Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Optimization of NMR Experiments using Genetic
Algorithm: Applications in Quantum Information
Processing, Design of Composite Operators and
Quantitative Experiments
A Thesis
Submitted For the Degree of
Doctor of Philosophy
in the Faculty of Science
by
V. S. MANU
Department of Physics
Indian Institute of Science
BANGALORE – 560 012
December 2013
Declaration
I hereby declare that the work reported in this thesis is entirely original. It was
carried out by me in the Department of Physics, Indian Institute of Science, Ban-
galore. I further declare that it has not formed the basis for the award of any
degree, diploma, membership, associateship or similar title of any university or
institution.
December 15, 2013 (V. S. MANU)
iii
“I study myself more than any other subject. That is my metaphysics, that is my
physics.”
Michel de Montaigne
Publications based on this Thesis
1. Non-Destructive Discrimination of Arbitrary Set of Orthogonal Quantum
States by NMR using Quantum Phase Estimation,
V. S. Manu and Anil Kumar,
AIP Conference Proceedings. 1384 229(2011).
2. Singlet-state Creation and Universal Quantum Computation in NMR using
a Genetic Algorithm,
V. S. Manu and Anil Kumar,
Phys. Rev. A 86 022324(2012).
3. Fast and Accurate Quantification using Genetic Algorithm Optimized 1H-13C Refocused Constant-Time INEPT,
V. S. Manu and Anil Kumar,
Journal of Magnetic Resonance 234 106 (2013).
4. Quantum Simulation using Fidelity Profile Optimization,
V. S. Manu and Anil Kumar,
Phys. Rev. A (Under Review).
5. Efficient Creation of NOON States in NMR,
V. S. Manu and Anil Kumar (To be communicated),
6. Composite Operator Design in NMR using Genetic Algorithm,
V. S. Manu and Anil Kumar (To be communicated).
vii
Abstract
Genetic algorithms (GA) are stochastic global search methods based on the me-
chanics of natural biological evolution, proposed by John Holland in 1975. Here
in this thesis, we have exploited possible utilities of Genetic Algorithm optimiza-
tion in Nuclear Magnetic Resonance (NMR) experiments. We have performed
(i) Pulse sequence generation and optimization for NMR Quantum Information
Processing, (ii) efficient creation of NOON states, (iii) Composite operator design
and (iv) delay optimization for refocused quantitative INEPT. We have generated
time optimal as well as robust pulse sequences for popular quantum gates. A
Matlab package is developed for basic Target unitary operator to pulse sequence
optimization and is explained with an example.
Chapter 1 contains a brief introduction to NMR, Quantum computation and Ge-
netic algorithm optimization. Experimental unitary operator decomposition using
Genetic Algorithm is explained in Chapter 2. Starting from a two spin homonu-
clear system (5-Bromofuroic acid), we have generated hard pulse sequences for
performing (i) single qubit rotation, (ii) controlled NOT gates and (iii) pseudo
pure state creation, which demonstrates universal quantum computation in such
systems. The total length of the pulse sequence for the single qubit rotation of an
angle π/2 is less than 500µs, whereas the conventional method (using a selective
soft pulse) would need a 2ms shaped pulse. This substantial shortening in time
can lead to a significant advantage in quantum circuits. We also demonstrate the
creation of Long Lived Singlet State and other Bell states, directly from thermal
equilibrium state, with the shortest known pulse sequence. All the pulse sequences
generated here are generic i.e., independent of the system and the spectrometer.
We further generalized this unitary operator decomposition technique for a vari-
able operators termed as Fidelity Profile Optimization (FPO) (Chapter 3) and
performed quantum simulations of Hamiltonian such as Heisenberg XY interaction
and Dzyaloshinskii-Moriya interaction. Exact phase (φ) dependent experimental
unitary decompositions of Controlled-φ and Controlled Controlled-φ are solved
using first order FPO. Unitary operator decomposition for experimental quantum
simulation of Dzyaloshinskii-Moriya interaction in the presence of Heisenberg XY
interaction is solved using second order FPO for any relative strengths of interac-
tions (γ) and evolution time (τ). Experimental gate time for this decomposition
ix
is invariant under γ or τ , which can be used for relaxation independent studies of
the system dynamics. Using these decompositions, we have experimentally veri-
fied the entanglement preservation mechanism suggested by Hou et al. [Annals of
Physics, 327 292 (2012)].
NOON state or Schrodinger cat state is a maximally entangled N qubit state
with superposition of all individual qubits being at |0〉 and being at |1〉. NOON
states have received much attention recently for their high precession phase mea-
surements, which enables the design of high sensitivity sensors in optical interfer-
ometry and NMR [Jones et al. Science, 324 1166(2009)]. We have used Genetic
algorithm optimization for efficient creation of NOON states in NMR (Chapter 4).
The decompositions are, (i) a minimal in terms of required experimental resources
– radio frequency pulses and delays – and have (ii) good experimental fidelity.
A composite pulse is a cluster of nearly connected rf pulses which emulate the
effect of a simple spin operator with robust response over common experimental
imperfections. Composite pulses are mainly used for improving broadband de-
coupling, population inversion, coherence transfer and in nuclear overhauser effect
experiments. Composite operator is a generalized idea where a basic operator
(such as rotation or evolution of zz coupling) is made robust against common
experimental errors (such as inhomogeneity / miscalibration of rf power or er-
ror in evaluation of zz coupling strength) by using a sequence of basic operators
available for the system. Using Genetic Algorithm optimization, we have designed
and experimentally verified following composite operators, (i) broadband rotation
pulses, (ii) rf inhomogeneity compensated rotation pulses and (iii) zz evolution
operator with robust response over a range of zz coupling strengths (Chapter 5).
We also performed rf inhomogeneity compensated Controlled NOT gate.
Extending Genetic Algorithm optimization in classical NMR applications, we have
improved the quantitative refocused constant-time INEPT experiment (Q-INEPT-
CT) of Makela et al. [JMR 204(2010) 124-130] with various optimization con-
straints. The improved ‘average polarization transfer’ and ‘min-max difference’
of new delay sets effectively reduces the experimental time by a factor of two
(compared with Q-INEPT-CT, Makela et al.) without compromising on accuracy
(Chapter 6). We also introduced a quantitative spectral editing technique based
on average polarization transfer. These optimized quantitative experiments are
also described in Chapter 6.
Time optimal pulse sequences for popular quantum gates such as, (i) Controlled
Hadamard (C-H) gate, (ii) Controlled-Controlled-NOT (CCNOT) Gate and (iii)
Controlled SWAP (C-S) gate are optimized using Genetic Algorithm (Appendix.
A). We also generated optimal sequences for Quantum Counter circuits, Quan-
tum Probability Splitter circuits and efficient creation of three spin W state. We
have developed a Matlab package based on GA optimization for three spin target
operator to pulse sequence generator. The package is named as UOD (Unitary
Operator Decomposition) is explained with an example of Controlled SWAP gate
in Appendix. B.
An algorithm based on quantum phase estimation, which discriminates quantum
states non-destructively within a set of arbitrary orthogonal states, is described
and experimentally verified by a NMR quantum information processor (Appendix.
C). The procedure is scalable and can be applied to any set of orthogonal states.
Scalability is demonstrated through Matlab simulation.
Acknowledgements
Completion of this doctoral dissertation was possible with the support of several
people. I would like to express my sincere gratitude to all of them. Foremost, I
would like to express my sincere gratitude to my advisor Prof. Anil Kumar, who
has supported me throughout my thesis with his patience, immense knowledge
and for allowing me to work in my own way. His guidance helped me in all the
time of research and writing of this thesis. He has always made himself available
to clarify my doubts despite his busy schedules. I would like to thank Prof. Rahul
Pandit for his inspirational advises and motivation as my research supervisor. One
could not wish for a friendlier supervisor and I am blessed with two.
I thank Prof. K. V. Ramanathan for his beautiful way of introducing the world
of Nuclear Magnetic Resonance and for numerous fruitful discussions. I would
like to thank Prof. N. Suryaprakash for introductory lectures on NMR. He is ever
smiling and always ready to help. I humbly acknowledge Dr. S. Raghothama and
Dr. Athreya for their basic NMR courses, which helped me a lot in understanding
the NMR methodologies. I thank Mr. P. T. Wilson for helping in solving problems
with NMR spectrometers.
I thank Prof. Apoorva Patel for his introductory courses on Quantum Computa-
tion and clarifying the concepts in Quantum world. I thank Prof. P. C. Mathias
and Vincent for the efficient maintenance of the spectrometers which helped a lot
in saving time. I thank Chandru for his help in spectrometer problems.
I would like to thank T. S. Mahesh for constructive criticisms and valuable sug-
gestions. I would like to acknowledge Koteswar, my only lab mate for his support
and encouragements. I like to thank Krishnarjuna for many focused discussions
and encouragements.
In my daily work I have been blessed with a friendly and cheerful group of lab
mates. I gratefully acknowledge my seniors Avik Mithra, Bibhuthi, Jharana,
Jayanthi, Nithin Lobo and Uday and Sankeerth for their help in understand-
ing NMR. I thank Nilamoni, Monalisa, Garima, Shibdas Roy, Dr. Bolanath, U.
V. Reddy, Subbareddy, Rajesh Sonti, Thirupathi, Kowsalya, Sudheer, Shivanand,
Sachin, Lokesh, Somnath, Pratik, Srinu, Vasu,Venkat, Santhosh and Durgesh for
numerous academic and non-academic discussions I had with them and for making
my stay here memorable.
xiii
I would like to thank Raghav G. Mavinkurve, an all-rounder helping hand, through-
out my Ph.D life. His way of handling problem is an inspiration to me all the time.
