The British Society for the Philosophy of Science

How Is Quantum Field Theory Possible? by S. Y. Auyang

Review by: Michael Redhead
The British Journal for the Philosophy of Science, Vol. 49, No. 3 (Sep., 1998), pp. 499-507
Published by: Oxford University Press on behalf of The British Society for the Philosophy of
Science
Stable URL: http://www.jstor.org/stable/688086 .
Accessed: 25/06/2014 07:04

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Oxford University Press and The British Society for the Philosophy of Science are collaborating with JSTOR to
digitize, preserve and extend access to The British Journal for the Philosophy of Science.

http://www.jstor.org

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions
 Brit. J. Phil. Sci. 49 (1998), 499-507

REVIEW

S. Y. AUYANG
How is QuantumField TheoryPossible?
Oxford, OxfordUniversity Press, 1995, paper?21.95

Michael Redhead
Centrefor Philosophy of Natural and Social Science,
LondonSchool of Economics and Political Science

This book provides a sustainedattemptto give an interpretationof quantum

field theory along broadly Kantianlines. Kant is sometimes representedas
trying to bolster Euclidean geometry and Newtonian physics as necessary to
articulating intelligible phenomenal experience, but Auyang attempts to
abstracta categorical frameworkat a level of generality which will accom-
modate everyday experience, classical physics, quantumphysics, and ulti-
mately quantumfield theory.I am sure this is the right way to approachKant,
but Auyang does not attemptto apply the dense jungle of Kantiandistinctions
and terminology, or to reproduce a genuinely transcendentalargument to
justify the synthetic a priori as Kant himself does. RatherAuyang's project
is in the traditionof Strawsoniandescriptivemetaphysics,but attemptingat a
deep level of abstractionto capturethe general concepts of object, property,
quantity,and relation that she sees as underpinningall intelligible discourse
about the physical world. This is the Kantianturn.
Let us startwith the generalconcept of object. ForAuyangthis is at a higher
level of abstractionthanthe substantialconcept of 'thing', but nevertheless,in
accordancewith her programme,an everydaything such as a table falls under
the general concept of object, which incorporatesfor Auyang two essential
features.Firstly there is the variety of representationscorrespondingcrudely,
for example, to viewing the table from different directions or perspectives
which are linked by definite transformationrules, and secondly there is the
crucial ingredientof binding togetheror transcendingthese representationsin
a way that objectifies the experiences as being of something,the object, over
and above the phenomenal representations.Auyang argues that the general
concept of object applies equally in quantummechanics. Quantumobjects
display a variety of properties corresponding to differing measurement
contexts. The objective feature is the state vector which can be represented
by expanding in terms of various complete sets of eigenstates of different

? OxfordUniversityPress 1998

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions
500 Review

possible 'observables'(in the simple case of a discrete spectrumof eigenva-

lues). The transformationtheory of quantummechanics shows how these
different representationsare related by unitary transformations.Again we
have the basic structureof representationslinked by appropriatetransforma-
tions and the objective state vector uniting and transcendingand hence, says
Auyang, objectifyingthe state of the quantumobjects.
In full'generalitywe have a state space M which encompassesall possible
states of an object. What Auyang calls 'representativerules' f, and fg assign
definite predicates in the representationsf,(M) and fg(M) for the objective
state x. As Auyang puts it (p. 96), 'the transformationfg.f,-' does not merely
relate the predicatesin different representations,it is a composite map that
points to the objective state x. The objectivity of x is guaranteedby its
invarianceunderall transformations,and thus abstractsfrom them all.'
ForAuyangit is this representation-transformation-invariance structurethat
the of
underpins generalconcept object. There is no privilegedrepresentation
either in observation, such as the sense datum, or in reality, providing the
unique God's Eye point of view. Thus Auyang rejects both phenomenalism
and metaphysicalrealism in the sense in which Putnamuses that term. If we
were given only the representationsthis, says Auyang, would lead to relati-
vism, which she also rejects. Every part of the structureexplained above is
necessary, accordingto Auyang, for the general concept of object.
When discussing the example of the table we explainedthe representations
as viewing the tablefromdifferentperspectives,butin discussionof Macbeth's
hallucinatorydagger Auyang makes it plain that representationmust involve
all the relevantmodalitiesof sight, touch, and so on, in the case of an everyday
'thing'. It is not perhapstotally clear what the relevant transformationrules
would be, but presumablywould comprisethe coherentlinkage of visual and
tactile impressions, etc. At all events Auyang wants her account to give
necessary, not just sufficient, conditions for objectivity. It should also be
stressedthatthe distinctionbetween the physical object and its representations
is not the distinctionbetween realityand appearance.The topic of knowledge,
whichAuyangrefersto as the 'empiricalobject',involvesthe fusionof thewhole
integratedstructure,of which the physicalobjectis just one conceptualelement.
Auyang sometimesrefersto the representationsof an object as conventions.
All she meansby this is the choice availableto us, which representationto use,
but again the concept of object involves all the possible choices, takenat one
gulp, so to speak.
It is clear thatAuyang arrivesat her analysis of object from the correspond-
ing definitionsof a manifoldin differentialgeometry,where x is a point in the
manifoldand fa(x), fp(x) are differentcoordinaterepresentationsof the point.
The composite map fp.f)-' is thenjust a passive coordinatetransformationon
the manifold. There are also active point transformationsof the manifold

