Today:
Uncertainty
relations for Spin
In QM, many physical
systems have complementary pairs of observables
which cannot be measured at the same time. Position (x) and momentum (px)
are one such pair. The product of the uncertainties must exceed
.
Another physical
quantity, spin, will be important in arguments to follow. Think of a spinning
ball. Its spin angular momentum points along the axis of rotation and has a
length equal to the rate of rotation times the moment of inertia. It is a
vector, , and all three
components can be specified.
That is the classical
picture. In QM, pairs of spin components satisfy uncertainty relations
. This means
that one can only know, or measure, one component of
spin at a time. (One can know the magnitude as well.)
When one measures sx, one
always obtains a multiple of .
When one measures s2, one always obtains
. s is a multiple of 1/2. Results of
spin measurements are quantized. A new feature: The
angular momentum of states in an atom can be zero. This runs counter
to the classical picture of orbits. In fact, the lowest energy state of the
hydrogen atom has L=0.
Experimental
implications
If you separate a beam
of neutrons into sx= +1/2 and sx =-1/2 beams (by running through magnet
pole-faces), you can discard the (-1/2) part and make a beam of pure sx =1/2 neutrons.
Now try measuring sy (just using a magnet turned 90°): you find that the measurements still give ±1/2, with a random pattern of + and - results.
If you take either the sy= 1/2 or the sy=
1/2 beam, and again try measuring sx, you
also find random results. The neutrons don't seem to be able to remember both
values at once. (the uncertainty relation)
But if you recombine the sy=
1/2 and the sy= 1/2 beams without measuring, i.e. without letting them interact
with some sort of detector, the resulting beam is still all sx=
+1/2.
Each sx= +1/2 was BOTH sy= +1/2 and sy= -1/2, and follows BOTH pathways . Only a "measurement" makes it choose
one or the other. Apparently sy is not
specified by a hidden variable, because each sx= +1/2 neutron seems to have both
values of sy.
A Science Fiction Story
Say that you want to
find out why people like pepperoni, mushrooms, and olives on pizza.
You ask many people,
and they give Yes/No answers when asked about each. (Say 50% yes for each.) But
you don't notice any distinguishing properties of the people who say Y.
Does that mean you can
conclude that the answer is random, some sort of momentary glitch that occurs
in a person's response?
Of course not- you may
just have missed the "hidden variables" needed to understand taste.
There MAY be something in each person's mind ahead of time that determines
their answer, or maybe not.
Try asking couples. You
find that each couple gives the same answer to the one question you ask them,
so you can be sure that there was already something in their heads that determined
the answer. Otherwise how could they all get their stories straight? So
there WAS some hidden variable determining the outcome.
(There's a slight
complication- all these folks seem to get confused after one question. If you
ask each member of the couples a second question, the perfect correlation
between them is lost. So you're only allowed to ask one of each person.)
Now you try asking one
person about pepperoni, and the other about mushrooms. You find that 85% of the
time, they give the same answer (YY or NN). So you conclude that whatever makes
people like mushrooms usually makes them like pepperoni.
You try the same thing
asking about mushrooms and olives. Again 85% of the time, they agree. (YY or
NN)
Now you ask about
pepperoni and olives. What extent of agreement do you expect? The 15% who had
different opinions on M-P and the 15% with different opinions on M-O might be
the same people. Then there would be 100% agreement on P-O. Or maybe they're
all different people. Then there would be 100%-15%-15% = 70% agreement on P-O.
Or it could come out anywhere in between.
You now ask the couples
do you like pepperoni/olives?
And you find they give
the same answer 50% of the time.
What gives?
Maybe there was a
statistical fluke. You try the same thing again with 100,000 couples- and get
the same result.
Maybe the couples were
secretly signaling each other, getting their stories coordinated (on some
questions) AFTER the questions were asked.
You ask the questions
in sealed boxes. Same result.
You ask the questions
SIMULTANEOUSLY (in Earth frame). Same result.
Maybe the couples were
getting their stories straight only on the questions which they knew they would
be asked.
You draw the questions out
of hats, in the sealed boxes, simultaneously. And get the same result.
Conclusion: this story
is obviously fiction. The identical results on any ONE question tell you that
(whether or not you believe that people in general are spontaneous and random)
on these particular issues there had to be some prior content in their heads
determining the answers. But there is NO WAY of assigning answers to the 3
questions (M,O,P) which gives these results for the
disagreements:
Violation
of triangle inequality!
Now
we tell exactly the same story about particles, where "couples" are
pairs emitted together in a decay process.
