized and discussed the possibility of interconversions between the strainless conformations he had proposed, but believed that they did not happen under regular conditions, especially since these interconversions involved a complete or partial passage through the strained conformation defined by von Baeyer. This led to the prediction of a number of isomers whose existence could not be confirmed. Nearly 30 years later, Ernst Mohr, professor of chemistry at Heidelberg, elaborated on Sachse’s hypothesis (Mohr 1918). Mohr suggested in particular that the interconversion, or version, of conformations in cyclohexane happened constantly at room temperature, which, if true, made experimental verification of Sachse’s hypothesis impossible.

Mohr’s illustrations and language in this article make it clear that he is not considering the configuration of disembodied structures, but rather of structures as embodied in models_m (whose origins and characteristics he does not otherwise discuss). This is particularly striking in his discussion of the phenomenon of interconversion (Figure 2). He thus instructs his readers, for example, that "one can transform the trigonal model (Abb. 3) into its mirror image model (Abb. 4), by turning up the atom models directed downward and turning down the atom models directed upwards. Through a still simpler alteration of shape, the rhombic model can be changed into the trigonal model (and vice versa): turn down the 4^th carbon atom in Abb. 1 with a simultaneous rotation of the 3^rd and 5^th atom, so as to obtain Figure 2, which is the trigonal model in a somewhat different orientation than depicted in Figures 3 and 4" (p. 317, my translation). Mohr is obviously inviting his readers to virtually witness manipulation performed on models_m, manipulations which they could repeat should they doubt the description, or should they simply wish get a better understanding of the phenomena. Like Sachse, Mohr reports experiments performed with models_m. Thus, in his discussion of fused double ring structures, he tells us that "As experiments with models show one can construct a fairly large number of such combinations of atoms in a strain-free form. They are without exception spatial constructions; the centers of their carbon atoms do not lie on a plane […]" (p. 321, my translation). On this basis, Mohr postulated two isomers (cis and trans) for decahydronaphtalin which are not interconvertible, since the bond that forms the shared side of the two rings locks the conformations in place.[18] In other words, interconversion cannot take place without breaking that bond.

Seven years later, Wilfred A. Wightman, chemist at the University of Leeds, offered again a modeling study of the cyclohexane structure (1925). Summing up the postulates previously proposed by Sachse and Mohr, he pointed out that "the principal objection to the acceptance of Sachse’s structures as stable configurations is the large number of isomerides they would require" (p. 1422). The boat structures (which Wightman called type II), in particular, would require the existence of several monosubstituted cyclohexanes. Nevertheless, he continues by suggesting that "it is of interest […] to determine whether co-ordinated relative rotations about the single bonds are possible without strain and to examine the consequences of such a phenomenon; to this end, the author has constructed models in which these mechanical requirements are fulfilled" (p. 1423).

Wightman provided no detailed description of the models_m he used,[19] although from the illustrations accompanying his text, it seems quite likely that they were very similar to the models_m used by Sachse and Mohr (i.e. a variation of the ball-and-stick motif). Not surprisingly, perhaps, he repeated the observation that while the chair form of the cyclohexane structure is rigid, the boat form is "loose", i.e. that it can without strain take on different conformations, a situation which, he concludes, "annihilate[s] the isomeric possibilities above mentioned" (idem.). On the other hand, he does not mention that these observations had already been reported by both Sachse and Mohr, a fact he might have conveniently ignored to give more relevance to his own results.[20]

Wightman furthermore concluded that investigations with models_m indeed "confirmed" the theoretical existence of a cis and trans form of decahydronaphtalene, as postulated by Mohr. The paper nevertheless ends with a word of caution, spurred by a recent announcement that a third isomer of decahydronaphtalene had been isolated: "It has therefore seemed desirable to publish the above detailed examination of the isomeric possibilities of this substance, in view of the probability that the existence of more than two isomerides will lead to the abandonment of the Mohr postulate" (p. 1424).