I would like to thank all my friends Alex, Akhil, Arunbabu, Ben, Naveen (Yeshu),
Pradeep, Rajesh, Ranjith, Sajesh, Sanjay, Simil, Vineeth, Vivek, Aneesh, Anoop,
Eby and Joji for many unforgettable moments I shared with them.
Past and present project assistants of NMR Research Center are also gratefully
acknowledged for their smooth running of the spectrometers. Thanks to all Physics
and NRC office staffs, who have helped me effortlessly get through the official
formalities.
I would like to thank Biju sir and Mathew sir (BSc teachers) for their constant
support and encouragement.
I wish to thank my parents, sister, brother in-law and nephews (Kannan & Kunju)
for their moral support and love. Their love was my driving force during my Ph.D
life. I owe them everything and wish I could show them just how much I love and
appreciate them.
Finally but not the least I would like to acknowledge Department of Physics and
NMR Research Centre for providing the support and equipment I have needed to
produce and complete my thesis. UGC and CQIQC are acknowledged for financial
support.
Contents
Declaration of Authorship iii
Publications based on this Thesis vii
Abstract ix
Acknowledgements xiii
List of Figures xix
List of Tables xxiii
Abbreviations xxv
Physical Constants xxvii
1 Introduction 1
1.1 Nuclear Magnetic Resonance . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Interactions in NMR . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Thermal Equilibrium state in NMR (ρeq) . . . . . . . . . . . 6
1.2 Quantum Information Processing (QIP) . . . . . . . . . . . . . . . 7
1.2.1 NMR Quantum Information Processing . . . . . . . . . . . . 10
1.2.1.1 Pseudo Pure State Creation . . . . . . . . . . . . . 11
1.2.1.2 Quantum State Tomography (QST) . . . . . . . . 12
1.3 Genetic Algorithm optimization . . . . . . . . . . . . . . . . . . . . 13
1.3.1 Population Initialization . . . . . . . . . . . . . . . . . . . . 17
1.3.2 Selection Operators . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.3 Reproduction Operators . . . . . . . . . . . . . . . . . . . . 18
1.3.3.1 Crossover Operators: . . . . . . . . . . . . . . . . . 18
1.3.3.2 Mutation Operators . . . . . . . . . . . . . . . . . 19
2 Genetic Algorithm Optimization in NMR Quantum InformationProcessing 23
xv
Contents xvi
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Genetic Algorithm for NMR Pulse Sequence Generation . . . . . . . 24
2.2.1 Representation Scheme . . . . . . . . . . . . . . . . . . . . . 24
2.2.2 Fitness Function . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Two qubit homonuclear case . . . . . . . . . . . . . . . . . . . . . . 26
2.3.1 Operator Optimization . . . . . . . . . . . . . . . . . . . . . 26
2.3.1.1 Single Qubit Rotations using non-selective pulses . 27
2.3.1.2 Controlled NOT gate . . . . . . . . . . . . . . . . . 28
2.3.2 State to State Optimization . . . . . . . . . . . . . . . . . . 30
2.3.2.1 Pseudo-Pure State Creation . . . . . . . . . . . . . 30
2.3.2.2 Creation of Bell States Directly from the ThermalState . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3 Quantum Simulation Using Fidelity Profile Optimization 37
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Theory - Fidelity Profile Optimization (FPO) . . . . . . . . . . . . 39
3.2.1 First Order FPOs . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.1.1 Controlled-Phase gate (C-φ) . . . . . . . . . . . . 41
3.2.1.2 Controlled-Controlled-Phase gate (CC-φ) . . . . . 41
3.2.1.3 Quantum Simulation of three spin Heisenberg XYHamiltonian . . . . . . . . . . . . . . . . . . . . . . 42
3.2.1.4 Three Spin DM interaction . . . . . . . . . . . . . 43
3.2.1.5 Adiabatic Quantum Evolution . . . . . . . . . . . . 44
3.2.2 Second order FPO . . . . . . . . . . . . . . . . . . . . . . . 46
3.3 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.1 Controlled Phase (C-φ) gate . . . . . . . . . . . . . . . . . . 51
3.3.2 Quantum Simulation of three spin Heisenberg XY Hamiltonian 51
3.3.3 Quantum simulation of DM Hamiltonian and entanglementpreservation . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4 Efficient Creation of NOON States in NMR 57
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.1 NOON State Creation (NSC) operators . . . . . . . . . . . 58
4.2.1.1 Spin chain with nearest neighbor couplings . . . . . 60
4.2.1.2 Spin star topology . . . . . . . . . . . . . . . . . . 61
4.2.2 Disentangling NOON states . . . . . . . . . . . . . . . . . . 62
4.2.3 Robust NOON state creation . . . . . . . . . . . . . . . . . 62
4.3 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5 Composite Operator Design Using Genetic Algorithm 69
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.2 Theory and Experimental . . . . . . . . . . . . . . . . . . . . . . . 70
5.2.1 Representation Scheme . . . . . . . . . . . . . . . . . . . . . 72
Contents xvii
5.2.2 Fitness Function . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2.3 Broadband excitation pulses . . . . . . . . . . . . . . . . . . 74
5.2.4 Broadband Inversion pulses . . . . . . . . . . . . . . . . . . 75
5.2.5 Broadband π/2 pulses . . . . . . . . . . . . . . . . . . . . . 77
5.2.6 Broadband π pulses . . . . . . . . . . . . . . . . . . . . . . . 79
5.2.7 rf inhomogeneity compensated composite pulses . . . . . . . 81
5.2.8 Composite Polarization transfer and Evolution operator . . . 83
5.2.9 RF inhomogeneity compensated Controlled NOT gate . . . . 86
5.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6 Fast and Accurate Quantification Using GA Optimized 1H −13 CRefocused Constant-Time INEPT 89
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2.1 Near uniform and equal response for CH1, CH2 and CH3
transfer functions . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2.2 Near uniform and separate response for CH1, CH2 and CH3
transfer functions . . . . . . . . . . . . . . . . . . . . . . . . 94
6.2.3 Near uniform and separate response for CH2 and CH3 only 94
6.2.4 Spectral editing by delay sets . . . . . . . . . . . . . . . . . 96
6.3 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
A Unitary Decomposition of Quantum Gates Using GA Optimiza-tion 103
A.1 Controlled Hadamard (C-H) . . . . . . . . . . . . . . . . . . . . . . 103
A.2 Controlled-Controlled-NOT (CCNOT) Gate . . . . . . . . . . . . . 104
A.3 Controlled SWAP (C-S) . . . . . . . . . . . . . . . . . . . . . . . . 105
A.4 Quantum Counter circuits . . . . . . . . . . . . . . . . . . . . . . . 106
A.5 W State Creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
A.6 Quantum Probability Splitter (QPS) . . . . . . . . . . . . . . . . . 110
B Matlab UOD Package 113
C Non-Destructive Discrimination of Arbitrary Set of OrthogonalQuantum States by NMR Using Quantum Phase Estimation 139
C.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
C.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
C.2.1 The General Procedure (n-qubit case): . . . . . . . . . . . . 140
C.2.2 Single qubit case: . . . . . . . . . . . . . . . . . . . . . . . . 142
C.2.3 Two qubit case: . . . . . . . . . . . . . . . . . . . . . . . . . 143
C.2.3.1 Special case (α = β = 1√2): . . . . . . . . . . . . . . 145
C.2.4 Bell state discrimination: . . . . . . . . . . . . . . . . . . . . 145
C.2.5 GHZ state discrimination: . . . . . . . . . . . . . . . . . . . 146
C.3 Experimental Implementation by NMR . . . . . . . . . . . . . . . . 147
Contents xviii
C.3.1 Non-Destructive Discrimination of two qubit orthogonal states:147
C.3.1.1 Implementation of Controlled-U1 and U2: . . . . . 149
C.4 Three qubit GHZ state Discrimination using Matlab® Simulation: 151
C.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Bibliography 157
Index 171
List of Figures
1.1 Quantization of Spin 1/2 nucleus in presence of an applied field B0. 2
1.2 Energy Level diagram of a spin 1/2 nucleus in applied field B0. . . . 3
1.3 Bloch Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Quantum circuit representation of Controlled-NOT gate (Eqn. 1.29). 9
1.5 Function profile of g(x1, x2) (Eqn. 1.35). Maximas (mx1, mx2 andmx3) and minimas (mn1) of the function are marked. . . . . . . . . 14
1.6 Mechanism of biological evolution. . . . . . . . . . . . . . . . . . . . 15
1.7 Basic outline of Genetic Algorithm. . . . . . . . . . . . . . . . . . . 17
1.8 Roulette Wheel selection method for a popuation of six individualsA, B, C, D, E, F. The area for each individual is proportional toits fitness value. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.9 Three different types of Children from existing generation to newgeneration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.10 Different Crossover operators. . . . . . . . . . . . . . . . . . . . . . 20
2.1 A two-spin system with chemical shifts ±δ and coupling J . . . . . 26
2.2 Pulse sequence for single qubit rotation. . . . . . . . . . . . . . . . 27
2.3 (a) Chemical structure of 5-Bromofuroic acid, (b) Equilibrium spec-trum, (c) (π/2)y SQR pulse on spin-1. (d) (π/2)y SQR pulse onspin-2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Matlab simulation study of fidelity variation of SQR with differentvalues of J/δ and θ. The special case for 5-Bromofuroic acid (J/δ =0.014) is shown with an arrow in x axis. . . . . . . . . . . . . . . . 30
2.5 (a) Pulse sequence for CNOT gate, (b) Diagonal element tomog-raphy of: (i) equilibrium state, and states obtained after applying(ii) C1NOT2, (iii) C2NOT1, (iv) C1NOT2, (v) C2NOT1, (c) Fi-delity (F ) vs. (J/δ) plot for CNOT gate. . . . . . . . . . . . . . . . 31
