To the editor

In their commentary, Looking for Design in Materials Design, Eberhart and Clougherty addressed the important and challenging issue of controllable, efficient design of new materials with pre-defined properties1. However, we feel that the authors have fallen victim to some misconceptions, which lead them to direct as well as implied assumptions in their argument that are inappropriate.

They propose that starting with desired properties of a hypothetical material, one should design a chemical compound that possesses these qualities. Although this is a very appealing thought in principle, such an approach does not seem to be feasible. From a given structure of a solid, that is, atom positions and associated features of the electron distribution, one can derive its physical properties, but this step is not reversible. Just the opposite: a given property, such as ionic or electronic conductivity, Young's modulus, bandgap or magnetic susceptibility, can be exhibited by many solids with different structures and compositions.

We regard a second aspect of the authors' argument to be even more problematic. In the 'design process' as visualized by them, the criterion controlling the selection process of structures and compositions consists of some prescribed physical property, with the underlying assumption that the compound generated in the process is capable of existence. However, the possible equilibrium structures of an ensemble of atoms are determined exclusively by their free energies and the barriers surrounding them on their energy landscapes, for given thermodynamic boundary conditions. Thus, 'design' of a material with a given property turns out to be a far more involved task than the authors present it to be.

Lack of control and predictability are notorious characteristics of the synthesis of new materials, indeed. We have tried to address this distressing and intellectually unacceptable situation over the past 12 years2,3,4, and have developed a general modular approach to achieve the goal of unbiased prediction of new solid compounds. Its foundation is the representation of the whole material world — that is, the known and not-yet-known compounds — on an energy landscape, which gives information about the free energies of these compounds. From this, it follows that all chemical compounds capable of existence are present on this landscape, and the chemical synthesis always corresponds to the discovery of compounds, not their creation4.

Predicting the structure and properties of not-yet-synthesized crystalline compounds requires information about the possible stable structures in a chemical system, their thermodynamic weight, and their kinetic stability. Because such (meta)stable compounds correspond to locally ergodic regions on the energy landscape of the system surrounded by sufficiently high energetic and entropic barriers, we need to undertake a global study of the landscape. Our approach can be divided into four basic steps (see Fig. 1): Identification of locally ergodic regions, determination of their kinetic and thermodynamic stability, computation of their local free energies, and calculation of their physical properties. The first three steps establish whether the compound will be able to exist as a thermodynamically or kinetically stable structure. As a final step, we can then use standard computational methods5,6 to compute the physical properties of the predicted solid compound.

Figure 1
figure 1

Flow chart of the general approach to prediction of not-yet-synthesized materials by global exploration of energy landscapes.

Full size image

We note that once this information has been acquired, we can then attempt to select from among the stable structures the ones that possess the desired properties2. A beautiful example of such materials optimization is the fine-tuning of the bandgap in the AlxGa1−xAs alloy by Franceschetti and Zunger7 using simulated annealing to achieve an optimal arrangement of Al and Ga atoms on their joint sub-lattice.

The flow chart displayed in Fig. 1 only represents the skeleton of our approach. When dealing with real applications, a number of restrictions and refinements of the method have to be taken into account2,3, while still remaining as general as possible as far as the feasible structures are concerned. As indicated in the flow chart, one can include the desired property into the global optimization process as an additional criterion of acceptance2 after the landscape exploration phase. However, we prefer to save all structure candidates and to establish libraries of compounds capable of existence. These libraries can then serve as the basis for a detailed screening, not only for properties currently being investigated, but also for other features of the compounds that might be of interest in the future.

We would like to emphasize here that the creative action of a materials scientist consists of the selection from what nature offers — there is no freedom for 'abstract design' outside the realm of compounds capable of existence. Thus, we feel that the term 'design', although very uplifting, is not really appropriate, and one should better speak of, for example, optimal selection instead.