In these briefly described cases, we get a sense of how molecular models_m, as analogic devices, became proximal objects of investigations, investigations which sought to explore and determine the geometrical and mechanical properties of the structure they embodied. Through these were postulated and demonstrated envelopes of theoretically possible configurations and their attendant conformations. In other words, models_m were an intrinsic part of the definition of these possible conformations, not simply a means of illustrating them.

The key question was of course whether the knowledge derived from considering the reality of models_m was true of molecular substances, or indeed of any relevance at all, and if so, to what extent? The empirical verification of these structural postulates remained for many years problematic. Evidence from test tubes was not always forthcoming, and if it was, it rarely proved unambiguous and decisive. Yet, whether they were considered viable working hypotheses or misleading speculations, there is little doubt that by the 1920s, the structural phenomena investigated, experienced, and demonstrated through molecular models_m were becoming an integral part of chemistry’s conceptual landscape.

The fact that the conformations predicted by models_m could actually be considered in agreement with experimental results was sometimes the object of surprise. In 1922, the Dutch chemist H.G. Derx (1922) published an investigation of the conformations of cyclopentane, -hexane, and -heptane based on the boric acid method. In this article, he considers strain-free versions of these structures and their various possible conformations (or arrangements, as he called them) as derived from simple skeletal models_m, photographs of which can be found in his text. According to Derx, these models_m were essential in order to visualize and demonstrate what he called the phenomena of rigidity and flexibility inherent to such structures, phenomena which he considered otherwise almost impossible to describe. He concluded that the conformations derived from the models_m agreed with the experimental results, a fact which he described as "even more surprising if one takes into consideration that these models are but a crude imitation of the real relations that exist within the structure of a molecule" (p. 315, my translation).

It is important to note that the rigidities and flexibilities denoted here by Derx are not so much general properties of the structures that he considered, but rather properties of these structures as assembled from a particular type of models_m. Choices made regarding the type of models_m to be used in structural investigations, thus of the properties exhibited, could have clear influence on the way hunches and hypotheses were pursued. This comes across clearly, for example, in a paper by American chemists Carl Marvel and C.A. Glass on monosubstituted cyclononane (1938). These investigators explore the hypothesis that a monosubstituted cyclononane might exist in two optically active forms. They report that investigations with space-filling models_m of the Stuart type confirmed the possible existence of enantiomers, which prompted them to synthesize and resolve the substance. No evidence of resolution was obtained, which led to the conclusion that the structure of monosubstituted cyclononane was less stable than what the model had let on. This negative result is turned into a lesson on how to interpret specific aspects of space-filling models_m:

3. Surveying structures: Models_m in early conformational analysis

Conformational analysis, by introducing structural considerations into areas where they had been hitherto absent,[26] made molecular models_m much more relevant to most organic chemists than they had been till then. Now much could be understood and explained by the simple inspection of carefully assembled models_m. Suggesting that polar (axial) bonds in cyclohexane are more hindered than equatorial bonds, Barton would, for example, indicate that "an inspection of models makes this reasonable for a polar bond is always close in space to two other polar bonds each attached to the next but one carbon atom, whereas there is no similar relationship for equatorial bonds" (Barton 1950, p. 318). It is quite likely that a common use of models_m was for such evaluation of the relative spatial proximity of groups or atoms in attempts to understand, explain, or predict specific reactions.[27]

Beyond the qualitative inspection of structures and conformations, models_m were also used to derive structural variables important to both theoretical work and the interpretation of physical measurements. Thus, when he established the preferred conformations of decalin and cyclohexane by calculating the interaction energy of non-bonded atoms in various conformations, Barton used equations which required that the distance between the nuclei of every possible non-bonded atom pair be known (Barton 1948). In order to obtain these distances, Barton simply made the measurements with the help of a ruler on skeletal models_m he had specially designed for that purpose (Barton, personal communication). Barton did not discuss these particular models_m in his early papers on conformational analysis. He did make extensive use of them in talks and lectures, which led to exposure and numerous requests for information (Barton, personal communication). There are indications that, through informal channels, they had started to be adopted in laboratories for tasks such as the study of conformations (Braude and Sondheimer 1955) and the calculation of dipole moments (Nace and Turner 1953). In 1956, Barton published a description of the models_m, with details about their construction (Barton 1956). In this publication, Barton discussed the use of the models_m briefly, stating simply that "the accuracy of manufacture and scale of these models is such that quite satisfactory measurements with a metre rule can be made of the distance between atomic centres" (p. 1137).

