Transcript The Learnability of Quantum States

New Evidence That Quantum Mechanics Is
Hard to Simulate on Classical Computers

Scott Aaronson (MIT)
Joint work with Alex Arkhipov
Computer Scientist / Physicist
Nonaggression Pact
You tolerate these complexity classes:
P NP BPP BQP #P PH
And I don’t inflict these on you:
AM AWPP BQP/qpoly MA P/poly PSPACE QCMA
QIP QMA SZK YQP
In 1994, something big happened in the
foundations of computer science, whose meaning
is still debated today…
Why exactly was Shor’s algorithm important?
Boosters: Because it means we’ll build QCs!
Skeptics: Because it means we won’t build QCs!
Me: Even for reasons having nothing to do with building QCs!
Shor’s algorithm was a hardness result for
one of the central computational problems
of modern science: QUANTUM SIMULATION
Use of DoE supercomputers by area
(from a talk by Alán Aspuru-Guzik)
Shor’s Theorem:
QUANTUM SIMULATION is
not in probabilistic
polynomial time,
unless FACTORING is also
Today: A new kind of hardness result for
simulating quantum mechanics
Advantages:
Disadvantages:
Based on a more “generic”
complexity assumption than
the hardness of FACTORING
Applies to relational
problems (problems with many
possible valid outputs) or
sampling problems, not to
decision problems
Gives evidence that QCs have
capabilities outside the entire
polynomial hierarchy
Harder to convince a skeptic
that your QC is indeed
Only involves linear optics!
solving the relevant hard
(With single-photon Fock state inputs,
and nonadaptive multimode photon- problem
detection measurements)
Less relevant for the NSA
Before We Go Further, A Bestiary of
Complexity Classes…
COUNTING
P#P
PERMANENT
BQP
How complexity
theorists
PHsay
“such-and-such is damn
Xunlikely”:
YZ…
P
FACTORING
“If such-and-such is true,NP
then PH
BPP
collapses to a finite
level”
3SAT
Our Results
Suppose the output distribution of any linear-optics circuit
can be efficiently
sampled
classically
(e.g., byaMonte
If the PGC
is true,
then even
noisyCarlo).
#P=BPPNP).
Then thelinear-optics
polynomial hierarchy
collapses
(indeed
P
experiment can sample
Indeed,from
even ifasuch
a distribution
can be sampled
by a classical
probability
distribution
that
computer with an oracle for the polynomial hierarchy, still the
no classical computer can feasibly
polynomial hierarchy collapses.
sample from, unless the polynomial
Suppose the outputhierarchy
distribution
of any linear-optics circuit
collapses
can even be approximately sampled efficiently classically.
Then in BPPNP, one can nontrivially approximate the
permanent of a matrix of independent N(0,1) Gaussian
entries (with high probability over the choice of matrix).
“Permanent-of-Gaussians Conjecture” (PGC): The above
problem is #P-complete (i.e., as hard as worst-case PERMANENT)
Related Work
Knill, Laflamme, Milburn 2001: Linear optics with adaptive
measurements yields universal QC
Valiant 2002, Terhal-DiVincenzo 2002: Noninteracting fermions
can be simulated in P
A. 2004: Quantum computers with postselection on unlikely
measurement outcomes can solve hard counting problems
(PostBQP=PP)
Shepherd, Bremner 2009: “Instantaneous quantum
computing” can solve sampling problems that might be hard
classically
Bremner, Jozsa, Shepherd 2010: Efficient simulation of
instantaneous quantum computing would collapse PH
Particle Physics In One Slide
There are two basic types of particle in the universe…
All I can say is, the bosons
got the harder job
BOSONS
FERMIONS
Their transition amplitudes are given respectively by…
Per A 
n
a  


S n i 1
i,
i
Det A 
 1


sgn  
S n
n
a  
i,
i 1
i
Linear Optics for Dummies (or computer scientists)
Computational basis states have the form |S=|s1,…,sm,
where s1,…,sm are nonnegative integers such that s1+…+sm=n
n = # of identical photons
m = # of modes
For us, m>n
Starting from a fixed initial state—say, |I=|1,…,1,0,…0—
you get to choose any mm mode-mixing unitary U
U induces an
 m  n  1  m  n  1

