Title: Quantum Random Walks
1Quantum Random Walks
Combinatorial and Computational Aspects of
Statistical Physics/ Random Graphs and
Structures Cambridge, September 5, 2002
Julia Kempe Computer Science Division and
Department of Chemistry, University of
California, Berkeley CNRS LRI, Université de
Paris-Sud, France
2Towards nanotechnology
Gordon Moore 1965
Size of the components Number of
components Speed
Theoretical limitations reached in 2020 !!!
Apparition of quantum phenomena
prevent or use quantum effects ?
3Information is physical!
- Use the laws of quantum mechanics for the basic
components of an information processing machine! - Quantum computing
- Quantum cryptography
- Quantum information
-
4Main applications
- Cryptography
- Protocol of unconditionally secure secret key
distribution Bennett, Brassard 84 - Implementation 100 km
- Quantum information
- Teleportation B, B, Crépeau, Jozsa, Peres,
Wooters 93 - Implementation Bouwmeester, Pan, Mattle,
Eibl, Weinfurter, Zeilinger 97 - Algorithms
- Factoring, discrete logarithm, ... Shor 94
- Database search Grover 96
- Num. of qubits ? 1995 2, 1998 3, 2002 8
Chuang (IBM) - 10 Los Alamos
5The qubit
- Classical bit b?0,1
- Probabilistic bit probability distribution
d?R0,1 such that d1 1. ? d(p,1-p)
with p ?0,1 - Quantum bit ?? ? C0,1 such that ?? 21.
- ? ?? ? 0? ? 1? with ? 2 ? 21
- (Dirac notation)
6Qubit evolution
- Measure reads and modifies
-
0?
? 2
Measure
? 0? ? 1?
1?
? 2
? Superposition ? Probability distribution
- Unitary transformation U? C2?2 such that UUId
-
U
?? U ??
??
unitary ? reversible
U
??
U??
7Example
0?
1/3
Measure
??
1?
2/3
8Example
- Superposition
- Measure
- Unitary transformations
- NOT 0? ? 1?
- Hadamard
0?
1/3
Measure
??
1?
2/3
U
?? U ??
??
H
9Quantum computer n qubits
- n qubits ? tensor product ?? ? C0,1n such that
?? 21. - ? ?? ?x?0,1n ?x x? with ?x ?x 2 1
- Measure
- Partial Measure
?x 2
?x?0,1n ?x x?
x?
Measure
Second bit 0 (?2 ? 2 )
Measure
? 00? ? 01?? 10? ?11?
10Quantum computer n qubits
- n qubits ? tensor product ?? ? C0,1n such that
?? 21. - ? ?? ?x?0,1n ?x x? with ?x ?x 2 1
- Measure
- Partial Measure
- Unitary transformation ?? ?U?? with U? U(2n)
- ex XOR
?x 2
?x?0,1n ?x x?
x?
Measure
Second bit 0 (?2 ? 2 )
Measure
? 00? ? 01?? 10? ?11?
00? 01? 10? 11?
i?
i?
XOR(i,j)?
j?
11Quantum computing a function
- Let f 0,1n ? 0,1m
- x ? f(x)
- Reversible
- Rf 0,1nm ? 0,1nm
- (x,y) ? (x,y?f(x))
- Quantum
- Uf ?U(2nm) Cnm ? Cnm
- x?y? ? x?y?f(x) ?
12Simplest Quantum AlgorithmDeutschs Problem
- Input function f0,1?0,1 (in black box)
- Question f constant (f(0)f(1)) or balanced
(f(0)?f(1)) ? - Quantum black box (reversible)
- Algorithm one query only!!!
x?
x?
f
y?
y?f(x)?
H
H
0?
Measure
f
1?
H
0? -constant 1? -balanced
13Simplest Quantum AlgorithmDeutschs Problem
- Input function f0,1?0,1 (in black box)
- Question f constant (f(0)f(1)) or balanced
(f(0)?f(1)) ? - Quantum black box (reversible)
- Algorithm one query only!!!
x?
x?
f
y?
y?f(x)?
H
H
0?
Measure
f
1?
H
0? -constant 1? -balanced
0 if f constant
0 if f balanced
14Universal computation
- Classical circuit model
- Quantum circuit model
-
- evaluates boolean functions
- can be constructed from universal local gates
(ex. NAND, COPY)
?
