In response to: “Physics on Edge” (Vol. 3, No. 2).
To the editors:
In his essay, George Ellis has once again gone public with his arguments that a set of ideas currently being considered by cosmologists concerning the notion of a multiverse is not scientific. His main criticism is that he does not think multiverse theories are predictive, as scientific theories should be. Were this really true, it would indeed be a problem. But Ellis’s claim relies on several basic misconceptions about science, and is thus easily refuted. He does not appreciate the difference between a framework and a theory, he applies a poorly-defined notion of direct evidence, and he mistakes the process of science for the output.
Let us begin with frameworks and theories. Every beginning physicist learns both classical mechanics and electromagnetism. These subjects have a quite different character; in classical mechanics we are told that force equals mass times acceleration, but this equation is actually useless until we are told what force is. And there are all sorts of different forces we might consider: gravity, electromagnetic forces, friction, tension, and so on. Any particular set of forces may govern the motion of a particular collection of particles, but classical mechanics itself does not prefer any specific choice. It is a framework within which we can realize many different theories. By contrast, there are a small number of concrete equations we study in electromagnetism—Maxwell’s Equations and the Lorentz force law—and then everything else follows from these few equations. This is a theory. It quantitatively describes many aspects of our world in an unambiguous way. Either it is correct, within its regime of validity, or it is not. Ellis’s first criticism of the multiverse is that there are a variety of distinct possibilities being considered, and thus that it cannot make specific predictions. But this is like saying that we should throw out classical mechanics because there are all sorts of forces which could exist. Sure there are, but in any particular theory, such as electromagnetism or gravity, there will be a specific set of rules to test.
Ellis’s second criticism of the multiverse framework is that typically most of the multiverse is outside our causal horizon, and thus we cannot receive direct signals from it. The multiverse, he argues, is for this reason not directly testable. But upon further scrutiny, this notion of direct testability is rather naive. Consider the existence of quarks. Have you ever seen one? I certainly haven’t. We infer the existence of quarks indirectly based on what happens when we collide other particles such as protons and electrons. My eyes only detect photons, and even those are conveyed into my brain indirectly through nerve signals. What matters in testing theories is not whether the evidence is direct or indirect. What matters is that the theory predicts more observations than it has parameters. And the hope of most people working on the multiverse framework is that any particular multiverse theory will indeed make detailed testable predictions. If, say, for a multiverse theory with zero parameters I could compute with high accuracy the expected values of all of the nineteenth parameters in the standard model of particle physics, and the six to seven parameters of the ?CDM model of cosmology. If all were consistent with the latest measurements I suspect that even Ellis would be forced to give strong credence to this theory. We might not be able to directly observe the other regions in that multiverse, but we would still have strong experimental confirmation of the theory. We would never be sure it was correct, but this is also true for all of our other scientific theories. It could be the case that things work completely differently than we think. Our scientific skepticism will never allow us to rule this out entirely. We will never be able to test all the predictions of any theory. It is enough that we can test some of them.
We do not have such a predictive multiverse theory today and thus far the multiverse has mostly remained a framework. Despite some promising initial calculations, it is quite difficult to construct an actual theory, never mind test it against the data in the way I just discussed. The strongly-constrained structure of string theory suggests that it might lead to a unique multiverse theory, but formulating this theory with the precision of Maxwell’s electromagnetic theory will require us to solve some of the deepest problems in quantum gravity. This process may or may not succeed. We will not know until we try. This situation gives rise to Ellis’s third criticism of the multiverse framework, citing the philosopher Richard Dawid: people who work on it sometimes give non-empirical reasons for doing so. Of course they do! The motivations for considering the multiverse are empirical: the observed value of the cosmological constant, the observed spectrum of density perturbations in the early universe as reflected in the cosmic microwave background, the distribution of galaxies, and the observed existence of gravity. The space of potential theories is enormous. If we were not allowed to work on candidate theories without knowing they were true in advance, how would we ever develop any theories at all? I can think of few things that would be more destructive to science than this notion, were it to be adopted. The process of theoretical science is what happens prior to the theory being formulated well enough to test it against experiment. The non-empirical methods described by Dawid have been used throughout the history of science to help decide which theories to work on and how to refine them. Names like Newton, Maxwell, and Einstein spring to mind immediately, as do many others.
We do not yet know if a successful multiverse theory exists. But it is clear that the multiverse framework is well within the bounds of conventional science. It may eventually go the way of so many other promising ideas that did not pan out, but this would be for the usual reasons that scientific frameworks are discarded and not as a result of some philosophical infraction.
George Ellis replies:
“Ellis,” Daniel Harlow writes, “does not appreciate the difference between a framework and a theory, he applies a poorly-defined notion of direct evidence, and he mistakes the process of science for the output.”
As to frameworks and theories, the inflationary universe is a framework, with over a hundred theories developed within that framework. The multiverse proposals are not a frameworks in the same sense. They are a mixed bag of ideas, ranging from the extension of the universe beyond the visible horizon to chaotic inflation to conflations of the Everett wavefunction with chaotic inflation to proposals that the universe is a simulation.1
As to direct evidence, Harlow says:
But upon further scrutiny, this notion of direct testability is rather naive. Consider the existence of quarks. Have you ever seen one? I certainly haven’t. We infer the existence of quarks indirectly based on what happens when we collide other particles such as protons and electrons.
There are many different and repeatable experiments of various kinds avaialble to indirectly test the idea that such particles exist. We are, however, not talking about the existence of types of particles or forces, but about existence of domains in the universe, and about specific configurations of geometry and matter. We are dealing with geography, which is not an experimental science. It is an observational science. The issue is not testing the generic nature of what exists, which is what Harlow refers to, but the particular nature of the specific entities that exist. Repeatable tests can be used to affirm that electrons exist, but not that a specific multiverse bubble with any particular properties lies a hundred Hubble radii away in a particular direction. This is completely different from talking about generic properties of matter. We are not dealing with experimental science numerous repeatable experiments on many different samples of the same entity in identical circumstances can be undertaken. That is the key difference. Harlow’s example does not apply.
Like Carol Rovelli, I agree that one always uses non-empirical reasons for choosing what to do with regard to process versus output. I never said otherwise. And I do not say that one should not work on multiverses. “The process of theoretical science,” Harlow writes, “is what happens prior to the theory being formulated well enough to test it against experiment.” Indeed. In the case of the multiverse there are good reasons to believe that most models arising from the framework will never be testable by any kind of observation or experiment that can demonstrate the claimed space-time regions do indeed exist with the properties supposed. That is the issue. The process of science—exploring cosmology options, including the possible existence or not of a multiverse—is indeed what should happen. The scientific result is that there is no unique observable output predicted in multiverse proposals. This is because, as is often stated by proponents, anything that can happen does happen in most multiverses.2 Having reached this point, one has to step back and consider the scientific status of claims for their existence. The process of science must include this evaluation as well.
Daniel Harlow is Assistant Professor of Physics at MIT.
George Ellis is Emeritus Distinguished Professor of Complex Systems in the Department of Mathematics and Applied Mathematics at the University of Cape Town in South Africa.