Density functional theory (DFT) calculations have become a ubiquitous feature of journal articles in all of chemistry’s subdisciplines, used by researchers to electronically characterize molecules or reactions they’re studying.
The spread of user-friendly quantum chemistry software has aided the democratization of DFT by allowing nonexperts to do computational analyses that might once have required a theoretical chemist’s expertise. And today there’s another option: private companies that will perform DFT calculations for chemists.
It seems clear that DFT’s transformation into a broadly accessible tool has helped advance chemistry. But its evolution also raises questions: Have DFT calculations matured to the point that they have become a technical exercise—like obtaining a nuclear magnetic resonance spectrum—rather than an intellectual one? And if you hire a company to run calculations for you, how should you disclose its involvement, especially in the interest of reproducibility? The answers are complex, but a picture of this dynamic area of science emerges through interviews with DFT experts, computational chemistry companies, and the chemists who partner with them.
Chemists use DFT to calculate the approximate electronic structures of molecules or materials and thus predict their properties. The exact electronic structure of any system can be found by solving the Schrödinger equation, which mathematically describes a system’s quantum mechanical wave function. But unfortunately, solving the Schrödinger equation fully for systems with more electrons than a hydrogen atom is computationally impossible.
DFT gets around that problem by making assumptions about the system that simplify calculations and bring them back into the realm of computational possibility. Used correctly, DFT calculations can accurately predict properties in a reasonable amount of time—say, in 1 day rather than 1 week. And theoretical chemists continue to improve DFT computational tools, making calculations more accurate and applicable to a wider range of situations.
Research journals help reveal DFT’s growing role in chemistry. In 1998, 36 papers in the Journal of Organic Chemistry included the phrase “density functional theory.” In 2018, 183 did. Demand for DFT calculations now supports a small army of computational chemists who collaborate with experimentalists to help them predict or understand laboratory results. The ones who spoke with C&EN don’t have a problem with DFT’s evolution from intellectual to technical.
“Merely running a DFT calculation is a technical job,” Filipp Furche says. He develops new electronic structure methods, including DFT methods, at the University of California, Irvine, and says he’s fine with DFT calculations being treated the same as NMR spectra collection.
But he also says that there’s more to using DFT calculations than simply performing them, and that’s where expertise can make a big difference. The first issue is knowing how to ask the right questions when setting up calculations and asking them in the right way, Furche says. Asking DFT to predict something like atomic charge might produce an unhelpful or confusing answer. While chemists understand that concept in terms of integer values, quantum chemistry doesn’t give a unique answer, so the DFT result depends on how you ask the question. A computational chemist is also familiar with different DFT functionals and methods and during calculation planning can devise a strategy—which might include techniques other than DFT—to maximize accuracy and minimize the computational time, Furche says. He considers these to be intellectual, not technical, contributions.
His UCI colleague Kieron Burke agrees that DFT calculations can be treated like NMR measurements in some cases, pointing out that their university has facilities that provide both services to researchers. But, echoing Furche’s caveats, he also says the analogy is imperfect. With NMR, a chemist is usually asking a technician to produce only a spectrum or spectra for a given sample. Getting useful and accurate answers from DFT calculations, Burke says, requires an iterative approach that starts with basic methods and leads to more complex ones. Doing it right means having PhD-level training in quantum chemistry and insight into the science in question, he says.
Understanding DFT results is not trivial, either, Furche notes. Without the right combination of experience and insight, nonexperts might dismiss results that are actually useful, or they might read too much into results because they don’t understand the limitations of DFT calculations or the methods used.
For experimental chemists who want some computational insight—or think it will improve their chances of publishing, a motivation computational chemists agree exists—the solution is a collaborator. But that doesn’t have to mean an academic researcher. Almost since computational chemistry began, there have been companies that scientists can pay to perform DFT calculations.
These companies provide a range of services to a range of clients. Computational chemistry giants like Schrödinger sell software that chemists can use to do DFT calculations. They also offer consulting services, meaning computational chemists who can help clients use the software effectively—although the cost of those services likely puts them out of reach for most academic chemists.
ChemAlive is a newer entrant into the quantum chemistry software market, and cofounder Peter Jarowski says he wants to automate quantum chemistry methods like DFT so any chemist can do it. He says his firm has worked with oil and gas companies, paint makers, process chemists, and others to perform calculations. Jarowski thinks DFT is most useful for predicting chemical trends, including compounds’ reactivities and colors. “Chemists want to know if they should go left or right” when they’re searching for new molecules or materials, he says, and DFT calculations are good at answering that question.
Jarowski is clear that he is not a technician. When he used DFT to study phase changes in metallogels with researchers from the Aramco Research Center in Houston, he “didn’t just take some SMILES strings and do DFT,” he says, referring to strings of standardized text that describe the line structure of molecules. Jarowski says he was part of designing and conducting the study.
He is a coauthor on a paper describing that work (J. Am. Chem. Soc. 2017, DOI: 10.1021/jacs.7b07053). In the past, authorship has been the norm for computational chemists, whether they are academics or in the private sector, to acknowledge their contributions to a collaboration with experimentalists. Some computational chemistry companies may be changing that standard, however. And how their involvement in research is presented in journals might be a useful proxy for understanding DFT’s evolving stature in chemistry.