2.6 (a) Pulse sequence for PPS creation, (b) Tomography of the diag-onal elements after preparing (i) |00〉 PPS, (ii) |01〉 PPS, (iii) |10〉,and (iv) |11〉 PPS. (c) Theoretical fidelity (F ) vs. (J/δ) plot of PPSgeneration pulse sequence. . . . . . . . . . . . . . . . . . . . . . . . 33
2.7 (a) Pulse sequence for creating the Bell states directly form thethermal equilibrium state, (b) Observation of the singlet state afterapplying the operator U of Eq. 2.9, (c) Density matrix tomographyof the created singlet state. . . . . . . . . . . . . . . . . . . . . . . . 34
2.8 Anti-phase signal decay as a function of interval and fits to a singleexponential decay. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
xix
List of Figures xx
3.1 Fidelity profile of FPO (Eqn. 3.12) of Heisenberg XY interaction(Eqn. 3.11). Fidelities are calculated using Eqn. 3.3. . . . . . . . . 43
3.2 Pulse sequence for experimental simulation of three spin DM inter-action. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3 Pulse Sequence decomposition of Eqn. 3.18. . . . . . . . . . . . . . 45
3.4 Fidelity profile of the Hamiltonian (Eqn. 3.18). . . . . . . . . . . . 45
3.5 Fidelity profile of UOD given in Eqn. 3.24. (Fidelities are calculatedusing Eqn. 3.3.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.6 13C labeled Chloroform, two qubit NMR system, used for the quan-tum simulation of DM interaction (Eqn. 3.32) and controlled phasegate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.7 13C labeled CHFBr2 used for experimental verification of FPO ofHeisenberg XY interaction. . . . . . . . . . . . . . . . . . . . . . . . 49
3.8 Pulse sequece for two spin pesudo pure state creation. . . . . . . . . 51
3.9 Experimental implementation of Controlled Phase (φ) gate for var-ious phase angles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.10 Experimental quantum simulation of Heisenberg XY Hamiltonian. . 52
3.11 Experimental scheme used for entanglement preservation. . . . . . . 54
3.12 (a).Entanglement dynamics of 13C −H system under the Hamilto-nian Eqn. 3.21, (b). Entanglement preservation experiment usingEqn. 3.37. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.1 Quantum circuit for NOON state creation. N qubit system requires(N − 1) Controlled-NOT gates and a Hadamard gate. . . . . . . . . 59
4.2 Pulse sequence for for two spin NOON state creation. . . . . . . . . 61
4.3 (a). Spin ising chain with nearest neighbor couplings. J i,i+1 is thecoupling between ith and (i+ 1)th spin, (b). Spin Star network. AllB spins are coupled with spin A with equal strength and there is nocoupling between B spins, (c). A general multi-spin system (spincloud). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4 (a). 90 composite pulse (five pulse) response for flip angle error,(b). 180 composite pulse (five pulse) response for flip angle error. . 64
4.5 Flip angle and coupling error response of composite evolution (Eqn.4.20). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.6 13C labeled CHFBr2 used for experimental NSC. The spins 1H, 13Cand 19F forms a linear spin chain with NN coupling values JH,Czz =224.5 Hz and JC,Fzz = −310.5 Hz, if the two bond JH,Fzz = 49.7 Hzis refocused during the entire experimental time. The longitudinaland transverse relaxation time – T1, T2 (in seconds)– for 1H, 13Cand 19F are 6.7, 1.4, 1.9, 0.71 and 4.0, 0.7. . . . . . . . . . . . 65
4.7 Pulse sequece for three spin pesudo pure state creation. . . . . . . . 66
List of Figures xxi
4.8 Quantum state tomography of three spin NOON state created in(a). spin chain (Eqn. 4.11) and (b). spin star topology (Eqn.4.13) in combination with composite evolution sequence shown inEqn. 4.20. The labels in XY plane 1,2...8 correponds to thecomputational basis states |000〉, |001〉, ...... |111〉. The observedexperimental fidelity is 95.9 % for spin chain and 96.4 % for spinstar topology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.1 Computational time advantage (γ, Eqn. 5.7) efficiency of effec-tive operator evaluation using ‘CompositePulseOperator ’ and Mat-lab builtin function ‘expm’ [1]. The response is evaluated for a setof randomnly created N pulse composite sequence. (Matlab 7.14,32 bit Ubuntu 12.04 OS with core i5 3.6 GHz Intel processor) . . . 71
5.2 Matlab simulation of three pulse, five pulse and seven pulse broad-band excitation sequences. (State to state optimization with Iz asinitial state and Ix as final state.) . . . . . . . . . . . . . . . . . . . 75
5.3 Experimental proton spectral intensity variation with different off-set values of three pulse, five pulse and seven pulse broadband ex-citation sequences. . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.4 Matlab simulation of three pulse, five pulse and seven pulse broad-band inversion sequences. (State to state optimization with Iz asinitial state and −Iz as final state.) . . . . . . . . . . . . . . . . . . 77
5.5 Experimental proton spectral intensity variation with different off-set values of three pulse, five pulse and seven pulse broadband in-version sequences. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.6 Matlab simulation of three pulse, five pulse and seven pulse broad-band π/2 sequences. . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.7 Experimental proton spectral intensity variation with different off-set values of three pulse, five pulse and seven pulse broadband π/2sequences. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.8 Matlab simulation of three pulse, five pulse and seven pulse broad-band π sequences. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.9 Experimental proton spectral intensity variation with offset valuesof broadband three pulse, five pulse and seven pulse π compositepulses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.10 Matlab simulation of seven pulse and nine pulse composite excita-tion sequences with robust rf inhomogeneity profile. . . . . . . . . . 83
5.11 Matlab simulation of five pulse π/2 and π composite sequences withrobust rf inhomogeneity profile. . . . . . . . . . . . . . . . . . . . . 84
5.12 Experimental proton spectral intensity variation with rf values offive pulse π/2 and seven pulse excitation. . . . . . . . . . . . . . . . 84
5.13 Composite operator sequence for for J compensation. . . . . . . . . 85
5.14 Pulse Sequence for Controlled NOT gate. . . . . . . . . . . . . . . . 86
5.15 Inhomogeneity response profiles of controlled NOT gate. . . . . . . 86
6.1 GAQIC pulse sequence. . . . . . . . . . . . . . . . . . . . . . . . . . 92
List of Figures xxii
6.2 Comparison of polarization transfer response of refocused INEPTand GAQIC experiments. . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3 Polarization transfer responses for quantitative spectral editing de-lay sets. (a). four scan (S4 ), (b). sixteen scan (S16 ) . . . . . . . . 97
6.4 Angle sets (θ = Jτ) for (a) SI1, (b) SI2 and (c) SI3. . . . . . . . . 98
6.5 Samples used for GAQIC experiment (a) 4-Penten 2 ol (P), (b) α-methoxy phenyl acetic acid (M) and (c) 1-2-dimethyl cyclohexane(D). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.6 Equilibrium spectrum of the compound mixture (Fig. 6.5) . . . . . 99
6.7 Experimental implementation of GAQIC spectral editing delay set.(a). 1H decoupled 13C spectra and (b). using the delay set (S4). . . 100
A.1 Pulse sequence for Controlled Hadamard. . . . . . . . . . . . . . . . 104
A.2 Pulse sequence for Toffoli Gate. . . . . . . . . . . . . . . . . . . . . 105
A.3 Controlled-SWAP decomposition using C-NOT gates. . . . . . . . . 106
A.4 Pulse sequence for Controlled SWAP gate. . . . . . . . . . . . . . . 107
A.5 Three qubit quantum counter using CC-NOT and C-NOT Gates. . 107
A.6 Pulse sequence for Two Qubit Counter. . . . . . . . . . . . . . . . . 108
A.7 Pulse sequence for Three Qubit Counter. . . . . . . . . . . . . . . . 109
A.8 Pulse sequence for W state creation. . . . . . . . . . . . . . . . . . 110
A.9 Quantum circuit for probability splitter. . . . . . . . . . . . . . . . 110
A.10 Pulse sequence for Quantum Probability Splitter. . . . . . . . . . . 110
B.1 Controlled SWAP pulse sequence. . . . . . . . . . . . . . . . . . . 138
C.1 The general circuit for non-destructive Quantum State Discrimina-tion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
C.2 (a) Two qubit State discrimination circuit for Experimental im-plementation in three qubit NMR quantum computer, (b) and (c)are splitting of the circuit-(a) into two circuits with single ancillameasurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
C.3 The three qubit NMR sample used for experimental implementation.149
C.4 The pulse sequence for Controlled-U1 and Controlled-U2 operatorsfor two qubit orthogonal states (Eqn. C.9) . . . . . . . . . . . . . . 150
C.5 Ancilla (proton spin) spectra of final state for two qubit state dis-crimination algorithm for (i) |φ1〉, (ii) |φ2〉, (iii) |φ3〉, (iv) |φ4〉states. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
C.6 Density Matrix Tomography of the initial and final states of QSDcircuit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
C.7 Pulse sequence for Controlled operators in GHZ state discrimination.153
C.8 The chemical structure, the chemical shifts and spin-spin couplingof a 13C labelled Crotonic Acid. The four 13C spins act as fourqubits [2]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