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions
 The British Journalfor the Philosophy of Science 501

specified by the map f1 ~.f,. Auyang is perfectly clear on the mathematical

distinctionbetween coordinatetransformationsand point transformations,but
we shall see later that it may pose problemsfor her subsequentdiscussion of
field theories, in particulargeneral relativity.
So let us turnto Auyang's discussionof field theories.Objectificationis now
appliednot to thingsbut to events. Basically the events consist of the fields at a
given spacetimepoint being in a certain state. So attachedto each spacetime
point there is a 'quality space' comprisingpossible states of the field at that
point. These states may be representedin various ways, the representations
being connectedby a local symmetrygroup,local because it is indexed by the
spacetime point at which it acts. But the spacetime points are also variously
representedby differentcoordinatizationsof the spacetimemanifold,which in
turn are related by general coordinate transformations.So there are two
symmetry groups in play, one relating different representationsof the local
quality space, which Auyang refers to as the local symmetrygroup,the other
relatingrepresentationsof the 'spacetime' points, which Auyang refers to as
the spatio-temporalgroup.
We have been careful to introduce the two symmetry groups in their
passive versions, because that ties in with the ideal of different representa-
tions of objective entities. Auyang, however, tends to slide between the
passive and active interpretations,which could make for some confusion,
as we shall explain later.
But for the momentwe shall continuewith the exposition of Auyang's own
ideas. The invariantobject in the quality space tells us what kind of event we
aredealingwith;the invariantlocationin the spacetimemanifoldtells us which
particularevent of that kind we are referringto.
So for Auyang the role of spacetimeis to individuateevents. At this stage in
the discussion the individualevents are 'loose', there is no causal connection
between events. That will come later. But first, I want to explain Auyang's
ratherinterestingapproachto the natureof spacetime. It derives essentially
from interpretingthe structureswhich mathematicianscall fibre bundles.
Speakingvery crudely, a fibrebundle can be thoughtof in two ways. Firstly,
it can be thoughtof as constructedby attachingone sort of space, the fibre,to
each point of a second sortof space, the base space, so thatlocally the structure
is just the familiar Cartesianproduct.But there is a structuregroup imposed
which transformseach fibreinto itself, but in such a way that the transforma-
tions of differentfibres are independentof each other.
In the field-theorycase, the fibres are replicas of the quality space and the
base space is the spacetimemanifold.The local symmetrygroupis essentially
just the structuregroup of the bundle.
On this reading one can imagine the base space as existing, ontologically
speaking, in its own right quite independentlyof the fields which are then

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions
502 Review

'attached' to the points of spacetime. This would give a 'substantival'

