Physics and the Human Face of Causation Mathias

Physics and the Human Face of Causation Mathias Frisch University of Maryland currently: Ludwig-Maximilians-Universität München 1

Causal Skepticism in Physics • An extremely popular view, held e. g. by Bas van Fraassen, Jim Woodward, Hartry Field, Chris Hitchcock, Huw Price, Barry Loewer and many, many others: • Causal notions play a legitimate role in the special sciences and in common sense reasoning but NOT in physics because causal notions have a ‘human face’. 2

Putative contrasts • • • • Barry Loewer: “The main relevant differences between fundamental dynamical laws and special science laws are these: the candidates for fundamental dynamical laws are (i) global, (ii) temporally symmetric, (iii) exceptionless, (iv) fundamental (not further implemented) and (v) make no reference to causation. In contrast, typical special science laws are (i*) local, (ii*) temporally asymmetric, (iii*) multiply realized and implemented, (iv*) ceteris paribus, and (v*) often specify causal relations and mechanisms. ” 3

The form of the skeptical argument (i)There is a certain (pragmatic/userdependent/perspectival) feature x that is characteristic of causal models. (ii)Representations in fundamental physics in general do not (or even could not) have feature x. (iii)Therefore, causal notions can play no role in physics. 4

• Overview Examine putatively ‘human-faced’ features of causal representations: i. coarse-graining ii. distinction causes vs. background conditions iii. asymmetry iv. partiality v. (applicability of notion of manipulation and intervention) • • case study: modeling the Large Hadron Collider at CERN Conclusions: (i) If causation has a human face, then so does representation in physics; (ii) Physics, like other sciences, is deeply permeated by causal reasoning. 5

Fundamental physics = ‘established’ physical theories • • • quantum theories, space-time theories, classical mechanics, classical electrodynamics. . . NOT only a putative fundamental final theory of quantum gravity. 6

Coarse-graining • Jim Woodward: “the variables of upper level causal theories are extremely coarse-grained from the point of view of fundamental physics”: • it is an essential feature of causal representations that they are coarse-grained; • theories of fundamental physics are not coarse-grained. • therefore there’s no place for causal representations in physics. 7

Coarse-graining Different senses of ‘coarse-graining’. • • • The relevant sense here: Inter-theoretic relations “coarse-grained variables may fail to completely partition the full possibility space from the point of view of an underlying finegrained micro theory. ” (Woodward; see also Field, Loewer) true for any but the most fundamental theory of physics (Quantum-Gravity? ) 8

Background conditions • causal models permit and even require a distinction between causes and background conditions against which we select causes. • • see Horia Tarnavu’s presentation. models in fundamental physics do not permit such a distinction: • In physics we represent phenomena through complete models, constructed from the data on complete initial-value-surfaces (e. g. lightcone cross sections) as input. 9

Background conditions: completeness t Initial value surface space 10

Background conditions: completeness t Complete crosssection of backward lightcone space 11

Modeling in fundamental physics • Initial value-surface at 1 s before time of interest: Need complete data in spatial region with 300 000 km radius! • yet the dynamical equations require full initial data as input. • One solution: • Modeling decision what to include on that surface. • similar to construction of causal models. 12

The LHC at CERN: Coarse-graining and background conditions in action • Detector events and proton beams are not modeled in terms of complete QFT lightcone model: • • • proton beams modeled classically beams are modeled as being influenced only by fields in bending and focusing magnets. external influences on magnet modeled in coarse-grained way. 13

• • Modeling in fundamental physics Recall Loewer: models are “(i*) local [+background conditions] (iii*) multiply realized and implemented, (iv*) ceteris paribus, (ii*) temporally asymmetric, (v*) often specify causal relations and mechanisms. ” 14

Temporal Asymmetry • Time asymmetric causation would have to be something “over and above physics”--it would amount to a “hyperrealism” about causation--which “threatens to make it both epistemologically inaccessible and practically irrelevant. ” • Huw Price and Brad Weslake 15