0 0
0 1 0 1
?
bits
?
- unitary transformations U
0? 0? 1?
1? 0? 0?
U
qubits
Measure
15Quantum circuits
16Quantum Circuits
- Quantum circuits can simulate classical circuits
efficiently (with polynomial overhead) - Classical circuits can be efficiently simulated
by classical reversible circuits universal
reversible gate e.g. Toffoli-gate - Toffoli-gate can be generated with local unitary
gates on a quantum computer - -gt Classical circuits ? Quantum circuits
17Quantum algorithms
- Deutsch-Jozsa algorithm (92) determines if a
function (black box) is constant or 2-1 with only
one query - Simon s algorithm (94) period finding
18Quantum algorithms
- Deutsch-Jozsa algorithm (92) determines if a
function (black box) is constant or 2-1 with only
one query - Simon s algorithm (94) period finding
- Shor (95) efficient factoring
- general problem (factoring, discrete log)
hidden subgroup - Input function f G ? G s.t. f(x)f(xH) where
H? G - Output H (generators)
- efficient quantum algorithm if G - Abelian or
 specialÂ
19Quantum algorithms
- Deutsch-Jozsa algorithm (92) determines if a
function (black box) is constant or 2-1 with only
one query - Simon s algorithm (94) period finding
- Shor (95) efficient factoring
- general problem (factoring, discrete log)
hidden subgroup - Input function f G ? G s.t. f(x)f(xH) where
H? G - Output H (generators)
- efficient quantum algorithm if G - Abelian or
 special - Grover (96) Search of one entry in a database
of size N with queries
(Classical lower bound is ?(N)) - (quantum lower bound)
20Discrete Quantum Walks
Discrete-time walks on finite graphs (Mixing
Time) Dorit Aharonov
(Hebrew University) Andris Ambainis (IAS,
Princeton) J. K. (LRI, OrsayUC
Berkeley) Umesh Vazirani (UC
Berkeley) (STOC01)
Polynomial hitting time on the Hypercube J. K. (
02) hitting time on other graphs (numerical
Analytical studies) Neil Shenvi and J. K. (in
preparation 02)
Mixing on the Hypercube C. Moore and A. Russel
(quant-ph01)
21Markov chains
- Markov chains for algorithms
- Idea construct a Markov chain (simple, local
transitions only, efficiently implementable) - (1) whose stationary distribution gives the
solution to the problem ? Mixing time - or (2) which hits the desired solution ? Hitting
time -  Quantum Markov chains ?
22Example Random walk for 2SAT
Input Boolean formula ? (conjunction of clauses
of 2 variables) in X1, , Xn (ex.
) Question Is
? satisfaisable? (ex. YES, FFT is satisfying
assignment) Algorithm 1) initialise the
variables u.a. random (T-Â true, F-false)
2) if all clauses satisfied STOP,
otherwise 3) chose a non-satisfied clause,
chose one of its two variables and
flip its value return to 2)
23Example Random walk for 2SAT
Algorithm 1) initialise the variables u.a.
random (T-Â true, F-false) 2) if all
clauses satisfied STOP, otherwise 3)
chose a non-satisfied clause, chose one of its
two variables and flip its value
return to 2)
FFT
0
STOP
gt1/2
TFT
FFF
FTT
Hamming distance
1
lt1/2
gt1/2
FTF
TFF
TTT
2
lt1/2
gt1/2
TTF
3
Random walk on a line with n1 vertices ! After
t2n2 repetitions ( Hitting time ) the succes
probability is gt1/2 (if ? satisfiable).
24Random Walks...
- 3SAT - biased random walk with exponential
hitting time - in general local, simple Markov chain on
exponential domain
lt2/3
lt2/3
lt2/3
lt2/3
lt2/3
(fastest known 3-SAT algorithm based on random
walk Schöning99, Hofmeister, Schöning
Watanabe02)
STOP
0
1
2
3
4
5
gt1/3
gt1/3
gt1/3
gt1/3
gt1/3
25Random Walks...
- Random walk on the line Mixing timeHitting time
O(n2) stationary dist.uniform - Questions
- Stationary distribution? (ergodic gt
independent of initial state?) - Mixing time?
- Hitting time?