interpretationof spacetimeas an independentlyexisting entity.
But there is a second way of thinking of a fibre bundle. Take the whole
bundle as primaryand introducean equivalencerelation,effectively of points
that lie on the same fibre. Then define the base space as the quotient of the
bundle by this equivalencerelation.Applied to field theorythe whole arrayof
particularfield events is what we start with, and spacetime is a derivative
notion constructedas the quotientspace,but makingno sense independentlyof
the fields. As Auyang puts it, spacetime is absolute, but not substantival.It
arises as a structuralaspect of the field which is itself the primaryontological
entity. In particular,on this second reading, it makes no sense to talk of
'emptying' spacetime of the fields-if there are no fields then, for Auyang,
there is no spacetime.
It should be noted that the mathematicsdoes not dictate which readingto
adopt-that is essentially a philosophicaldecision on Auyang's partto avoid
the chimera of a substantival spacetime. But it is also importantthat the
mathematicscan be glossed that way.
Auyang also rejectsa relationaltheoryof spacetimeas imposed 'externally'
on events. The role of spacetimeis to confer identityon events, to accountfor
theirdiversity.As Auyangputsit (p. 139), 'space [is] ... neithera containernor
a relatorbut a kind of divider.'
I now wantto turnto Auyang's discussionof how to introduceinteractionsin
field theory, i.e. causal relations between events. To be specific, consider a
chargedmatterfield it(x). We treatthe field classically to startwith, so one can
thinkof i as a (first-quantized)Schr6dingerfield, or if we wantto be relativistic
a Klein-Gordon or Dirac field. The Lagrangianfor the matterfield is invariant
underglobal phasetransformationsI(x) +(x) ei0, which leads via Noether's
theorem to the conservation of charge. ---*
But suppose we want to impose
invariance under local phase transformations,i.e. iV(x)--+ i(x) eiO(x),where
0(x) is now an arbitraryfunction of the spacetimelocation x.
The presence of derivativesia,t in the Lagrangianspoils the local invar-
iance, but it can be restoredby 'correcting'the derivative with a correction
term that specifies what is to count as 'the same phase' at differentspacetime
locations.
So a,,? in the Lagrangianis replacedby (a, - ieA,)i where A, specifies
what mathematicianscall the connection, which ties adjacentphases together
and itself transforms as A, -+ A, + e-la,0(x) under the local symmetry
transformations.It is then easily checked that the corrected derivative is
indeed invariantunder the local symmetrytransformations.But the A, field
is formallyidenticalto what we would get by coupling the matterfield 4(x) to
the electromagneticpotential A,, and the constante just measuresthe strength
of the coupling, namely the electric charge.

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions
 The British Journalfor the Philosophy of Science 503

So, to summarize,we have derived the electromagneticcouplingby impos-

ing invarianceunderthe local symmetrygroup.For historicalreasonsthe local
symmetrygroup is known as a gauge group (strictly of the so-called second
kind-the correspondingglobal symmetry is a gauge symmetry of the first
kind) and the hope is widely shared that all interactions,including electro-
magnetic,strong,weak, andeven gravitational,can be derivedby imposingthe
appropriatelocal gauge symmetry.
As we have seen, gauge symmetries are guaranteedby introducingcon-
nections on the relevant fibre bundles, which serve to spread the conven-
tional choice of representationof the matterfield across spacetime, so tying
together the local symmetry transformations,at different spacetime points.
The connection is the additionalstructurethat binds the independentevents
at different spacetime points into an interactive unity, and underlies the
fundamentalconcept of causation, the final element, says Auyang, required
for the intelligibility of the physical world over and above the objectivity
and identity of the fundamentalevents.
Having expoundedAuyang's own ideas, I will now turnto some possible
criticisms.
Let us start with the reasons for imposing local gauge symmetries.
Auyang sees this as arising from, and indeed in some sense solving, the
general philosophical problem of consistent predication, how to relate the
blue of this cup to the blue of that cup. Why should not we call the second
cup yellow? After all, it is a free world! But this trivial semantic conven-
tionalism is not the metaphysicalproblem of what it is about the world that
makes it true to say that different objects exhibit the same colour property
(whatever we choose to call it). However, it is not clear that Auyang's
exposition of local gauge symmetry really deals with that problem at all.
Firstly, for Auyang the gauge group is interpretedpassively. The connection
relates representationsof objective states of affairs at different spacetime
points, but it does not relate the states of affairs themselves. It is a bit like
the trivial semantic conventionalismwe discussed a moment ago. Secondly,
it is not at all clear that conventions should be treatedas propagatinglocally
(indeed accordingto some authorswith subluminalspeed). Conventionsjust
aren't the sort of thing that relativistic constraintsapply to. Thirdly, even if
the first two points were met, physics does not admit local gauge invariance
for all predicates,but only for a very limited number,so it could not be the
solution to the general philosophical problem.
Indeed, more generally, how can symmetry under a mere choice of con-
ventional representationdictate any genuinelyphysical principle at all?
I do not thinkAuyang really deals with any of these points, so let me hazard
my own views, which are relatedto Auyang's, but neverthelessare, I believe,
distinctively different.

