quantum versus classical probabilitymuller/ctms/lectures/rau_090818.pdf · classical probability...
TRANSCRIPT
q-info.org
Quantum versus classical probability
Jochen Rau
Goethe University, Frankfurt
Duke University, August 18, 2009
arXiv:0710.2119v2 [quant-ph]
q-info.org
Motivation from quantum foundations
Reconstructing quantum theory:
In search of a physical principle
Duke University, August 2009 1J. Rau | Quantum vs classical probability
Special relativity Quantum mechanics
Mathematical
apparatus
Physical
principle
Lorentz transformations • States
• Unitary evolution
• Measurement
• space and time not absolute
but observer-dependent
• causality (light cone)
preserved in all frames?
q-info.org
Relevance for quantum gravity
The issue is pertinent in the quest for a theory of
quantum gravity
Duke University, August 2009 2J. Rau | Quantum vs classical probability
General relativity Quantum mechanics
Mathematical
apparatus
Physical
principle
• Pseudo-Riemannian manifold
• Einstein equation
• States
• Unitary evolution
• Measurement
• local flatness
• equivalence principle
• ? ?
q-info.org
Examples for reconstruction
There have been many reconstructive approaches –
none of them entirely satisfactory
Duke University, August 2009 3J. Rau | Quantum vs classical probability
Name Protagonists Principal idea Shortcomings
Quantum logic• Birkhoff, v. Neumann
• Jauch et al
(„Geneva School“)
• Generalise classical logic: lattice theory
• Propositions subspaces of Hilbert sp.
• Skew field remains unspecified
(R, C, H)
• Proof only for d≥3
Algebraic
• Jordan, v. Neumann,
Wigner
• Gelfand, Neumark, Segal
• Haag, Kastler
• Algebra of operators (C* algebra)
• Hermitian elements observables
• Physical understanding?
Operational,
convex cone
• Mackey
• Ludwig et al
(„Marburg School“)
• Varadarajan, Gudder
• Focus on primitive laboratory operations
• Convex geometry of state space
• Convexity does not suffice
• Additional mathematical
assumptions needed
Test spaces
• Foulis, Randall et al
(„Amherst School“)
• Generalise probability theory • How to single out quantum theory
from the multitude of conceivable
generalisations of classical
probability?
q-info.org
Motivation from quantum computing
Interest in quantum foundations has been revived
by the advent of quantum computation
Duke University, August 2009 4J. Rau | Quantum vs classical probability
classical probability
theory
sum rule, Bayes rule
classical information
Shannon
classical computing
Turing, network model
quantum probability
interference, entan-
glement, Bell,
Kochen-Specker
quantum
information
Holevo, Schumacher
other non-classical
probability theories
generalised
information
quantum computing
no-cloning,
teleportation
other non-classical
forms of informa-
tion processing
Physics / computer
science / logic
?
?
Today‘s talk
q-info.org
Major differences
Quantum theory and classical probability are often seen
as two very different theories...
Duke University, August 2009 5J. Rau | Quantum vs classical probability
classical
probability
quantum
theory
• amplitudes, interference
• non-commutativity
• uncertainty relations
• entanglement
• violation of Bell inequalities
• Kochen-Specker theorem
• ...
q-info.org
Overlap
...yet they share a substantial common core
Duke University, August 2009 6J. Rau | Quantum vs classical probability
class-
ical
quan-
tum
common
core
q-info.org
Logical structure
Both classical and quantum theory deal with propositions
and logical relations
Duke University, August 2009 7J. Rau | Quantum vs classical probability
orthomodular poset / orthoalgebra / test space
weaker than classical: , defined iff jointly decidable
q-info.org
Reasoning in the face of uncertainty
Probabilities satisfy sum and Bayes rules –
central to ensure consistency
Duke University, August 2009 8J. Rau | Quantum vs classical probability
Bayesian probability
embodies agent‘s
state of knowledge
degree of belief rather
than limit of relative
frequency
can be legitimately
assigned not just to
ensembles but also to
individual systems
Consistency
Different ways of using the
same information must lead
to the same conclusions,
irrespective of the particular
path chosen+
Cox 1946, Jaynes 2003
Quantitative rules
Sum rule
Bayes rule
quantum:
only if probabilities
pertain to propositions
that are jointly decidable
q-info.org
Agent intervention
Agents can intervene by selection or interrogation
Duke University, August 2009 9J. Rau | Quantum vs classical probability
machine
Generic setup
000
11..