"Like
pepperoni?" becomes "spin up along x?"
"Like
olives?" becomes "spin up along y?"
"Like
mushrooms?" becomes "spin up along direction halfway between x and
y?"
Experimentally the results are as we have described. (Most of the real experiments
use a slightly different set-up with photon polarization, with the same logical
structure.)
Obviously
when we concluded that this story is science fiction we assumed something
that's false.
What
did we assume about reality?
Note
that we assumed nothing whatever about QM- we didn't even mention it. It
happens that QM precisely predicts the results, but we are now finding how the world
differs from our assumptions, not how quantum theory differs from our
assumptions.
WE assumed
Realism.
Not necessarily that the world is deterministic, but only that those aspects of
it which can be predicted perfectly are determined by some element of reality.
You can predict the result on any particle by making the corresponding
measurement on its partner.
"If,
without in any way disturbing a system, we can predict with certainty (i.e.,
probability equal to unity) the value of a physical quantity, then there exists
an element of physical reality corresponding to this physical quantity."
(Einstein, Podolsky, Rosen, 1935)
Local causality. The value of a property
possessed by an isolated system cannot be affected by any operations carried
out at a sufficient (i.e., spacelike)
separation from it.
Induction. The properties of an ensemble are defined completely by
the preparation conditions. In particular, the distribution of "possessed
values" of a variable for the subensemble which
we actually measure is identical to the distribution for the complete ensemble.
(random sampling)
Which
one are you willing to give up?
There's a
conceivable loophole: the initial passage of the particles through the
angular momentum selectors are space-like separated. However, in the current
generation of experiments, the conversion of those micro-events to some
large-scale device setting occurs slowly enough for it to be conceivable that a
time-like signal could propagate between the detectors before
"measurement" is complete. This again is a real long-shot, but one
that can be addressed experimentally just by speeding up the detectors and
setting them farther apart.
This loophole was
recently narrowed substantially in some Austrian experiments, in which the
distance between the detectors was several hundred meters, and in which the
analyzer settings were set by local random number generators. Experiments are
planned in which the detectors will be on satellites separated by several
hundred km. (These expts are funded by agencies
interested in quantum cryptography.)
Does the
problem lie with probabilities?
Our discussions have
concentrated on probabilities and correlations. This may give the incorrect
impression that the fundamental issue is the probabilities. It is possible to
devise a 3-particle spin measurement in which the
distinction between QM and LR can be seen in every single measurement.
Prepare a 3-particle (each
spin 1/2) state with the z-components of the spins described by ( - ¯¯¯)/21/2. Measure the
x-components of the spins.
A single measurement of
the 3 spins could do the job. (See article by N. D. Mermin
in Physics Today ,
June 1990.)
Here's some old notes, left in for
historical interest:
"Although it is
conceivable that this hypothetical experiment, if it is done, will agree with
LR rather than QM, don't bet on it. It would be very strange for QM to do a
perfect job of predicting the violations of LR so far, then break down and have
LR work in the non-probabilistic context. There's no theory that predicts that,
and there was only a vague hunch that somehow the application of probabilistic
reasoning might have something to do with the failure of LR."
The new parable:
You can ask one
question each of Anne,(A) Belinda (B), or Cary (C).
The questions can be "olives?" (X) or "Onions?" (Y) . The answers can be "yes" (+1) or
"no" (-1).
You always find that if
you ask one "olives?" and two "onions?" the product of the
answers is +1. So it sounds as if each answer is determined by an "element
of reality" since it is predictable by asking two OTHER questions.
X1Y2Y3
= Y1X2Y3= Y1Y2X3=1
So if those are just
numbers we're talking about:
X1Y2Y3Y1X2Y3Y1Y2X3
= X1X2X3 Y12 Y22
Y32 = X1X2X3 = 1
ALWAYS
BUT QM says something
strange: X1X2X3 = -1 ALWAYS
Essentially this
experiment (using photons rather than spin=1/2) has just been done, and the
result supports QM.
More than 85% of the
time, X1X2X3 = -1, was found which is impossible classically. The remaining 15% or
so is attributed to imperfections in the analyzers, etc.
Now we can return to
the "measurement" question, but without any false hope of finding a
mundane solution in terms of local hidden variables.
Which have we lost:
1. realism?