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions
504 Review

There are two ways of dealing with the gauge freedom in physics which
come immediatelyto mind:
1. Remove the gauge freedom by just choosing a single representationin
Auyang's sense-that is, do what physicists call 'fixing the gauge'. But
there are two problemshere: (a) it may not be possible to fix the gauge
consistentlyacrossthe whole bundle.This goes in the tradeunderthe name
of the Gribov obstruction.And (b) when we quantize a gauge theory,
gauge fixing leads to a new sort of gauge freedom associated with the
requirementof preservingunitarity.This is a rigid fermionic symmetry
involving so-called ghost fields, known as the BRST symmetry after
its discoverers, Becchi, Rouet, Stora, and Tyutin. This symmetry is
fundamentalto the general proof of the renormalizabilityof gauge
theories.
2 Since the objective physical quantities are gauge-invariant,formulate
gauge theories in terms of these invariants. For example, the gauge
potentials specified by the connection depend on the gauge but the
gaugefields themselves, definedgeometricallyin termsof the curvature
of the connection, are gauge-invariant.So why not just use the fields as
opposed to the potentials in formulatinggauge theories?The difficulty
here is a rathersubtleone, thatpreventsthe formulationof a gauge theory
as a local theory at all. This is best seen by consideringthe Aharonov-
Bohm effect, which effectively measuresthe line integralof the connec-
tion (i.e. the potential)arounda closed curve enclosing a flux of 'curva-
ture', that is, a gauge field. This 'loop integral' is a gauge-invariant
quantity,but dependson the fields in regions in generalremotefrom the
loop in question.So if the fields are all thatis real theireffect on the non-
vanishingof the loop integralis in generalhighly non-local.Anotherway
of expressingthis situationis thatthe generalgauge-invariantquantities
are defined, not over a space of points in spacetimebut over a space of
loops in spacetime, again showing that they cannot in generalbe speci-
fied locally.
The reactionof most physicists to this situationis that the gauge potentials
are in some sense 'real', ratherthan being conventional, i.e. that the gauge
grouprelates not just representationsof events but events themselves. So the
gauge groupis now being given an active interpretation,but with the proviso
thatevents linked by gauge transformationsare observationallyindistinguish-
able. So, in a slogan, the real transcendsthe observable,but thatperhapsis not
too high a price to pay for restoringa truly local physics of the real.
So farwe have discussedthe first-quantizedversionof gauge theories.When
we turn to the second-quantizedversion, the full quantumfield theory, new
difficulties of interpretationarise.

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions
 The British Journalfor the Philosophy of Science 505

Firstly, Auyang claims that the fields are operator-valuedfunctions when

quantizationis employed. Technically that is not quite right:fields are opera-
tor-valued distributions defined over a space of test functions which are
themselves defined over spacetime. Crudely, the fields are not indexed by
points of spacetime, as Auyang opines, but are 'smeared' over regions of
spacetime. This may seem a technicality but requires some revision of
Auyang's account of spacetimepoints as conferringidentity on field events.
But, more importantly,let me stressthatin generalthe fields themselves, for
example a chargedmatterfield, are not quantum-mechanicalobservables,that
is, representedby self-adjoint operators.The observables are quantitieslike
charge densities or energy densities which are self-adjointconstructionsfrom
the fields. The fields themselves serve as intertwiningoperatorsconnecting
differentsuperselectionsectors of the theory. So the b(x) in the above discus-
sion of the first-quantizedgauge theoryis farremovedfromthe notion of event
in the second-quantizedtheory,even at the level of representationof the state
in a basis providedby 'observable'operators,as recommendedby Auyang in
her generaldiscussionof objectificationin quantummechanics.Putsimply, the
second-quantized charged fields are components of purely mathematical
structure.For example, the Dirac fields do not commute at spacelike separa-
tion. Far from it, they anti-commute.Moreover,any attemptto constructthe
fields out of local observablesis highly non-unique,so in a sense the fields may
be thought of as 'coordinatizing'the local observables. (This ambiguity in
specifying fields in observabletermswas firstformalizedby Borchers[1960],
althoughthe basic point really goes back to Dyson [1948].)
If we are to take the fields seriously in an ontological sense, then we may
need to invest the purely mathematicalingredientsin a theory with a kind of
reality. That, after all, was our conclusion when discussing the unobservable
gauge freedomin the first-quantizedversion of the theory.But all this leads us
far from Auyang's Kantiananalysis.
Finally I want to say something about Auyang's treatment of general
relativity (GR) as a gauge theory. There is considerable confusion in the
literatureas to what is meant by the gauge group of GR. Considered as a
constrained Hamiltonian system the gauge group is the group of general
coordinatetransformations.Noether's so-called second theoremresults,not in
conservationlaws, but in identities, indeed the contractedBianchi identities,
which reducefromten to six the numberof independentfield equations,andso
allow a gauge freedomassociatedwith four arbitraryfunctions,corresponding
to the arbitrarychoice of coordinate system. The general coordinate trans-
formations do not in general constitute a group from the global point of
view, since in general they cannotbe defined globally. But there is a globally
defined symmetry group, which is an invariance group of GR, namely the
diffeomorphismgroup,diff, which, from the local point of view, is the active