C.9 Matlab® simulation results for GHZ state discrimination. . . . . . 155
List of Tables
1.1 Spin quantum numbers of some of the common nuclei. . . . . . . . 2
1.2 Different physical implementation of quantum computer and crudeestimates of coherence time and gate operation time in seconds(from [3]). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Darwinian Evolution and GA optimization problem comparison. . . 15
2.1 Spin selection and phase φ of the SQR is controlled by the phasesφ1 and φ2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 (θ,φ) values for all four CNOT gates. . . . . . . . . . . . . . . . . . 29
2.3 Values of φ and d for Bell state preparation using the pulse sequenceshown in Fig. 2.7(a). . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.1 Min-Max Difference (MMD) and Average Polarization Transfer (APT)of all optimized delay sets and R-INEPT. . . . . . . . . . . . . . . 96
6.2 Experimentally measured relative quantities of compound 2 (RQ21)and compound 3 (RQ22). . . . . . . . . . . . . . . . . . . . . . . . . 100
A.1 Phase modulation values for different Toffoli gates . . . . . . . . . . 105
A.2 Truth table for two qubit quantum counter. . . . . . . . . . . . . . 107
C.1 State of ancilla qubits for different input states for two qubit or-thogonal states. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
C.2 State of ancilla qubits for different input states of Eqn. C.13 andC.19. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
xxiii
Abbreviations
AED Average Experimental Deviation
APT Attached Proton Test / Average Polarization Transfer
CNOT Controlled NOT
DEPT Distortionless Enhancement by Polarization Transfer
DJ Deutsch - Jozsa
DM Dzyaloshinsky-Moriya
FPO Fidelity Profile Optimization
GA Genetic Algorithm
GAQIC Genetic Algorithm optimised Quantitative INEPT - Constant time
GRAPE Gradient Ascent Pulse Engineering
INEPT Insensitive Nuclei Enhanced by Polarization Transfer
LLSS Long Lived Singlet State
LOCC Local Operations and Classical Communication
LS-NMR Liquid State - Nuclear Magnetic Resonance
MMD Min-Max Difference
MRI Magnetic Resonance Imaging
NMR Nuclear Magnetic Resonance
NN Nearest Neighbor
NOE Nuclear Overhauser Effect
NSC NOON State Creation
PEA Phase Estimation Algorithm
PPS Pseudo Pure States
PFG Pulsed Field Gradients
QIP Quantum Information Processing
xxv
Abbreviations xxvi
QSD Quantum State Discrimination
QST Quantum State Tomography
SA Spatial Averaging
SQR Single Qubit Rotation
SMP Strongly Modulated Pulses
SS State to State
SSQB Single Superposed Qubits
SST Spin Star Topology
TA Temporal Averaging
UOD Unitary Operator Decomposition
Physical Constants
Planck constant h = 6.62606957× 1034 Js
reduced Planck constant = 1.054571726× 1034 Js
vacuum permeability µ0 = 4π × 10−7NA−2
Bohr magneton µB = 9.27400968× 10−24JT−1
Avogadro’s number NA, L = 6.02214129× 1023mol−1
Boltzmann constant kB = 1.3806488× 10−23JK−1
elementary charge e = 1.602176565× 10−19C
electron mass me = 9.10938291× 10−31kg
proton mass mp = 1.672621777× 10−27kg
xxvii
To my Parents,
xxix
Chapter 1
Introduction
The heroic age of quantum physics starts from Stern-Gerlach experiment in 1922.
Using a beam of silver atoms, Stern and Gerlach showed that measurable values of
atomic magnetic moments are discrete. Later in 1924, Wolfgang Pauli postulated
the existence of a nuclear spin to explain the hyperfine structure of atomic spectra.
The existence of a nuclear spin was confirmed experimentally by Rabi in 1939.
Rabi’s experiment showed that a beam of hydrogen molecules passing through a
magnetic field absorbs radio frequency of a discrete wavelength. 1943 and 1944
Physics Nobel prizes were awarded to Otto Stern and Isidor Isaac Rabi for these
findings[4].
The first NMR experiment on condensed matter was carried out by Felix Bloch
[5] and Edward Purcell[6], who were jointly awarded 1952 Nobel Prize in Physics
for their discovery. Fourier transform NMR was introduced in 1964 by Prof. R.R.
Ernst (chemistry Nobel Prize 1991 [4]). As a scientific tool, NMR is used for
determining the physical and chemical properties of molecules. This leads to a
wide range of applications in chemistry[7], biomolecular structural and functional
studies [8–12] and in medicine (MRI[13, 14], Functional MRI [15] and Magnetic
Resonance Microscopy [16]). Nuclear magnetic resonance is extremely useful since
it is a non destructive and noninvasive technique, which is very important in study-
ing biological systems and compounds [10]. Acquisition of dynamic information is
one of the remarkable advantages of NMR [17–20]. NMR spectroscopy is also an
important technique for studying time dependent chemical phenomena [21, 22].
1
Chapter 1. Introduction 2
Figure 1.1: Quantization of Spin 1/2 nucleus in presence of an applied fieldB0.
Element 1H 2H 12C 13C 14N 15N 16O 17O 19F 31P 35ClSpin 1
21 0 1
21 1
20 5
212
12
32
Table 1.1: Spin quantum numbers of some of the common nuclei.
1.1 Nuclear Magnetic Resonance
Many atomic nuclei possess an intrinsic form of angular momentum called ‘Spin’.
These spinning charge generates a magnetic field and the resulting spin-magnet
has a magnetic moment (µ) proportional to the spin I. Magnetic moment of these
atomic nuclei are quantized as per the quantization nature of angular momentum
operator, which is 2I + 1 discrete levels for a nuclear spin with spin quantum
number I. In the presence of an applied magnetic field, nuclear magnetic moment
associated with these discrete nuclear spin states experiences different magnetic
potential energies. This difference in energy between the spin states increases
linearly with the strength of the field. Irradiation of the sample with energy
corresponding to the exact spin state separation of a specific set of nuclei causes
excitation of those set of nuclei from the lower energy state to the higher energy
state. For a spin 1/2 nucleus the quantized states are spin up – aligns with the
magnetic field – and one spin down – opposes applied field (Fig. 1.1). Spin
quantum numbers of some common nuclei are given in Tab. 1.1.
1.1.1 Interactions in NMR
Atomic nuclei with non zero angular momentum (~I) behave like tiny magnets,
which interacts with applied or local magnetic fields. This interaction was first
observed by Pieter Zeeman as splitting a spectral line into several components in
the presence of a static magnetic field (1902 Nobel Prize). Magnetic moment of a
Chapter 1. Introduction 3
Figure 1.2: Energy Level diagram of a spin 1/2 nucleus in applied field B0.
nucleus with spin ~I is given by,
µ = γ~I, (1.1)
where γ is the gyromagnetic ratio and = h/2π, h is the Plank’s constant.
For a spin I = 1/2 nuclei, there are 2I + 1 = 2 discrete levels, which corresponds
to parallel and anti-parallel configuration to external magnetic field ~B(Fig. 1.1).
Both configuration experience different magnetic potential energies proportional
to strength of external field (B).
The Zeeman interaction Hamiltonian is given by,
H = −µ. ~B, (1.2)
where ~B is effective magnetic field vector experienced by the spin. For a constant
magnetic field along z direction ( ~B = B0z), Eqn. 1.2 can be written as,
Hz = −γB0Iz = ωIz, (1.3)
where Iz is the z component of ~I and ω is known as Larmor frequency (named
after Sir Joseph Larmor) of the nucleus. The energy level diagram of a spin 1/2
nucleus is shown in Fig. 1.2.
Atomic nuclei are shielded by the electrons that surrounded them. In the presence
of an applied magnetic field (B0), the induced magnetic field opposes the applied
field B0 and is known as diamagnetic shielding. This modifies the Hamiltonian
(Eqn. 1.3) as,
Hz = −γ(1− σ)B0Iz, (1.4)
Chapter 1. Introduction 4
where σ is the isotropic value of shielding tensor. Fast tumbling motion of molecules
in liquid state results in a time averaged Hamiltonian as,
σ =1
3(σxx + σyy + σzz). (1.5)
Depending on the surrounding electron density, the strength of diamagnetic shield-
ing will vary and cause a shift in the Larmor frequency of the nuclei, called Chem-
ical Shift (Ω).
The basic NMR interactions are,
(i) Interaction of spin with external field – Zeeman interaction Hz (Eqn. 1.4),
(ii) interaction between spins – scalar coupling (HJ) and dipolar coupling (HD),
(iii) interaction with electric-field gradient generated by their surroundings – for
spins with ~I > 1/2 – Quadrupolar coupling (HQ) and
(iv) interaction with time varying external rf field (Hrf ).
Hence most general NMR Hamiltonian is given by,
H = Hz +HJ +HD +HQ +Hrf . (1.6)
Scalar coupling or spin-spin coupling Hamiltonian (HJ) is an indirect interaction
mediated through electrons in the chemical bond between nuclei.
HJ = 2π∑l<m
~Il.Jlm.~Im, (1.7)
where Jlm is the spin-spin coupling tensor between lth and mth spins. In isotropic
case (liquid state NMR) only the trace of Jlm survives. Hence the spin-spin inter-
action is given by,
HJ = 2π∑l<m
JlmIl.Im, (1.8)
Under weak coupling approximation, where spin-spin coupling strength (Jlm) <<
chemical shift difference of spin l and m(ω0l − ω0m),
HwJ = 2π
∑l<m
JlmIzlIzm. (1.9)
Chapter 1. Introduction 5
Through space direct interaction of magnetic dipoles give rise the dipolar coupling
interaction, which has the form,
HD = 2π∑l<m
~Il.Dlm.~Im, (1.10)
where Dlm is the dipole tensor. For ~rlm, vector connecting two dipolar coupled
nuclei, l and m, Eqn. 1.10 can be expanded as,
HD =∑l<m
dlm[~Il.~Im − 3(~Il.rlm)(~Im.rlm)
r2lm
], (1.11)
where dlm = µ0γlγm/(4πr3lm).