- Methods spectral gap, conductance, Log Sobolev,
coupling,
1/2
1/2
O(n2)
26Classical/quantum random walks
- Classical
- Transition matrix
- translationally invariant
- Dt(i)-distribution after time t
- stationary distribution
- measure of closeness total variation distance
- mixing time ? - time until ?ltconst.
1/2
1/2
O(n2)
27Classical/quantum random walks
- Classical Quantum
- Transition matrix
- translationally invariant
- Dt(i)-distribution after time t
- stationary distribution
- measure of closeness total variation distance
- mixing time ? - time until ?ltconst.
1/2
1/2
O(n2)
?
unitary? reversible? local translationally
invariant
28Quantum random walk
- Classical Markov process
- Quantum??? Unitary???
- Meyer 97 All local, translationary invariant
unitary matrices are simple translations.
R
L
29Classical random walk
R
L
- Incorporate coin-flip into walk!
- Classical walk in two steps
- ?,? ? (?,0),(?,0),(?,1),,(?,n-1),
(?,n-1) - flip direction coin C
- perform controlled shift S ? ? R Â Â
- ? ? L Â Â
- MSC
- Trace out (ignore, average over) the
direction-space
30Classical random walk
? ?
? ?
? ?
- ?,? ? (?,0),(?,0),(?,1),,(?,n-1),
(?,n-1) - Â
- MSC
- Trace out (ignore, average over) the
direction-space
- perform controlled shift ? ? R Â Â
- ? ? L Â Â
-
- S
31Quantum random walk
- Meyer 97 All local, translationary invariant
unitary matrices are simple translations. - coined walk in two steps
- ??,?? ?
- flip direction coin ( )
- perform controlled shift ?? ? R Â Â
- ?? ? L Â Â
R
L
H
H
unitary walk U
U collapses to the classical random walk if we
measure directions or positions at every step!
32Quantum random walk
R
L
- ?,? ? (?,0),(?,0),(?,1),,(?,n-1),
(?,n-1) - Â
- MSC
- After t steps measure
- Trace out (ignore, average over) the
direction-space
- perform controlled shift ? ? R Â Â
- ? ? L Â Â
-
- S
33Quantum random walks
0?
n-1?
1?
2?
- Example start in
- induces probability-dist. Pt(i) on the sites
(after measurement) - Convergence?
- NO! U is unitary ? reversible! (no stationary
distrib.) - Def. averaged distribution Qt (Cesaro limit)
- Theorem Qt converges to a stationary
distribution.
34Stationary distribution
0?
n-1?
1?
2?
- Theorem Qt converges to a stationary
distribution.
- Calculate eigenvectors/eigenvalues
of U - Expand initial state
- State at time t
- Stationary distribution
- if
35Stationary distribution
0?
n-1?
1?
2?
- Theorem Qt converges to a stationary
distribution.
- Stationary distribution
- uniform if G non-degenerate ( )
- If G also abelian -gt stationary distribution
uniform -
- characters of the abelian group (unit norm)
36Observations
- Classically real eigenvalues
- Quantum complex eigenvalues
- Classically behavior depends ( ) on second
largest eigenvalue - Quantum all eigenvalues equally important
- Ex mixing time determined by convergence of
- i.e. by
(minimum gap)
37Results on mixing time
0?
n-1?
1?
2?
- Cycle
- quantum walk converges towards uniform
distribution - Mixing time
- classical ? ?(N2 log(1/?))
- quantum ? O(N log N / ?3)
- Total variation distance
- Similar results in higher dimensions, for Cayley
graphs, graphs on abelian groups, walks with
different coins,
D. Aharonov,A. Ambainis, J.K., U.Vazirani-STOC01
38Results on mixing time
0?
n-1?
1?
2?