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions
506 Review

version of local coordinatetransformations.It is this groupthatleads to the so-

called hole argumentin GR. Briefly, a hole diffeomorphismis a diffeomorph-
ism that acts only inside a patch of spacetime, but reduces to the identity
outside the patch. From the invarianceunderdiff it follows that the dragged
fields, metric,etc. inside the patchcannotbe specifieduniquelyin termsof the
fields outside the patch, so determinismfails in the strongestpossible sense.
Now this state of affairsis often used as an argumentagainstsubstantivalism,
but it clearly also hits Auyang's absolutistconception of spacetimepoints. A
possible response is again to recognise that models of GR related by hole
diffeomorphismsare observationallyequivalent,so the failureof determinism
applies only to a reality which outstripsthe observational,a similar lesson to
the one we already seem to learn from trying to interpretgauge theories
generally, as in our discussion above.
Auyang signally fails to distinguish diff from coordinatetransformations,
and seems unaware of the extensive literatureon the hole argument.It is
perhapsthe most serious lacuna in the whole book.
But there is another sense in which one can understandGR as a gauge
theory,namelyby consideringthe tangentbundleof the spacetimemanifold,or
more appositely the related principalbundle, namely the frame bundle. The
local gauge group is GL(4, R) which reduces to SO(1, 3) if considerationis
restrictedto Lorentzianframes (or one might want to consider SL(2, C), the
covering groupof SO(1, 3) if spinorfields areto be introduced).Thereare now
two ways to go. Stick with the Lorentzgroup SO(1, 3) or introducean affine
structurein the fibres(to be sharplydistinguishedfrom an affineconnectionon
the bundle),so the local symmetrygroupbecomes the inhomogeneousLorentz
group,thatis, the Poincar6group.Auyang follows some influentialauthorsin
claiming thatthe Poincar6groupis needed if one wants to allow for torsionin
the spacetime manifold. I do not think this is right and refer the reader to
Invanenkoand Sardanashvily[1983] or Gickeler and Schiicker [1987], who
support,in my view correctly,a contraryview. We do not need an affinebundle
at all in orderto extend GR to the Einstein-CartanU4 theory incorporating
spin and torsion.
So far we have only discussed GR as a classical gauge theory.This, indeed,
is all thatAuyang does. But beyond this beckon the philosophicallyuncharted
waters of the quantumgravity programme.This potentiallyraises new diffi-
culties for Auyang's interpretationof the role of absolute spacetimepoints, if
in some sense spacetimeis itself being quantized,perhapseven at the level of
the local manifold structure.
So, in summary,I have some disagreementswith Auyang,both of a general
natureandalso at the level of some of the technicaldetail.But I wantto end this
review on a positive note. Auyang's book is beautifully written. She has
producedthe work, as far as I can tell, in intellectualisolation both from the

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions
 The British Journalfor the Philosophy of Science 507

philosophicalcommunity and indeed from the quantumfield theory commu-

nity. It is a noble effort to investigatethe philosophicalunderpinningsof field
theories and in particularof gauge theories. I have learnt much from reading
this book, not least in formulatingmy disagreementswith some of the argu-
ments. At all events I thoroughly agree with her final quotation from Ein-
stein:'Das Wirklicheis uns nicht gegeben sondernaufgegeben(nach Art eines
Ritsel).' As AuyangtranslatesEinstein:'Thereal is not given to us butis set us
as a task (by way of a riddle).'

References
Borchers,H. J. [1960]:'Uberdie Mannigfaltigkeit
derinterpolierenden
Felderzu einer
kausalenS-Matrix',NuovoCimento,15, pp.784-94.
Dyson, F. J. [1948]: 'The Interactions
of Nucleonswith MesonFields', Physical
Review,73, pp. 929-30.
Gockeler, M. and Schficker,T. [1987]: Differential Geometry,Gauge Theories, and
Gravity,Cambridge, UniversityPress.
Cambridge
D. andSardanashvily,
Ivanenko, G. [1983]:'TheGaugeTreatment
of Gravity',Physics
Reports,94, pp. 1-45.

This content downloaded from 195.78.108.40 on Wed, 25 Jun 2014 07:04:59 AM

All use subject to JSTOR Terms and Conditions