input
register
r
100
01..
output
register
p
1-p
Interrogation
finite or infinite sequence of reprod‘le
measurements, possibly stochastic
register = record of results
measurement setups may be
controlled by intermediate results
(feed forward, as in 1-way q. comp.)
no selection, no waste
I
Two examples
Selection
sequence of reprod‘le measurements
register = record of results
selection rule, possibly stochastic
keep or discard the system
S
1111
10..
algorithm
q-info.org
S(ki) = dim M({ki}) + S(ki)
Most general selection
Arbitrary stochastic selection rules are allowed –
selection can prepare arbitrary states
Duke University, August 2009 10J. Rau | Quantum vs classical probability
classical 0 ki
quantum Σijkikj ki2
prob(x|r) = in...i1 prob(x|in...i1,mn...m1,)prob(in...i1|mn...m1,)prob(s|in...i1,S)
record of results
(string of integers)
sequence of n
measurement setups
record remains
internal to machine
{{xi}|xi=I,d(xi)=ki} parameters of a non-normalised state
normalised toprob(s|S,mn...m1,)-1 selected
q-info.org
Most general interrogation
Arbitrary interrogation algorithms are allowed
Duke University, August 2009 11J. Rau | Quantum vs classical probability
record of results
(string of integers)
prob(x|r) = I,r prob(x|r,I,)prob(r|I,)prob(I|r,I)
sequence of
meas‘t setups
normalised
to one
Convexity
arbitrary mixtures constitute valid posteriors
states form a convex set
feed forwardstochastic
record remains
internal to machine
q-info.org
Key question
Which additional assumptions are needed to arrive
at the classical and quantum case, respectively?
Duke University, August 2009 12J. Rau | Quantum vs classical probability
class-
ical
quan-
tum
common
core
? ?
Not ħ: We consider probability
theory, not dynamics!
variable
value 1 value 2ideal:
q-info.org
Response to interrogation
Only in the quantum case it is possible to prepare arbitrary
states by mere interrogation
Duke University, August 2009 13J. Rau | Quantum vs classical probability
Preceding interrogation can make
any measurement emulate any other
measurement of the same resolution:
{xi},{yi}M({ki}) I:
prob(yi|I,) = prob(xi|) i,
based on
quantum Zeno effect
Interrogation can steer a system from
any pure state to any other pure state:
e,f I: prob(x|I,e) = prob(x|f) x
v. Neumann 1932
Quantum malleability
Interrogation has no effect:
prob(x|I,) = prob(x|) I, x,
Classical invariability
based on
joint decidability
Deterministic, static interrogation
q-info.org
Agent-dependency
Responsiveness to interrogation adds another layer
to agent-dependency
Duke University, August 2009 14J. Rau | Quantum vs classical probability
Classical
Quantum
elicitmay
inform
answers
(data)
prior
expectation + conclusion
questions
data conclusionOrthodox:
Bayes: dataprior
expectation + conclusion
: agent-dependent
q-info.org
Wheeler‘s game of 20 questions (1/2)
In quantum theory conclusions depend on questions asked
Duke University, August 2009 16J. Rau | Quantum vs classical probability
Wheeler 1983 („Law without law“)
q-info.org
The word does not already exist „out
there“ – rather, information about the
word is brought into being through the
questions raised
1
The answers given are internally
consistent; convergence to some
word is possible
2
Wheeler‘s game of 20 questions (2/2)
In quantum theory conclusions depend on questions asked
Duke University, August 2009 17J. Rau | Quantum vs classical probability
Wheeler 1983 („Law without law“)
The questioner has influence on the
outcome: a different sequence of
questions will lead to a different word
3
...like in quantum theory:
•There is no preexisting reality that is
merely revealed, rather than
influenced, by the act of measurement
•The image of reality that emerges
through acts of measurement reflects
as much the history of intervention as it
reflects the external world
q-info.org
World view
This new form of agent-dependency leads to a radically
different world view
Duke University, August 2009 18J. Rau | Quantum vs classical probability
Classical Quantum
reality with
independent
existence
manipulate in
experiment: grab,
dissect into parts,
select, transform
observe without
disturbance
?