For instance, a number of chemistry research papers in journals including the Journal of the American Chemical Society (2019, DOI: 10.1021/jacs.9b05351), Electrochimica Acta (2019, DOI: 10.1016/j.electacta.2019.06.061), and Catalysis Science and Technology (2018, DOI: 10.1039/c8cy02248h) acknowledge a company called Shenzhen Huasuan Technology for performing DFT calculations or providing theoretical chemistry support. An advertisement for the company lists several other journals that it says its clients’ work has appeared in, including Angewandte Chemie International Edition, although a Google Scholar search did not produce any mention of the company in that journal. Jingshan S. Du, a PhD student at Northwestern University, brought the Shenzhen Huasuan Technology ad and several papers the company was involved in to C&EN’s attention.
Shenzhen Huasuan Technology did not respond to multiple requests from C&EN for an interview about its process for working with chemists. Linxi Hou of Fuzhou University is a coauthor on a paper about tumor imaging that acknowledges the company for DFT calculations of electronic states (ACS Appl. Mater. Interfaces 2019, DOI: 10.1021/acsami.9b01205). Hou describes a collaborative relationship, including the sort of iterative work that Burke outlined. Hou says his group provided the company original experimental data and his group’s requirements for the calculations, and the group then worked with Shenzhen Huasuan Technology to interpret the results.
SciCalQ is another computational chemistry company that offers DFT calculations as a service. Chief Technology Officer Jianming Chen says SciCalQ works mostly with experimentalists in industry or academia. In a typical scenario, Chen says, clients would provide SciCalQ with data from synthesis or characterization experiments that they can’t explain. SciCalQ would meet with them and use DFT calculations or another computational technique to help understand the mechanism underlying those data.
“We hope DFT service for chemists can be as convenient as scanning tunneling microscopy,” Chen says, referring to another molecular analysis technique routinely done by technicians.
Chen says SciCalQ has worked with scientists on research that has been reported in journals including Chemical Communications, Angewandte Chemie International Edition, and Organic Letters, although he says the company’s agreements with those chemists prevent him from identifying specific papers. C&EN was unable to find SciCalQ mentioned in those or other journals, though Chen notes that in some cases, SciCalQ’s involvement may be minimal, such as consulting with chemists doing their own computational work. Starting at the beginning of 2019, Chen says, SciCalQ has generally required its customers to mention their consultation with the company in the acknowledgment section of research papers.
Contracting out routine chemistry work is nothing new, says Christopher J. Cramer, a theoretical chemist and vice president of research at the University of Minnesota Twin Cities. He says synthesis, characterization, and modeling have all been outsourced to contractors. He thinks what has changed is there is now enough demand for routine computational contract work to sustain companies offering DFT as a service.
In Chen’s opinion, chemists may find it more convenient to work with SciCalQ rather than finding a more traditional collaborator or coauthor. He thinks letting SciCalQ do the work can save time and money, letting chemists focus on their research questions. He also sees advantages for researchers who use a contractor rather than a collaborator in avoiding arguments about the order of authors on a paper or about intellectual property. SciCalQ’s contracts stipulate that all intellectual property rights belong to its clients, according to Chen.
Acknowledgments in research papers are a way to recognize contributions that don’t qualify as intellectual or otherwise meet the standards for authorship. “When such contract services do not include any particular intellectual contribution, it seems reasonable to provide only an acknowledgment,” Cramer says. The American Chemical Society, which publishes C&EN, defines authors as those who have made a significant contribution to the reported research and share responsibility for the results. Other publishers, like the Royal Society of Chemistry and John Wiley & Sons, offer similar guidelines.
But if DFT results are viewed as a technical contribution, not an intellectual one, they may not be spelled out in a paper, which can be a problem for other chemists trying to better understand or replicate the work. That appears to be the case in at least some research papers that acknowledge Shenzhen Huasuan Technology’s contributions.
Axel D. Becke of Dalhousie University, who was one of the first to apply DFT to chemistry questions, calls the computational section of one paper that acknowledges Shenzhen Huasuan Technology’s contributions totally unacceptable (J. Am. Chem. Soc. 2019, DOI: 10.1021/jacs.9b05351). “They simply refer to DFT calculations. There is no mention of functionals, basis sets, or computer codes,” Becke says. “So this is nonreproducible nonsense.” Becke says that although other papers connected to the company do a better job by naming the algorithms and software used and describing the method, their descriptions could include more detail (Electrochim. Acta2019, DOI: 10.1016/j.electacta.2019.06.061 and ACS Appl. Mater. Interfaces 2019, DOI: 10.1021/acsami.9b01205).
The editor in chief of ACS Applied Materials and Interfaces, Kirk S. Schanze of the University of Texas at San Antonio, says the journal does not weigh in on questions of authorship, a policy shared by most chemistry journals. After C&EN contacted Schanze about the Shenzhen Huasuan Technology paper, he discussed it with the journal’s editors during their most recent monthly teleconference. He says that they agreed to look out for papers like it that acknowledge technical services and that in the future, editors might suggest authors review ACS’s guidelines and think about authorship choices in certain cases. Schanze acknowledges that the authorship of a paper could affect the clarity or reproducibility of results presented in a paper.
The questions that DFT calculations can answer, and how well, will continue to change, and the ways that researchers use DFT calculations will as well. Computational chemists, experimentalists, companies, and publishers will have to adapt as DFT grows. Some teething pain may be inevitable, but DFT’s continued march into the mainstream seems all but assured.