- Cycle
- quantum ? O(N log N / ?3)Â
-  Warmstart  to get logarithmic ?-dependence
- Initialize in
- Run quantum walk for steps -gt measure (node v)
- Restart new walk in (d-random)
- Repeat k-times
- Resulting distribution is -close to the
stationary distribution - (works if stationary distribution is independent
of initial state) -
D. Aharonov,A. Ambainis, J.K., U.Vazirani-STOC01
39Results on mixing time
- Conductance-type lower bound for mixing time of
any quantum walk on bounded degree graph - capacitance flow
- conductance
- Theorem (Jerrum,Sinclair89)
Classical
Quantum d-max.degree
D. Aharonov,A. Ambainis, J.K., U.Vazirani-STOC01
40Conductance
- Quantum d-max.degree
- Cut (X,X) of G, boundary
- Idea start with state concentrated in X and show
that at each time step leakage into X is
bounded by . - Then after steps
- And hence
41Quantum Hitting Time on Hypercube
42Quantum Hitting Time on Hypercube
- Space
- Walk
- Conditional Shift
- Coin C (respects permutational symmetry of
hypercube)
43Quantum Hitting Time on Hypercube
- Space
- Walk
- Conditional Shift
- Coin C (respects permutational symmetry of
hypercube) - Initial state
Symmetric superposition over all directions
Mixing time classical quantum (coupon
collector) (MooreRussel01)
44Hitting time?
- Dilemma constant measurement of position will
collapse U to the classical walk - Two options
- One-shot q-hitting-time (T,p)
- Measure only at time T
- Hits desired target-state x with probability gtp
- Concurrent q-hitting-time (T,p)
- Partial measurement (Am I at x/Am I not at x?)
at all times - Stop walk if x is hit. Probability gtp to hit x
before time T
45Results on hitting time
- Classical from v to opposite v hitting-time
- Quantum
- One-shot hitting-time from v to v (T,p)
and
(T-n) even,
J.K.02
46Results on hitting time
- Classical from v to opposite v hitting-time
- Quantum
- One-shot hitting-time from v to v (T,p)
- Need to know with accuracy when to
measure, success ?1 in linear time! - Concurrent hitting-time from v to v (T,p)
- No information on when to measure needed, with
amplification success ?1 in TO(n2)!
and
(T-n) even,
J.K.02
47Details
- Use symmetry to calculate eigenvalues/eigenvectors
of unmeasured walk U - Assymptotics to calculate hitting probability
at T - ? one-shot hitting time (T,p)
- For concurrent hitting time give a lower bound on
hitting probability in terms of unmeasured walk
U - Lemma
48Robustness of initial condition
- Polynomial hitting time to opposite corner, how
long from other sites (or to sites close to
corner)? - close initial
- states give similar polynomial behavior
- Upper bound
- Region around v of polynomial hitting time to v
at most - (otherwise we could find search algorithm that
beats the lower bound for quantum searching
(Grover))
49Open graphs
Example
start
hit
n-level binary tree
Reduces to assymetric walk on the line
(classically and quantum).
2/3
2/3
2/3
2/3
1/3
1/3
1
2
n
n1
1/3
1/3
1/3
2/3
2/3
2/3
A.Childs, E.Farhi, S. Gutman, quant-ph/01
50Open graphs
Example
Classical O(exp(n)) hitting time
start
hit
Quantum (numeric) poly(n) hitting
time (N.Shenvi J.K.02)
n-level binary tree
Reduces to assymetric walk on the line
(classically and quantum).
2/3
2/3
2/3
2/3
1/3
1/3
1
2
n
n1
1/3
1/3
1/3
2/3
2/3
2/3
A.Childs, E.Farhi, S. Gutman, quant-ph/01
51Outlook/Open questions
- In general which graphs have exponential
quantum/classical gaps in hitting times ? - How robust is this gap w.r.t. initial
position/distribution? - Mixing times for non-abelian walks ?
- Mixing times for walks on non-bounded degree
graphs? - For degenrate or non-abelian groups stationary
distribution depends on initial state
-algorithmic use? - Algorithmic use?
52Outlook/Open questions
- General THEORY???!!!
- Connection to classical
- Markov chains?
53Collaborators and related work
Discrete-time walks (Mixing Time)
(On the Line) Dorit Aharonov
(Hebrew University) A. Ambainis, E.
Bach, A. Nayak, Andris Ambainis (IAS, Princeton)
A. Vishwanath, J. Watrous J. K. (LRI,
OrsayUC Berkeley) (STOC01) Umesh Vazirani
(UC Berkeley) (STOC01)
Polynomial hitting time on the Hypercube J. K.
(submitted 02) hitting time on other graphs
(numerical Analytical studies) Neil Shenvi
and J. K. (in preparation 02)
Mixing on the Hypercube C. Moore and A. Russel
(quant-ph01)