!
(partial)
record of
answers:
0111001
1101100
....ongoing
interrogation
image of reality
q-info.org
Conjecture
Can agent-dependency serve as a foundational principle
for quantum theory?
Duke University, August 2009 19J. Rau | Quantum vs classical probability
Special relativity Quantum mechanics
Mathematical
apparatus
Physical
principle
Lorentz transformations • States
• Unitary evolution
• Measurement
• space and time not absolute
but observer-dependent
• causality (light cone)
preserved in all frames
• image of reality not absolute
but agent-dependent
• consistency of reasoning
preserved for all agents
?
q-info.org
Response to interrogation
Only in the quantum case it is possible to prepare arbitrary
states by mere interrogation
Duke University, August 2009 20J. Rau | Quantum vs classical probability
Preceding interrogation can make
any measurement emulate any other
measurement of the same resolution:
{xi},{yi}M({ki}) I:
prob(yi|I,) = prob(xi|) i,
based on
quantum Zeno effect
Interrogation can steer a system from
any pure state to any other pure state:
e,f I: prob(x|I,e) = prob(x|f) x
v. Neumann 1932
Quantum malleability
Interrogation has no effect:
prob(x|I,) = prob(x|) I, x,
Classical invariability
based on
joint decidability
Hidden, deterministic, static interrogation
q-info.org
Smoothness
Maximum agent-dependency rests on the Zeno effect,
which in turn presupposes smoothness
Duke University, August 2009 21J. Rau | Quantum vs classical probability
For finite granularity d:
1 Set of pure states X(d) is a continuous, compact, simply connected manifold
X(d) Robustness under small preparation
inaccuracies
>0 >0: prob(x|e)>1- eB(e0;)
Continuity condition:
ee0
B(e0;)
prob(x|e)
(x e0)
2 Probabilities change in a continuous fashion:
q-info.org
Analysis of symmetry group
Continuity in combination with common core leads to
quantum theory in Hilbert space over R, C or H
Duke University, August 2009 23J. Rau | Quantum vs classical probability
Constraint Implication for group
probabilities change in a
continuous fashion
dim G(d) =
dim X(2)/2d(d-1) + dim G(1)d
X(d) continuous Lie group
dimensions match dim X(d) =
dim G(d) - dim G(d-1) - dim G(1)
Definition
Automorphisms of a
proposition system
preserve
logical relations
, , \
granularity d
Irreducible building
block
Group acts transitively
on collections M({ki}):
M({ki}) ~ G(ki)/iG(ki)
X(d) compact and simply
connected (hull of convex set)
compact and simple
up to Abelian factor
X(d) ~ M({1,d-1}) transitive on X(d)
G(d) {SO(nd), U(nd), Sp(nd) | nN}
q-info.org
Conclusion
Classical and quantum case derive from common core
via additional assumptions on agent-dependency
Duke University, August 2009 24J. Rau | Quantum vs classical probability
class-
ical
quan-
tum
common
core
agent-
dependencymin
prob(x|I,)=prob(x|)
max
e,f I: prob(x|I,e)=prob(x|f)
R, C or H?1
?
intermediate
cases2
q-info.org
Summary
Duke University, August 2009 28J. Rau | Quantum vs classical probability
Quantum theory shares with classical probability theory
a substantial core of common properties
Quantum theory and classical probability theory mark opposite
extremes in their response to interrogation – quantum theory
maximises agent-dependency
No reference to unitary evolution – primitive operations are
measurements only1
1 see also: T. Rudolph; measurement-based quantum computing
q-info.org
Duke University, August 2009 29J. Rau | Quantum vs classical probability
Further details:
arXiv:0710.2119v2 [quant-ph]