For spins with angular momentum I > 1/2, interaction with electric field gradient
results in Quadrupolar interaction of the form,
HQ =∑l
~Il.Ql.~Il, (1.12)
where Ql is the quadrupolar coupling tensor. In terms of electric field gradient
tensor (Vl), at the site of spin l, Ql is given by,
Ql =eQl
2Il(2Il − 1)Vl, (1.13)
where Ql is the quadrupole moment of spin l.
In isotropic solution the dipolar and quadrupolar interactions average to zero in
the first order and do not contribute to the system Hamiltonian (Eqn. 1.6). The
interactions can be brought back by orienting the molecules in liquid crystal sol-
vents yielding reduced dipolar and quadrupolar couplings. However in this thesis,
we are dealing only with isotropic solutions.
The dynamics of the spin can be controlled externally using radio frequency pulses.
The spin Hamiltonian of isotropic solution during rf pulse is given by,
H = Hz +HJ +Hrf (t), (1.14)
where Hz = ω0Iz is the Zeeman Hamiltonian (ω0 = −γB0). And
Hrf = ω1Cos(ωrf t+ φ)Ix + Sin(ωrf t+ φ)Iy, (1.15)
Chapter 1. Introduction 6
where ω1 = −γB1 (B1 −→ amplitude of rf), ωrf is the frequency of applied rf
pulse with phase φ.
By a rotating frame transformation, using the operator [14],
U = eiωrf Izt, (1.16)
the rf Hamiltonian (Eqn. 1.15) can simplified to a time independent form, yielding
an effective Hamiltonian (Heff ) in the rotating frame as,
Heff = ΩIz + ω1(IxCosφ+ IySinφ), (1.17)
where Ω is the offset,
Ω = ω0(1− σ)− ωrf . (1.18)
1.1.2 Thermal Equilibrium state in NMR (ρeq)
A quantum spin system left undisturbed for a long time in contact with its sur-
rounding (or lattice) will reach a thermal equilibrium state (ρeq),
ρeq =e−H/kBT
Tr[e−H/kBT ], (1.19)
where kB (= 1.3806488 × 10−23JK−1) is the Boltzmann constant and T is the
lattice temperature. The coherence between any two quantum states in thermal
equilibrium is zero,
ρeqkl = 0, for k 6= l. (1.20)
The distribution of population of molecules in different energy levels of thermal
equilibrium state (diagonal elements of ρeq) follows Boltzmann statistics. At room
temperature and usual B0 fields, kBT becomes much larger than difference between
spin energy eigenvalues of H. In this high temperature approximation (upto first
order) of Eqn. 1.19 can be written as,
ρeq u1− H/kBT
Tr[1− H/kBT ], (1.21)
For an ensemble of N spins, the Eqn. 1.21 can be written as,
ρeq u1
2N(1− H
kBT). (1.22)
Chapter 1. Introduction 7
Substituting the Hamiltonian(H), Eqn. 1.22 can be simplified as,
ρeq u1
2N(1 +
B0
kBT
N∑i=1
γiIiz),
u1
2N(1 + ερdev),
(1.23)
where γi is the gyromagnetic ratio, B0 is the static external magnetic field and
ρdev is the deviation density matrix.
1.2 Quantum Information Processing (QIP)
Processing information stored in quantum states of a system, using quantum me-
chanical resources such as superposition and entanglement, can solve certain prob-
lems much faster than their classical counterparts. This made QIP, an intensively
investigated field in Physics over last few decades. The 2012 Nobel Prize in Physics
was awarded to Serge Haroche and David. J. Wineland “for ground breaking ex-
perimental methods that enable measuring and manipulation of individual quan-
tum systems”.
Quantum information unit in QIP is quantum bit (or qubit). Compared with
classical bit, which can assume only two states (either ‘0’ or ’1’), qubit can have
any superpositions of ‘0’ and ‘1’, yielding infinite number of quantum states. A
general single qubit state can be represented as,
|ψ〉 = Cos(θ/2)|0〉+ e−iφSin(θ/2)|1〉. (1.24)
The state (Eqn. 1.24) can be represented as a point in a unit sphere with position
(θ, φ), known as Bloch sphere(Fig. 1.3). All the points in the surface of Bloch
sphere are pure states and those inside are mixed state. A general single qubit
density matrix (ρ) can be represented as,
ρ =1
2(1 + ~r.~σ), (1.25)
where ~r is Bloch vector and ~σ = σx, σy, σz. The magnitude of Bloch vector
(r = |~r|) is unity for pure states and remains unchanged during unitary evolutions.
For a mixed state Trace(ρ2) < 1 ⇒ r < 1, hence all points inside Bloch sphere
represents mixed state. r = 0 represents a maximally mixed state.
Chapter 1. Introduction 8
Figure 1.3: Bloch Sphere
Quantum gates are unitary operations which can perform reversible computation.
Quantum gate operation transform a quantum state from one point of the Hilbert
space to another one. DiVicenzo in 1995 showed that a set of (i) general Single
Qubit Rotation (SQR) and (ii) a two qubit entangling gate forms a universal set
of gates for quantum computing.
Single Qubit Rotation (SQR) : A single qubit state can be visualised as a point
(θ, φ) on Bloch sphere (Fig. 1.24). Any trace preserving unitary operation move
this point over the surface and can be considered as a rotation of the state vector.
Using Pauli matrices,
σx =
[0 1
1 0
], σy =
[0 −ii 0
], σz =
[1 0
0 −1
], (1.26)
the rotation operators about x, y and z axes (Rx, Ry and Rz) can be written as,
Rx(θ) = e−iθ/2σx =
[Cos(θ/2) −iSin(θ/2)
−iSin(θ/2) Cos(θ/2)
],
Ry(θ) = e−iθ/2σy =
[Cos(θ/2) −Sin(θ/2)
Sin(θ/2) Cos(θ/2)
],
Rz(θ) = e−iθ/2σz =
[e−iθ/2 0
0 eiθ/2
].
(1.27)
Chapter 1. Introduction 9
Figure 1.4: Quantum circuit representation of Controlled-NOT gate (Eqn.1.29).
Important single qubit gates are,
NOT gate , UNOT = e−iπIx = −i
[0 1
1 0
]
Hadamard gate , UH = eiπ/2Iye−iπIx = −i
[0 1
1 0
]
Phase (φ) gate , Uφ = e−iφIz = e−iπ/2Iye−iφIxe−iπ/2Iy = e−iφ/2
[1 0
0 eiφ
] (1.28)
An operation on a system which is controlled by the state of another system is
called Controlled-Operators. In quantum computing these Controlled-Operators
are very important. A prototypical controlled operation is Controlled-NOT (Eqn.
1.29),
UCNOT =
1 0 0 0
0 1 0 0
0 0 0 1
0 0 1 0
(1.29)
When a quantum system in ground state is evolved under a slowly time varying
Hamiltonian, it stays in ground state of the instantaneous Hamiltonian. Adia-
batic quantum computation explores this idea for solving problems like finding
the ground state of a Hamiltonian, which is not easy to find. Recently molecular
ground state energy of Hydrogen is experimentally upto 45 bit accuracy by a two
spin NMR quantum computer [23].
Physical realizations of quantum computer
The basic qualities of a quantum system to be used as quantum information pro-
cessor are,
Chapter 1. Introduction 10
System Coherence time Gate operation timeNuclear spin 10−2 – 102 10−3 – 10−6
Electron spin 10−3 10−7
Ion Trap (In+) 10−1 10−14
Electron - Au 10−8 10−14
Electron - GaAs 10−10 10−13
Quantum dot 10−6 10−9
Optical cavity 10−5 10−14
Microwave cavity 100 10−4
Table 1.2: Different physical implementation of quantum computer and crudeestimates of coherence time and gate operation time in seconds (from [3]).
(i) Robust representation of quantum information as the states of the system,
(ii) Universal set of basic experimental unitary operations,
(iii) Sufficient coherence and gate operation time,
(iv) Ability to initialize the system quantum states for computation,
(v) Output measurement capability.
Many efforts are being made to realize a scalable quantum computer using the
techniques such as trapped ions, optical lattices, superconductor based qubits,
Bose-Einstein condensate based quantum computer [24] and NMR. First com-
mercial quantum computer was built by D WAVE (2011, in collaboration with
Lockheed Martin) based on superconducting flux qubits . In May 2013 Google
and NASA announced Quantum Artificial Intelligence lab with D WAVE’s 512
qubit quantum computer.
A comparison of coherence time and gate times of major candidates in experimen-
tal realization are given in Tab. 1.2 (from [3]).
1.2.1 NMR Quantum Information Processing
Long-lasting coherence and high fidelity controls in nuclear magnetic resonance
(NMR) are ideal for quantum information processing. NMR has become an
important experimental tool for demonstrating quantum algorithms, simulating
quantum systems, and for verifying various tenets of quantum mechanics. Experi-
mental implementation of quantum algorithms (Deutsch-Jozsa algorithm, Grover’s
Chapter 1. Introduction 11
search algorithm and Shor’s algorithm of factorization), testing basic principles of
quantum mechanics (no-hiding theorem [25], Leggett-Garg inequality [26]) and
quantum simulation (hydrogen molecule [23] and system with competing two and
three body interactions [27]) have been performed using Liquid state NMR. NMR
quantum computing uses the spin states of nuclei as qubits. The main limitations
of liquid state NMR quantum computer are,
(i) NMR quantum computation uses an ensemble of quantum systems (molecules),
typically of the size, N ≈ 1018. Thus it is an ensemble computing.