Partial evidence and causal representations • Implicit assumption: • All inferences/explanations in fundamental physics can be exhaustively characterized in terms of the ‘dynamicallaws-plus-particular-initial-and-boundaryconditions’-model. • But: Causal representations allows us to draw inferences from incomplete information, where dynamical laws alone are not applicable. • Limits on the available evidence result in an underdetermination problem that can be solved with the help of causal assumptions. 16

Example: inference to the existence of a star as common cause of our localized observations. final value surface Oi S

Partial evidence and causal representations • infer common cause from correlated localized events. • similar to an ‘existential abductive hypothesis’ in medicine? see Aliseda • Also need to assume that any source-free incoming fields from distant regions are independent. • i. e. exogenous variables are independent. • Enables us to bring formal apparatus of causal structural models (e. g. Pearl) to bear. • • Causal Markov Theorem: If M is an acyclic, deterministic functional model with independent exogenous variables, then the causal graph that corresponds to M satisfies the Markov condition. 18

Time-asymmetric causal representations and inferences at the LHC: the accelerator Control systems (Steinhagen, “LHC Beam Stability and Feedback control”): “The signal flow and causality are indicated through arrows. ” (Steinhagen) Feedback systems ensure robustness of the beam-system through layers of signals, feedbacks, and regulation (see Sandy Mitchell’s talk; also Mike Joffe: feedbakc mechanisms across many disciplines. ) Use of ‘causal’ delay terms: introduces asymmetry in mathematical representation: “It is obvious that causality forbids the inversion of the exponential delay term. ” (Steinhagen) see linear response theory: time-asymmetric causal assumption invoked to restrict solution space of a time-symmetric equation. 19

• • Modeling in fundamental physics Recall Loewer: models are “(i*) local [+background conditions] (ii*) temporally asymmetric, (iii*) multiply realized and implemented, (iv*) ceteris paribus, and (v*) often specify causal relations and mechanisms. ” 20

Objections • You’ve only shown that models in applied physics also have a ‘human face’. • But truly fundamental physics does not merely provide partial and coarsegrained representations and thus causal notions do not apply there. • At the very least, a Laplacian demon would have no need for causal notions. 21

1. Applied vs. Fundamental Models • Distinguish: • the possible worlds allowed by a theory (represented by model-theoretic models of a theory’s laws) (‘theoretical models’) • ‘models constructed with the help of a theory’ (‘representational models’) • We use theories to represent the world. What a theory tells us about the world is given by the structures we actually can and do use to represent phenomena. • what is empirically confirmed is the success of those models actually used to represent phenomena. 22

1. Applied vs. Fundamental Models • Confirmation of causal dynamical model, not merely dynamical model: Causal structures (e. g. DAGs) Dynamical equations causal dynamical model confirms data model 23

3. Partiality and causal representations • Does that mean a Laplacian demon could do without causal notions? • Perhaps, but if the demon is given the complete ‘Humean mosaic’, he could equally well do without laws! t initial value surface space The Humean Mosaic 24

• Pat Suppes: Conclusion “What we are able to get a grip on is a variety of heterogeneous, partial relationships. In the rough and ready sense of ordinary experience, these partial relationships often express causal relations, and it is only natural to talk about causes in very much the same way that we do in ordinary experience. ” • Hitchcock’s reading of this: causal talk is preliminary and “(P*) there advanced stages in the study of certain phenomena when it becomes appropriate to eliminate causal talk in favor of mathematical relationships (or other more precise characterizations). ” • My argument here: The ‘Human Face’ of Causation is the face of scientific representation more generally: • • • there is nothing preliminary about the way LHC is modelled. scientific representations are partial, coarse-grained, involve distinction between salient factors and background conditions and often appeal to time-asymmetric constraints. partiality of representations, even in physics, makes causal representations especially useful. • Causal structures represent things just as objectively real or unreal as (other) aspects of models in physics. 25