Among the explicitly reported uses[28] of the Barton models_m in the next decade there are, as mentioned above, the calculation of dipole moments (Nace & Turner 1960), the calculation of interaction energies (Barton et al. 1960), the calculation of diamagnetic shielding (Lenz & Heeschen 1961), and the calculation of coupling constants (Cross 1964; Lemieux 1961). Most reported uses revolved around measurement of structural variables such as interatomic distances and dihedral angles between neighboring molecular groups. In a number of papers, models_m are simply used to obtain dihedral angle values which are then compared to nuclear magnetic resonance measurements performed on molecular substances (Coxon & Hall 1964; Hall 1964; Perlin 1964). In other words, models_m were not in those circumstances simply a conventional means for chemists to represent the molecular objects_c ‘behind’ the instrumental inscriptions, but part and parcel of the process of giving meaning to these inscriptions in the first place, part of the possibility of interpreting these inscriptions as signifiers for these molecular objects_c.

Measurements performed on models_m were not always considered a simple affair and were done (and reported) with the care one would expect in the case of empirical manipulation, to avoid possible errors. Thus, Cross reports the following method for measuring the distance (r) between two specific atoms in 26 different steroid structures:

Figure 3. One of the device designed by J.W. ApSimon and his colleagues to measure angle and distance between two groups in the model_m of a molecular structure. Reproduced courtesy of J.W. ApSimon.

4. Conclusions

First, historically, the investigations of specific molecular structures through models_m has been a mean of revealing the mechanical properties of these structures as physical objects in their own right, properties which emerged out of explicit and implicit choices made in the course of their production. These properties could then be ascribed, tentatively, to the molecules these material structures stood for, in an analogical abduction process which sometimes proved not only successful but also enlightening (while it could, of course, sometimes prove misleading). It seems clear, for example, that models_m did not become post hoc a clever and convenient means of illustrating preexisting concepts such as the interconversion in cyclohexane, concepts which would have been derived simply from cogitation on mental representation.[34] In other words, the production of concepts can involve at its core not only the ‘handling’ of mental representation, but also the handling of material representations. If the word dematerialization has any use here, it does not so much describe a formal property of chemical analytical processes, but rather this tendency, rather common in the formal chemical literature, to treat work with models_m as if it never happened.[35]

Second, we also have seen how models_m have been used as devices to ‘survey’ molecular structures, to gain qualitative and quantitative knowledge about the spatial arrangements of atoms and groups in these structures. This point brings up two perspectives on models. On the one hand, the model can be seen as a ‘map’, or a quasi-inscription, the result of a long chain of re-representation that extends much further back than the local site of use,[36] an object that shows relative gain in universality, standardization and capacity to circulate. Yet, they worked precisely because in many ways they were exactly the inverse of what is explicitly defined as inscription. On the other hand, there is the model as a quasi-specimen, raw ‘matière à investigation’. In laboratory settings, as the cases I have discussed thus far show, models_m are not simply observed. They are submitted to various manipulations, assembled, probed, and measured. In other words, the qualitative and quantitative properties of the models_c are not always a given on which scientific investigations could rest. They often need to be worked out and sorted out in the process of performing these investigations, and here, as we saw, models_m sometimes proved useful.

Simply attending to the materiality of models_m in the course of their deployment in chemical research, as I have done here, certainly does not constitute a complete account of the many interwoven practical and epistemological problems at play in molecular modeling_mcas a research practice, nor does it provide more than a glimpse into the development of chemistry’s increasingly complex understanding of molecular structures. Yet, it does compel us to consider that such devices, at the fulcrum of the articulation of theories and concepts, are also what chemistry has been about.
 
 

Acknowledgments

Notes

Bibliography