(ii) Creation of pure state is difficult in NMR. Pseudo Pure State (PPS), which
mimics the pure state is the initial state of most of the NMR quantum
information processor.
(iii) Scalability of NMR-QC depends on,
(a) availability of quantum spin system,
(b) robustness of the external controls and
(c) coherence time of the system.
1.2.1.1 Pseudo Pure State Creation
Equilibrium state in NMR is a statistical mixture of all the energy eigen states in
Boltzmann distribution. Cory et al. [28] showed an elegant method for creating
an effective pure state (or Pseudo Pure State - PPS) in NMR. PPS mimics the
pure state based on the idea: NMR experiments are sensitive only to traceless
deviation part of the density matrix.
In terms of product operators, the deviation density matrix (ρdev, Eqn. 1.23) of a
two spin system can be shown as (Eqn. 1.30),
ρdev = γ1I1z + γ2I
2z , (1.30)
where γ1 and γ2 are gyromagnetic ratios of first and second spins. For a homo-
nuclear spin system γ1 = γ2. The relative population of the eigen states |α α〉, |α β〉, |β α〉, |β β〉are 1, 0, 0,−1. Pseudo pure |α α〉 state is represented by,
ραα = I1z + I2
z + 2I1z I
2z . (1.31)
Chapter 1. Introduction 12
and the relative populations are 32,−1
2,−1
2,−1
2 and is equivalent to 2, 0, 0, 0
with an addition of identity matrix with proper weight factor (12). Other two spin
pseudo pure state density matrices are,
ραβ =− I1z + I2
z + 2I1z I
2z .
ρβα =I1z − I2
z + 2I1z I
2z .
ρββ =I1z + I2
z − 2I1z I
2z .
(1.32)
There are many methods for creating PPS in NMR. Most common methods are
(i) temporal averaging, (ii) spatial averaging [28], (iii) logical labelling [29] and
(iv) SALLT [30].
Temporal averaging (TA) uses a set of states with average state behaves like pure
state. TA does not really create a logical pure state, but average result can simulate
the execution of a logical operation.
Spatial Averaging (SA) is the most commonly used method for PPS creation
in NMR [28]. SA is based on dividing the system in spatially averaged sub-
ensembles. These sub-ensembles can be manipulated independently using rf pulses
and gradients. We have used SA method for creating PPS in quantum simulation
experiments described in chapters 2 and 4.
Logical or state labeling was first introduced by Gershenfeld and Chuang [29]. In
this method, for an n qubit system, one qubit is used for label the state, while
the (n − 1) qubits are put in a PPS configuration. Spatially Averaged Logical
Labeling Technique (SALLT) [30], combines spatial averaging and logical labeling
techniques. This method uses one labeling qubit and pseudopure state of any
number of qubits can be created in the subsystem manifolds of the labeling qubit.
SALLT does not scale with number of qubits.
1.2.1.2 Quantum State Tomography (QST)
Readout in NMR is performed by recording the single quantum spectra – equiv-
alent to measuring x and y component of the magnetization or the expectation
value 〈Ix + iIy〉 – which is a function of the final observed density matrix (ρ).
Density matrix - spectrum is a many to one mapping and needs around 2n × 2n
spectra for exact density matrix reconstruction (or quantum state tomography).
QST is a systematic way to measure all the elements of the density matrix ρ. The
Chapter 1. Introduction 13
first NMR tomography was developed by Chuang et al. [31].
Consider a general deviation density matrix of a coupled two spin system,
∆ρ =
x11 x12 + iy12 x13 + iy13 x14 + iy14
x12 − iy12 x22 x23 + iy23 x24 + iy24
x13 − iy13 x23 − iy23 x33 x34 + iy34
x14 − iy14 x24 − iy24 x34 − iy34 x44
(1.33)
∆ρ is a traceless Hermitian ⇒ x11 + x22 + x33 + x44 = 0.
Single quantum coherences of ∆ρ (Eqn. 1.33) for spin I1 and I2 are (x12 +
iy12), (x34 + iy34) and (x13 + iy13), (x24 + iy24) respectively. These ∆ρ12, ∆ρ34,
∆ρ13 and ∆ρ24 are the ‘reading positions’ of the density matrix. By unitary
transformations, any desired element of density matrix can be brought in to these
‘reading positions’. Explicit solution for two spin NMR system is given in [32].
1.3 Genetic Algorithm optimization
“The gene is by far the most sophisticated program around.”
- Bill Gates, Business Week, June 27, 1994.
Optimization is a mathematical procedure of finding the best element from a set
of available alternatives and is central to any problem involving decision making.
Finding optimal parameter values or conditions for most favorable performance
of a function makes it popular in all areas of science and technology. Aided by
rapid growth in computer science, optimization has received enormous attention
in recent years.
The basic optimization problem can be stated as,
minimize or maximize f(x)
subject to x ∈ Ω(1.34)
Chapter 1. Introduction 14
The function f : Rn −→ R, is a real values function and is called the objective
function or cost function. ‘x’ is a vector of ‘n’ independent variables referred to
as ‘decision variable’. The set Ω is a subset of Rn. Out of all possible x (∈ Ω)
vectors, optimization algorithm has to find the ‘best’ x (say xb) vector which will
give the minimal or maximal value – f(xb) – for the function. As an example,
function profile of g(x1, x2) (Eqn. 1.35) – known as ‘peak ’ in Matlab, obtained by
translating and scaling Gaussian distributions – is shown in Fig. 1.5.
g(x1, x2) = 3(1− x1)2 × exp[−x21 − (x2 + 1)2]− 10(x1/5− x3
1 − x52)
×exp[−x21 − x2
2]− 1
3× exp[−(x1 + 1)2 − x2
2](1.35)
In the two dimensional space defined by the vectors x1, x2, function g has many
local maximas (mx1, mx2 and mx3) and minimas (mn1). An optimization algo-
rithm with defined problem, maximize g(x), will search this two dimensional space
for the global maxima point mx1.
Figure 1.5: Function profile of g(x1, x2) (Eqn. 1.35). Maximas (mx1, mx2
and mx3) and minimas (mn1) of the function are marked.
Genetic algorithms (GA) are stochastic global search method [33] based on the
mechanics of natural biological evolution (Fig. 1.6). It was proposed by John
Holland in 1975[34]. GA operates on a population of solutions of a specific prob-
lem by encoding the solutions to a simple chromosome-like data structure, and
applies recombination operators. At each generation, a new population is created
by breeding individuals (selected according to their fitness value) together using
operators borrowed from natural genetics. This process leads to the evolution of
individuals and generate populations that are better suited to their environment.
A comparison of Darwinian evolution and GA evolution is given in Tab. 1.3.
GAs differs substantially from other traditional optimization methods, which are,
Chapter 1. Introduction 15
Darwinian Evolution Genetic AlgorithmEnvironment. −→ Optimization problem.Individuals living in that envi-ronment.
−→ Feasible solutions.
Individual’s degree of adaptationto its surrounding environment.
−→ quality of solutions or fitnessfunction.
A population of organisms(species).
−→ A set of feasible solutions.
Selection, recombination andmutation in natures evolutionaryprocess.
−→ Stochastic operators.
Evolution of populations to suittheir environment.
−→ Iteratively applying a set ofstochastic operators on a set offeasible solutions.
Table 1.3: Darwinian Evolution and GA optimization problem comparison.
Figure 1.6: Mechanism of biological evolution.
(i) Traditional methods generate a single point at each iteration while GA search
a population in parallel.
(ii) Selects the next population by probabilistic transition rules (by using random
number generators).
(iii) Do not require any auxiliary information about the problem (like derivatives
in gradient ascent algorithm).
GAs are attractive in engineering design and applications because they are easy
to use and are likely to find the globally best design or solution, which is superior
to any other design or solution [35]. Genetic Algorithm optimization has found
applications in various fields of physics such as molecular geometry optimization
[36], prediction of ultrahigh-pressure phases of ice [37], optimization of silicon clus-
ters [38], and determination of best-fit potential parameters for a reactive force
field [39]. Genetic algorithm has been used in NMR for designing new experiments
[40], for improving excitation and inversion accuracy of rf pulses [41] and pulse
Chapter 1. Introduction 16
sequence optimization [42].
The first step in GA optimization is defining a good genetic representation that
is expressive and evolvable. Encoding solutions of the problem to individuals in
genetic evolution is performed using this representation scheme and optimality
of these solutions or individuals are defined by fitness function. Flow of genetic
evolution process of the population is mainly controlled by fitness function and is
crucial for any GA optimization. There can be many fitness function for the same
problem and selection is usually based on sensitivity of the function – change in
fitness with change in generated solutions – and computational complexity.
In our work we have explained representation scheme for designing pulse sequence
and for delay optimization in quantitative INEPT . An objective function that
prescribes the optimality of a solution or individual in this representation space is
called fitness function.
Outline of the basic Genetic Algorithm optimization is (Fig. 1.7) given by,
(i) The optimization begins by creating a random initial population of individ-
uals (or valid solutions to the problem).
(ii) Creates a sequence of new populations based on following steps:
(a) Evaluate the fitness of each member of the current population.
(b) Select two Parent individuals from the population according to their
fitness (the better fitness, the bigger chance to be selected).
(c) Some of the individuals in the current population that have lower fit-
ness are chosen as elite. These elite individuals are passed to the next
population (Fig. 1.9).
(d) Produces Children from the Parents. Children are produced either by
making random changes to a single Parent – mutation – or by combining
the vector entries of a pair of Parents – crossover.
