Computer Science Curricula Must Reintroduce Physics Now

After a couple of years of performing research in quantum computing at the National Center for Supercomputing Applications and the Illinois Quantum Information Science and Technology Center , and teaching computer science courses at two US institutions, it has become apparent that various changes are needed to fully insert our discipline into the quantum revolution more broadly. These changes go well beyond inserting Quantum Information Science courses or setting up minors: a major overhaul of CS is needed to satisfy increasing workforce development needs in a world where academia and industry are redefining what computation as a whole means.

Useful metaphors. For a while, computer science has been at its core an exercise in abstraction. From computer organization to data structures and algorithms, we hide the physics of devices beneath well-behaved entities that we can reason about formally, and use prescriptively to dictate how a computing platform should operate given a set of instructions. Even the notion of what an instruction is happens to be a useful metaphor that modularizes complexity. Both modularity and indirection have served the role of universal mechanisms to make intellectually tractable what would otherwise be hard to articulate as information processing.

Multiple events have supported this trend toward higher abstraction disconnected from the physical basis. In part, the quality of silicon manufacturing technologies for semiconductors has reached the staggering figure for single bit error rates of 10^-24, which means that only data centers and supercomputers need to be concerned with it. The physics behind classical computers (mostly classical electrodynamics) can be reduced to a series of principles and components which are amenable to standardization. This is why CS curricula teach about digital design in computer organization and architecture without the apparent need to worry about the solid state physics behind a bipolar junction transistor or how a microprocessor oscillator works, and why our abstract machines do not make reference to circuit elements. Our world of classical information processing, despite the progress and challenges it has brought forth in the world, is covered by a small set of relatively simple governing laws.

Metaphors break. Quantum information science changes the governing laws behind computing to those spanned by quantum mechanics. The more we dig into quantum computing, the more we discover how many of our abstractions have to be pried open just to understand the way they differ from classical counterparts. What we call algorithms and circuits still refers to mostly analog systems we cajole into behaving somewhat nicely in actual hardware. Various flavors of quantum abstract machines do make reference to state vectors and unitary operators, unthinkable in the classical world –and possibly unproductive . We have not yet moved away from describing programs with circuits, despite the fact that they are not ideal to be "read by humans and only incidentally for computers to execute" (Abelson, SICP 1996) ; while efforts are under way from some of us by getting the community together and creating new programming languages from old CS roots , finding better abstractions seems like a genuinely hard problem to solve. Paraphrasing Feynman , it is an important question "and by golly it's a wonderful problem, because it doesn't look so easy".

This all seems to mean at first sight that, for a while, we need to live in a dual world of classical and quantum information and learn to live with the discomfort until the dust settles, quantum systems become fault-tolerant, and we reconcile both. Or, do we?

New metaphors. I wish to suggest here that there is a better alternative for computer science to insert itself productively into this quantum computing revolution. We need to bring back physics into CS curricula in strategic ways as we navigate the dissonance between classical and quantum information. Recognizing that computing devices are first and foremost physical devices goes a long way in understanding which principles survive, which ones need revisions, and which ones need to be stated anew.

From the point of view of workforce development, it seems increasingly clear that attracting people to quantum computing depends on a number of limiting rates, one of which is the acquisition of principles from both computing and quantum physics. Although anecdotal since no formal study on the matter has been carried out, my own experience and that of my colleagues in physics and computing suggests that it is easier to obtain computer science foundations for people who hold a degree in physics than the other way around. It is more intuitive to distill computation once understanding of physical principles is present, than to buttress oneself with enough physics starting from CS.

CS needs Physics. The fact that the universe admits another form of computation besides that derived from classical mechanics is an enthralling fact in itself, and one that calls for a new unification of what we mean by computing. Unifications are what makes science progress once facts differ enough, and these appear when motivated individuals have acquired enough understanding and language to converse confidently in both worlds. We are not there yet in computer science, and will not get there by continuing to ignore the physical basis behind information processing.

This apparent predicament can be interpreted as an opportunity to rethink computer science education and how it connects to a substantially more complex world. The physics of harmonic oscillators is used to both mark time in classical CPUs and provide the basis for QPUs; there are several instances where the same principle, with the same mathematical form, applies in different ways. Having those physical intuitions at hand may also provide insights into why, for instance, continuous variable optimization problems are in P while ones with discrete variables are NP-HARD, and why quantum resources do become advantageous for the latter case.

Physics needs CS. Building quantum computers, regardless of their underlying principles, cannot escape the consequences of computability and good hardware design. We have new quantum resources (superposition, interference, entanglement and possibly coherence), yet computational complexity does not still account for them in the same way that we do for space and time; this is a sign of the need to involve more fully CS theorists, which requires understanding of the mathematics behind quantum objects at depth.

Good hardware design means separation of concerns via collectively agreed modularity in a way that allows advancing architectures along multiple fronts simultaneously without stepping on each other's toes; the current quantum hardware design space is now opening to more integrated concerns, yet there seems to be some degree of reinvention of principles that have been well known in CS for a long time. CS can bring decades of tools and methods that allow reasoning about systems to then proactively suggest changes at the hardware-software interface. Yet, to engage fully and responsibly, we need CS expertise that includes from the start the ability to reason about physical systems for computation.

Conclusion. Rethinking QIS education, and more broadly CS education in the context of QIS is an imminent and rewarding challenge. The National Quantum Initiative Act and its recent reauthorization have both highlighted the significance and urgency of workforce development and training: doing so must be done in a scalable manner, which for CS curricula implies providing enough physics language and intuitions as early as possible so that transferable skills are present for those interested in QIS technologies. This is a global challenge that needs to be addressed broadly, soon, and in earnest.