  

 n   n 
states, defined by
unitary (U) on n-photon
S  U T 
PerU S ,T 
s1! sm!t1!tm!
Here US,T is an nn matrix obtained by taking si copies of the
ith row of U and tj copies of the jth column, for all i,j
Then you get to measure (U)|I in the computational basis
Upper Bounds on the Power of Linear Optics
Theorem (Feynman 1982, Abrams-Lloyd 1996): Linear-optics
computation can be simulated in BQP
Proof Idea: Decompose the mm unitary U into a product of
O(m2) elementary “linear-optics gates” (beamsplitters and
phaseshifters), then simulate each gate using polylog(n)
standard qubit gates
Theorem (Bartlett-Sanders et al.): If the inputs are Gaussian
states and the measurements are homodyne, then linearoptics computation can be simulated in P
Theorem (Gurvits): There exist classical algorithms to
approximate S|(U)|T to additive error  in randomized
poly(n,1/) time, and to compute the marginal distribution on
photon numbers in k modes in nO(k) time
By contrast, exactly sampling the distribution over
all n photons is extremely hard! Here’s why …
Given any matrix ACnn, we can construct an mm modemixing unitary U (where m2n) as follows:
 A B 

U  
 C D
Suppose we start with |I=|1,…,1,0,…,0 (one photon in
each of the first n modes), apply (U), and measure.
Then the probability of observing |I again is
p : I  U  I
2

2n
Per A
2
Claim 1: p is #P-complete to
estimate (up to a constant factor)
Idea: Valiant proved that the
PERMANENT is #P-complete.
Can use known (classical)
reductions to go from a
multiplicative approximation
of |Per(A)|2 to Per(A) itself.
Claim 2: Suppose we had a
fast classical algorithm for
linear-optics sampling. Then
we could estimate p in BPPNP
Idea: Let M be our classical
sampling algorithm, and let r
be its randomness. Use
approximate counting to
estimate PrM r  outputs I
: I  U
I a fast
 classical
Per A algorithm
Conclusion:pSuppose
we had
for linear-optics sampling. Then P#P=BPPNP.
2
2n
r
2

High-Level Idea
Estimating a sum of exponentially many positive or
negative numbers: #P-complete
Estimating a sum of exponentially many nonnegative
numbers: Still hard, but known to be in BPPNP  PH
If quantum mechanics could be efficiently simulated
classically, then these two problems would become
equivalent—thereby placing #P in PH, and collapsing PH
So why aren’t we done?
Because real quantum experiments are subject to noise
Would an efficient classical algorithm that sampled from a
noisy distribution still collapse the polynomial hierarchy?
Main Result: Take a system of n identical photons with
m=O(n2) modes. Put each photon in a known mode, then
apply a Haar-random mm unitary transformation U:
U
Permanent-of-Gaussians
Conjecture (PGC): This
problem is #P-complete
Let D be the distribution that results from measuring the
photons. Suppose there’s a fast classical algorithm that takes
U as input, and samples any distribution even 1/poly(n)-close
to D in variation distance. Then in BPPNP, one can estimate
the permanent of a matrix A of i.i.d. N(0,1) complex
Gaussians, to additive error
n! with high probability
,
over A.
O 1
n
PGC  Hardness of Linear-Optics Sampling
Idea: Given a Gaussian random matrix A, we’ll “smuggle” A
into the unitary transition matrix U for m=O(n2) photons—in
such a way that S|(U)|I=Per(A), for some basis state |S
Useful fact we rely on: given a Haar-random mm unitary matrix,
an nn submatrix looks approximately Gaussian
Then the classical sampler has “no way of knowing” which
submatrix of U we care about—so even if it has 1/poly(n)
error, with high probability it will return |S with probability
|Per(A)|2
Then, just like before, we can use approximate counting to
estimate Pr[|S]|Per(A)|2 in BPPNP, and thereby solve a
#P-complete problem
Problem: Bosons like to pile on top of each other!
Call a basis state S=(s1,…,sm) good if every si is 0 or 1 (i.e.,
there are no collisions between photons), and bad otherwise
If bad basis states dominated, then our sampling algorithm
might “work,” without ever having to solve a hard
PERMANENT instance
Furthermore, the “bosonic birthday paradox” is even worse
than the classical one!
2
Prboth particles land in the same box   ,
3
rather than ½ as with classical particles
Fortunately, we show that with n bosons and mkn2 modes,
the probability of a collision is still at most (say) ½
Experimental Prospects
What would it take to implement
the requisite experiment?
• Reliable phase-shifters and
beamsplitters, to implement an
arbitrary unitary on m photon
modes
• Reliable single-photon sources
Fock states, not coherent states
• Photodetector arrays that can
reliably distinguish 0 vs. 1 photon
But crucially, no nonlinear optics
or postselected measurements!
Our Proposal:
Concentrate on (say)
n30 photons and
m1000 modes, so that
classical simulation is
difficult but not
impossible
Open Problems
What are the exact resource requirements? E.g., can our
experiment be done using a log(n)-depth linear-optics circuit?
Prove the Permanent of Gaussians Conjecture!
Would imply that even approximate classical simulation of
linear-optics circuits would collapse PH
140Fr
Are there other quantum systems for which approximate
classical simulation would collapse PH?
Do a linear-optics experiment that solves a
classically-intractable sampling problem!
?