(e) Replaces the current population with the Children to form the next
generation.
The algorithm stops when one of the stopping criteria is met.
Chapter 1. Introduction 17
Figure 1.7: Basic outline of Genetic Algorithm.
1.3.1 Population Initialization
Random generation of the individual parameters, which follows the specific con-
straint levels of the problem, is the most common way of creating initial population
in GA optimization. Prior knowledge about the solution properties can be utilized
for creating a fast converging initial population. The size of the population has
to be decided by the user based on the complexity of the problem and available
computational power.
1.3.2 Selection Operators
Selection options specify how the genetic algorithm chooses Parents for the next
generation. Common selection methods are random selection (each individuals
have equal probability for getting selected without any reference to its fitness
value), proportional selection (based on the fitness value of the individual) and
rank-based selection (based on the rank order of the fitness value).
The specific proportional selection functions we have used for optimization are,
Stochastic uniform: This method lays out a line in which each Parent corresponds
to a section of the line of length proportional to its scaled value. The algorithm
moves along the line in steps of equal size. At each step, the algorithm allocates
a Parent from the section it lands on. The first step is a uniform random number
less than the step size.
Roulette: Roulette selection chooses Parents by simulating a roulette wheel, in
which the area of the section of the wheel corresponding to an individual is pro-
portional to the individual’s expectation. The algorithm uses a random number
Chapter 1. Introduction 18
Figure 1.8: Roulette Wheel selection method for a popuation of six individualsA, B, C, D, E, F. The area for each individual is proportional to its fitnessvalue.
Figure 1.9: Three different types of Children from existing generation to newgeneration.
to select one of the sections with a probability equal to its area.
1.3.3 Reproduction Operators
The selected individuals or Parents are combined, using reproduction operators.
These operators are borrowed from natural genetics to generate new generation.
The reproduction operators are designed in such a way that the favorable genetic
material should be preserved throughout generations and at the same time pop-
ulation must have genetic diversity. Children are always feasible with respect to
specific constraints of the problem. One or more most fit individuals are allowed
to propagate through successive generations known as elite Children. This elistic
strategy will make sure an incremental or steady state evolution. The most com-
mon reproduction operators are crossover and mutation (Fig. 1.9).
1.3.3.1 Crossover Operators:
Crossover specifies how pairs of Parents should be combined to form a Child for
the next generation (Fig. 1.9). Crossover function randomly selects an entry of
Chapter 1. Introduction 19
the Child vector and randomly replace with same coordinate entry from one of
the two Parents. The crossover operators used in our optimisation are,
(i) Single point : randomly select one point (n) in the Child vector and replace
the entries before (1 to n) and after (> n + 1) the point from Parent1 and
Parent2 respectively (Fig. 1.10(a)).
(ii) Two point : selects two random points (m and n) in the Child vector. The
two point crossover operator replace the entries of the Child vector 1 to m
and n + 1 to final from Parent1 and remaining entries from Parent2 (Fig.
1.10(b)).
(iii) Scattered : based on a randomly generated entries of a binary array (0 or 1),
entries of the Child vector is selected from Parent1 (if binary element is 0)
or Parent2 (Fig. 1.10(c)).
(iv) Intermediate: Child creation is based on weighted average of the Parents.
The parameter ‘Ratio’ can be used to control the ‘Child ’.
Child = Parent1 + rand×Ratio× (Parent2− Parent1). (1.36)
where rand is a random number.
(v) Heuristic: Heuristic crossover operator creates Child from Parent1 and
Parent2 based on the following formula,
Child = Parent2 +Ratio× (Parent1− Parent2), (1.37)
which creates a Child that lies on the line containing the two Parents, closer
to the Parent with the better fitness value. The ‘Ratio’ can be used to con-
trol the Child generation.
(vi) Arithmetic: creates Child that are the weighted arithmetic mean of Parent1
and Parent2.
1.3.3.2 Mutation Operators
Mutation Operators randomly change the genes of individual Parents and intro-
duce new genetic material. Mutation provides genetic diversity which enable the
Chapter 1. Introduction 20
(a) Single Point Crossover
(b) Two Point Crossover
(c) Scattered Crossover
Figure 1.10: Different Crossover operators.
genetic algorithm to search a broader space. Mutation is usually applied with low
probability to preserve favorable genetic material. The mutation operators used
in our optimisation are,
(i) Gaussian: Adds a random number taken from a Gaussian distribution with
mean 0 to each entry of the Parent vector. The standard deviation of this
distribution is determined by the parameters Scale and Shrink,
(ii) Uniform: Selects a fraction of the vector entries of an individual for mu-
tation, where each entry has a probability rate of being mutated and the
algorithm replaces each selected entry by a random number selected uni-
formly from the range for that entry.
(iii) Adaptive Feasible: Randomly generates directions that are adaptive with
respect to the last successful or unsuccessful generation.
Stopping conditions of GA optimization is mainly implemented using fitness limit
or optimization time or generations or function tolerance. We have used GA
optimisation for,
(i) pulse sequence generation for a given operator / quantum gate,
(ii) quantum state generation (with specified initial and final state),
(iii) generation of broadband and rf composite pulses and
(iv) delay set optimisation for quantitative INEPT (GAQIC). The GA optimi-
sation details such as representation scheme, fitness function, mutation op-
erators, crossover operators, population size, number of generations etc are
different in each case and are explained with details in respective chapters.
Chapter 1. Introduction 21
All the programs are written in Matlab®-7.14 in combination with Matlab’s opti-
mization toolbox. All optimizations are performed in a 3.6 GHz Intel® Core i5
Processor processor with Ubuntu 12.04 operating system (32 bit).
Chapter 2
Genetic Algorithm Optimization
in NMR Quantum Information
Processing
“Hilbert space is a big place”
- Carlton Caves.
2.1 Introduction
Decomposing a unitary operator as a sequence of experimentally preferable oper-
ators is the main task in experimental implementation of a quantum algorithm.
There are several proposals for such decomposition in NMR QIP, such as SMPs
(Strongly Modulated Pulses) by Fortunato et al. [43], GRAPE (Gradient Ascent
Pulse Engineering) by Khaneja et al. [44], and algorithmic approach by Ajoy et
al. [45]. Here we investigate the use of Genetic Algorithm for direct numerical
optimization of pulse sequences, and devise a probabilistic method for perform-
ing universal quantum computation using hard pulses. GA optimization, being a
global search algorithm, yields unitary decompositions that are more general, and
hence can be applied to any spin system with different values of J couplings and
chemical shifts. We also investigate quantum state preparation using GA opti-
mization. For performing non-unitary transformations, we have included Pulsed
23
Chapter 2. Genetic Algorithm Optimization in NMR Computing 24
Field Gradients (PFG) [46] in GA optimization. We demonstrate hard pulse uni-
tary decomposition for preparation of Pseudo-Pure State (PPS) and Long Lived
Singlet State (LLSS) (along with other three Bell states) directly from thermal
equilibrium in a two qubit homonulear NMR system. Section 2.2 describes the
optimization procedure and Sec. 2.3 outlines the experimental implementations.
2.2 Genetic Algorithm for NMR Pulse Sequence
Generation
In liquid state NMR, the system Hamiltonian is composed of the interactions of
nuclear spins with the external magnetic field and with each other. Combining
that with external radio frequency (RF) pulses (with specific frequency, amplitude
and phase), one can simulate any preferred effective Hamiltonian [14]. Hence the
unitary operator decomposition problem in NMR can be treated as an optimiza-
tion problem that gives optimal values of pulse parameters and delay durations.
Optimality is determined here by a proper fitness function, which depends on the
target Hamiltonian or the target state.
We have performed pulse sequence optimization using GA for quantum logic gates
(operator optimization) and quantum state preparation (state-to-state optimiza-
tion) [43]. State-to-state optimization converges faster than operator optimization
(there can be many operators which can perform the same state-to-state transfer).
In the discussion given below, Single Qubit Rotation (SQR) pulses and two-qubit
Controlled-NOT (CNOT) gates are operator optimizations, whereas creation of
PPS and Bell states are state-to-state optimizations.
2.2.1 Representation Scheme
Representation scheme is the method used for encoding the solution of the problem
in individuals undergoing genetic evolution. We have selected the gene as a combi-
nation of (i) an array of pulses that we apply simultaneously on each channel with
arbitrary amplitudes (θ) and phases (φ), and (ii) arbitrary delays (d) between the
pulses. It can be easily shown that repeated application of the above gene forms
the most general pulse sequence in NMR. Let an individual representing a valid
Chapter 2. Genetic Algorithm Optimization in NMR Computing 25
solution have m genes, and let n be the number of channels or spins. Then the
individual can be described as a matrix of size (n+ 1)× 2m as shown in Eq. 2.1.
θ11 φ11 . . θm1 φm1
θ12 φ12 . . θm2 φm2
. . . . . .
. . . . . .
θ1n φ1n . . θmn φmn
d1 0 . . dm 0
(2.1)
This matrix has to be optimized, according to the optimality condition expressed
by the fitness function (Sec. 2.2.2).
To begin with, we guess the number of genes (m) as a number that depends on
the complexity of the problem to be solved. If the fidelity, i.e. the fitness of the
best individual in the population, calculated using the fitness function crosses a
cutoff value of (say > 99%), the optimization program tries to reduce the number
of genes by assigning zero values to a few gene parameters. If the fitness function
does not cross the cutoff value, the program will re-run with more number of
genes. We have carried out the optimization procedure using a population size
of 100 individuals evolving for 1000 generations. All the programs are written in
Matlab® in combination with Matlab’s optimization toolbox.