2. Local causality?
3. Induction?
Microscopic realism
The CI explicitly
abandons microscopic realism. However, the structure of our experiment
said nothing about assigning "reality" to the wave function or any
other microscopic construct. We argued only about macroscopic
outcomes: which detectors said 'up' or 'down'. There was a slight loophole- the
time it would take to send a signal between detectors allows only partial
conversion to a full-scale macro-outcome. However, recent experiments with the
detectors farther apart have kept pushing up the size of events to which we
cannot apply ordinary local realism. On a time scale of microseconds, very many
logic operations in a computer can be performed, etc.
Extrapolation of the
abandonment of realism to a slightly larger size scale would imply that it is
not true that counterfactual statements about fully macroscopic (brain-size)
events have definite truth value. The postulate of counterfactual
definiteness would be false. As Asher Peres said, "Unperformed
experiments have no results."
It becomes impossible
to say that IF the horizontal spin had been measured then the result would have
been (+ or -) *
This claim does not
deny the definiteness of results of experiments which are actually performed
("would have" ≠ "did").
* unless
you change the rules of logic, so that you cannot "distribute"
"or" i.e.
(Up &(left or right)) doesn't imply ((up & left) or (up & right))
If Induction is
lost, keeping local realism:
The principle of
induction says that the properties of the subensemble
on which some measurements are made are the same as those of the entire ensemble
of prepared systems. This seems natural if we are selecting the objects
"at random" from the ensemble. Is it possible that we are wrong about
this?
Breakdown of induction
implies that either
1. the emitter already knows what is
going to be measured and produces the appropriate subensemble.
The particles get their stories straight on just those questions which they
shall be asked, and not others. Or
2. the measurer is clued as to just
what questions are ok to ask, i.e. what questions the emitter will be ready to
answer.
But in recent expts, those questions are set by some sort of
"random" chance (e.g. another quantum event) in a space-like
separated (non-causal) way, off near the detectors.
That separation
requires either communication from the future to the past (detectors signaling
emitter) or that somehow the "random" events picking what will be
measured at the detectors have some prior cause which also sets which subensemble the emitter will produce, i.e. some very
intricate causal web of self-consistent connections between remote events, all
of which look random in local measurements.
So
dropping induction while keeping local realist causation really is the same as
postulating some entirely different theory of intricate local causes producing
distant connections which manage to look completely random locally. It is
impossible in principle to rule out all such conspiracies.
Why can we not rule out
conspiracies?
The
Formally,
However, any particular
conspiracy theory should have some testable additional implications.
If Local
causality is lost:
Remember what we needed
in Special Relativity:
1. Nothing with mass can travel
faster than c.
2. No conserved quantity can travel
faster than c.
3. No information can travel faster
than c.
We found it tempting to
generalize this to the idea that "ALL connections between events are a
causal network flowing forward in time at speeds not greater than c" but
that generalization was not logically necessary.
What happens if we say
"The correlation between the spins of the two particles exists, but not at
any particular locus in space-time." Someone trying to describe the
measurement by a picture completely imbedded in space-time will say "One
particle measured 'up", and that info was transmitted faster than c to the
other, making it spin 'down'". An observer in another frame would be
forced to say that the measurements happened in reverse order, and would
describe the flow of information as going the other way.
Does that mean signals
could be sent backward in time?
SRI said to the Navy:
You have to threaten retaliation if there's a nuclear attack. You don't want
your missiles blown up, and you certainly don't want to accidentally fire when
there's a false warning. SRI requested $$$$ to develop a way to wait until
there was a confirmed attack, then signal backwards in time and shoot first!
Fortunately, the Navy didn't bite.
What was nuts about
the SRI idea?
The correlated events
which one might interpret as having backwards causation are random quantum events.
They are not predictable on the basis of anything else in the universe (if QM
is correct) and thus cannot be used to send information. The output of any
receiver is a random string of bits, regardless of whether some other receiver
is getting a correlated random string of bits. You cannot send information via
random bit streams if neither you nor anything else in the universe can do
anything to affect what's in those bit streams
Notice that everything
is ok so long as the apparently random QM variables are really and truly
completely random, i.e. uninfluenced by any prior cause.
Things are also ok (if
obscure and terribly complicated) if these variables all have some
hidden deterministic cause.
A mixed case would seem
to lead to contradictions, in that signals could be sent backward in time.
Akin to the uncertainty
relations: consistent as a whole, but inconsistent if you removed just one.
Local causality
lost? (cont)
In other words, this
option requires that certain facts about the world not have any "place"
in it.
The direction of
"causation" (if you wish to use that concept) for this class of
events ceases to be Lorentz invariant.
The world has become
much stranger than we had imagined, but not, perhaps, stranger than we can imagine-
until we return to the current descriptions of the measurement process.