Professional Documents
Culture Documents
Outline
Simulations
QA SA
2011
1983
Hardwares
CA
~10.000 b.C. Guglielmo Mazzola | QuAASI 2016 | 3
Implement “Quantum Mechanics”
SQA
Implement classical
optimization
QA SA
QA SA
Quantum Hardware
SA
Santoro et. al. Science 295, 2427 (2001)
QA
cl ee
as d
does not display
sp
si up
du m
ca
p
sp antu
quantum speed-
l
ee
up compared to
qu
SA.
QA SA
QA SA
2014: D-Wave
does not display
N.B. stoquastic drivers such
as transverse field quantum speed-
hamiltonians up compared to
always assumed
SA.
Guglielmo Mazzola | QuAASI 2016 | 9
QA SA
2014: D-Wave
does not display
N.B. stoquastic drivers such
as transverse field quantum speed-
hamiltonians up compared to
always assumed
SA.
Guglielmo Mazzola | QuAASI 2016 | 10
Outline
| |
Quantum tunneling in the smallest droplet of water
Guglielmo Mazzola | | 15
Santoro et. al. Science 295, 2427 (2001) AQC 2016
Path Integral Monte Carlo
H H
Density matrix is an operator: ⇢=e Z = Tr e
Matrix elements are difficult to H 0 H = H1 + H 2
hq|e |q i because [H1 , H2 ] 6= 0
compute:
h iM
⌧H ⌧ H1 ⌧ H2 H H1 H2
instead e ⇡e e e = lim e M e M
M !1
Z
hq|e H
|q 0 i = dq1 dq2 · · · dqM hq|e ⌧H
|q1 i q
hq1 |e ⌧H
|q2 i · · · hqM |e ⌧H 0
|q i
··· q0
0 ⌧ 2⌧ 3⌧
H H
Density matrix is an operator: ⇢=e Z = Tr e
Matrix elements are difficult to H 0 H = H1 + H 2
hq|e |q i because [H1 , H2 ] 6= 0
compute:
h iM
⌧H ⌧ H1 ⌧ H2 H H1 H2
instead e ⇡e e e = lim e M e M
M !1
Z
hq|e H
|q 0 i = dq1 dq2 · · · dqM hq|e ⌧H
|q1 i q
⌧H ⌧H 0 q0
hq1 |e |q2 i · · · hqM |e |q i
q↵
H H
Density matrix is an operator: ⇢=e Z = Tr e
Matrix elements are difficult to H 0 H = H1 + H 2
hq|e |q i because [H1 , H2 ] 6= 0
compute:
h iM
⌧H ⌧ H1 ⌧ H2 H H1 H2
instead e ⇡e e e = lim e M e M
M !1
Z
H ⌧H
hq|e |qi = dq1 dq2 · · · dqM hq|e |q1 i q
⌧H ⌧H
hq1 |e |q2 i · · · hqM |e |qi q↵
Z
H ⌧H
hq|e |qi = dq1 dq2 · · · dqM hq|e |q1 i q
⌧H ⌧H
hq1 |e |q2 i · · · hqM |e |qi
X
S[path]
Z= e
NB: only possible for stoquastic H
paths
X X
z z
H= Jij i j + sxi
ij i
?
J
M =4
M =8 M = 16 M =1
Continuous time limit (CT)
One could also optimise M at given
Outline
The runtime of the (ideal) Quantum How does the runtime of a QMC
device is dictated by the simulated annealing algorithm scale?
Quantum Adiabatic theorem:
2 TQM C / TQA ?
TQA /
Guglielmo Mazzola | QuAASI 2016 | 24
Path Integral Monte Carlo pseudodynamics
Doing the integral with Monte Carlo by sampling ring-
S[q(⌧ )]
polymer configurations (paths) with Metropolis weight e
y
q
Evolution of the classical
path as a function of the
simulation time t
q(⌧, t)
given by the Metropolis
pseudo-dynamics
(updates).
⌧ (updates).
d
| (t)i = iH(t)| (t)i
dt
S[x⇤⇤ (⌧ )] 2
e ⇠
q ⇤⇤ (⌧ )
q(⌧, t = 0) 0
y Guglielmo Mazzola | AQC 2016 |
⌧
27
Transition states of the dynamics: instantons
Consider double well problems @q(⌧, t) S[q(⌧, t)]
= + ⌘(⌧, t)
@t q(⌧, t)
S[q(⌧, t)] Transition state (TS)
Transition states defined by: =0 is the saddle point
q(⌧, t) with non-trivial
boundary conditions
x
⇤⇤
for double well is known: the instanton x (⌧ )
S[x⇤⇤ (⌧ )] 2
e ⇠
0
Guglielmo Mazzola | AQC 2016 |
⌧
28
Transition states of the dynamics: instantons
The classical field, in a PIMC simulation, evolves through a Langevin equation,
S[q ⇤⇤ (⌧ )] 2
k/e ⇠
2
↵
q ⇤⇤ (⌧, t = Ttunn )
1/
q(⌧, t = 0)
Ttunn : Number of QMC updates required to generate q ⇤⇤ (⌧ ) x0
Guglielmo Mazzola | AQC 2016 | 30
Ferromagnetic Ising system
X X
z z x
Consider now a spin system. H=J i j + i
i,j i
L R
z
1 m =< > 1
2
< TP BC >⇠ 1/
x x
S[x⇤ (⌧ )]
e ⇠
⇤
x (⌧ )
⇤⇤
x (⌧ )
y ⌧
0 Guglielmo Mazzola | QuAASI 2016 | 33
QMC tunneling rate in ferromagnetic Ising system
We are currently working on finding possible exceptions to this matching, exploring models
(borrowed from chemistry) which feature multidimensional tunnelling.
Clearly, also Hamiltonians with sign-problem (non stoquastic), beyond the Transverse
field, may show scaling advantage compared to QMC.
Guglielmo Mazzola | QuAASI 2016 | 35
Thank you!
36
| |