2.2.2 Fitness Function
A fitness function is a particular type of objective function that describes the op-
timality of a solution or individual. In operator optimization, GA tries to reach
a preferred target unitary operator (Utar) from an initial random guess pulse se-
quence operator (Upul). We have selected the fitness function Fpul to be the pro-
jection of Upul onto Utar,
Fpul = Trace(Upul × U †tar). (2.2)
It is normalized to give the maximum value 1.0 when Upul=Utar.
In state-to-state optimization, the optimization program will run over different
possibilities of Upul, to prepare a preferred target state ρtar from the initial state
Chapter 2. Genetic Algorithm Optimization in NMR Computing 26
Figure 2.1: A two-spin system with chemical shifts ±δ and coupling J
ρin. Then we choose the fitness function to be
Fstate = Trace(Upul × ρin × U−1pul × ρ
†tar). (2.3)
In both cases, the optimization has to maximize the fitness function.
2.3 Two qubit homonuclear case
Consider the two qubit NMR homonuclear system (Fig. 2.1) with chemical shifts
±δ and coupling J . Assuming weak coupling (δ J), the Hamiltonian can be
written as [14],
H = Hcs +HJ = 2πδ(Iz1 − Iz2) + 2πJ(Iz1Iz2). (2.4)
For single qubit rotations in this system, one can use spin selective pulses (low
power, long RF pulses) which will excite a small spectral region around the selected
spin [47]. On the other hand, by using the natural chemical shift difference between
two spins, we show here how to implement SQR with global hard (non-selective)
pulses. Later we extend this method to perform two qubit homonuclear universal
quantum computation using only global hard pulses.
2.3.1 Operator Optimization
Operator optimization deals with pulse sequence generation for quantum logic
gates. Here we look at two essential unitary operators for universal quantum
computation: SQR and the two-qubit CNOT gate.
Chapter 2. Genetic Algorithm Optimization in NMR Computing 27
Figure 2.2: Pulse sequence for single qubit rotation. First two filled pulsesare (π/2). Flip angle of the third pulse is (θ/2) with phase φ. The parenthesesabove each pulse contain the flip angle (first number) and the phase (secondnumber).
2.3.1.1 Single Qubit Rotations using non-selective pulses
For a two qubit homonuclear NMR system, we first consider the case J = 0,
H = Hcs = 2πδ(Iz1 − Iz2). (2.5)
Evolution under such a Hamiltonian creates a relative phase among spins, pro-
portional to the chemical shift difference (2δ) and the evolution time. Combining
this relative phase with global rotation hard pulses, single qubit operations can be
performed.
The target operator for SQR is,
Utar = exp(−iθIφk), (2.6)
where k=1 or 2, θ is the flip angle and φ is the phase. The optimized pulse
sequence for SQR is shown in Fig. 2.2. At the start, we selected m = 3, i.e. three
hard pulses and three delays. The optimized sequence has three hard pulses and
a single delay. The flip angle (θ) of the SQR pulse is determined by the delay and
the flip angle of the third pulse, whereas the phase of SQR (φ) and spin selection
is determined by phases of all the three pulses (Tab. 2.1). It may be pointed out
that single qubit rotation with θ = π is a NOT gate and θ = π/2 is a pseudo
Hadamard gate [30].
Experimental verification of the optimized SQR pulse sequence in 5-Bromofuroic
acid (Fig. 2.3(a)) (in C6D6) is shown in Figs. 2.3(c) and 2.3(d). The total length
of the pulse sequence for the (π/2) SQR pulse is less than 500 µs, whereas the
conventional method (using a selective soft pulse) would need a 2 ms shaped pulse.
Chapter 2. Genetic Algorithm Optimization in NMR Computing 28
Spin to be Excited φ1 φ2
1 (φ− π/2) (φ+ π/2)2 (φ+ π/2) (φ− π/2)
Table 2.1: Spin selection and phase φ of the SQR is controlled by the phasesφ1 and φ2.
This substantial shortening in time can lead to a significant advantage in quantum
circuits.
The above analysis also holds for J 6= 0, so long as γB1 δ, J , except that
introduction of the J coupling dephases the final state and results in a fidelity loss.
The fidelity of the pulse sequence (fitness function Eq. 2.2) is studied using Matlab
simulation (Fig. 2.4) and is > 99.8% for J/δ < 0.1 and θ < π/2. (J/δ < 0.1 is the
limit for weakly coupled spins and here we are dealing with only weakly coupled
spins). For π/2 < θ < π, the theoretical fidelity is still quite good (>99.5%) as
can be seen in Fig. 2.4.
2.3.1.2 Controlled NOT gate
The CNOT gate is an essential component in the construction of a universal quan-
tum computer. Any quantum circuit can be simulated to an arbitrary degree of
precision using a combination of CNOT gates and single qubit rotations [3]. The
target unitary operator for C1NOT2 (control on first qubit and NOT on second
qubit) is
Utar = exp(−i(−π4I +
π
2Ix2 +
π
2Iz1 − πIz1Ix2)). (2.7)
The optimized pulse sequence for CNOT is shown in Fig. 2.5(a), obtained using
the Hamiltonian of Eq. 2.5. All the four CNOT gates can be obtained by tuning
(θ, φ) as shown in Tab. 2.2. The pulse sequence is identical for all the four CNOT
gates except for the angles θ and φ.
The experimental implementation of various CNOT gates in 5-Bromofuroic acid
is illustrated in Fig. 2.5(b). We achieved an average experimental fidelity greater
than 99%.
The theoretical fidelity of the CNOT operation (as per Eq. 2.2) is dependent on
the ratio (J/δ). Matlab simulation of fidelity as a function of (J/δ) for the CNOT
gate pulse sequence is shown in Fig. 2.5(c). We find that fidelity is > 99.99% for
Chapter 2. Genetic Algorithm Optimization in NMR Computing 29
(a)
(b)
(c)
(d)
Figure 2.3: (a) Chemical structure of 5-Bromofuroic acid. Diagonal elementsin the table contain the chemical shifts of protons at 500 MHz and the non-diagonal element represents the J coupling (sample dissolved in C6D6). (b)Equilibrium spectrum. (c) (π/2)y SQR pulse on spin-1. (d) (π/2)y SQR pulseon spin-2. The average experimental fidelity (calculated using the standarddefinition [43] that compares spectral intensities with the equilibrium spectrum)for the SQR pulse is greater than 99%.
Gate θ φC1NOT2 π/4 π/2C1NOT2 π/4 0C2NOT1 3π/4 0C2NOT1 3π/4 π/2
Table 2.2: (θ,φ) values for all four CNOT gates. The notation used is CiNOTj ,denoting control qubit i and target qubit j. Also Ci means that NOT operationacts on the target when the control qubit is in the |0〉 state.
Chapter 2. Genetic Algorithm Optimization in NMR Computing 30
Figure 2.4: Matlab simulation study of fidelity variation of SQR with differentvalues of J/δ and θ. The special case for 5-Bromofuroic acid (J/δ = 0.014) isshown with an arrow in x axis.
(J/δ) = 0.01 and drops to 99.84% for (J/δ) = 0.1. This means that even if one
needs ten CNOT gates in a quantum circuit, the fidelity can exceed 99%.
2.3.2 State to State Optimization
State to state optimization deals with pulse sequence generation for quantum state
preparation. Here, we have added gradient pulses to the optimization procedure,
which enabled us to perform non-unitary transformations. We performed two
important quantum state preparations: pseudo-pure state creation, and Bell state
creation directly from the mixed thermal equilibrium state, using global hard
pulses.
2.3.2.1 Pseudo-Pure State Creation
Quantum information processing by NMR spectroscopy uses PPS to mimic the
evolution and observations on pure states [28]. There are several methods for
creating PPS from thermal equilibrium [28–30]. We closely follow the Spatial
Averaging Method (SAM) of Cory et al.. SAM uses qubit selective pulses, which
in homonuclear qubit systems become soft long pulses. Instead, we obtain a novel
pulse sequence using only non-selective (hard) pulses for a homonuclear two qubit
system. The optimization problem has the thermal equilibrium state ∆ρeq = Iz1 +
Iz2 as the initial state and the ∆ρ00 = Iz1 +Iz2 +2Iz1Iz2 as the target state. SAM
also uses pulse field gradient (PFG) pulses to defocus all transverse magnetization
components and retain only longitudinal components of magnetization. These are
Chapter 2. Genetic Algorithm Optimization in NMR Computing 31
(a)
(b)
(c)
Figure 2.5: (a) Pulse sequence for CNOT gate. Parentheses above each pulsecontain flip angle (first number) and phase (second number). (b) Diagonalelement tomography of: (i) equilibrium state, and states obtained after applying(ii) C1NOT2, (iii) C2NOT1, (iv) C1NOT2, (v) C2NOT1. The labels 1, 2, 3,4 represent the states |00〉, |01〉, |10〉, |11〉. An average experimental fidelitygreater than 99% is observed (calculated according to Ref. [43] from diagonalelements of the density matrix). (c) Theoretical fidelity (F ) vs. (J/δ) plot forCNOT gate.
Chapter 2. Genetic Algorithm Optimization in NMR Computing 32
Bell State φ1 φ2 φ3 d1 d2
|ψ+〉 = 1√2(|00〉+ |11〉) 3π/4 9π/8 3π/4 1/16δ 0
|ψ−〉 = 1√2(|00〉 − |11〉) 3π/4 9π/8 π/4 1/16δ 0
|φ+〉 = 1√2(|01〉+ |10〉) 0 5π/8 3π/4 9/48δ 9/8δ
|φ−〉 = 1√2(|01〉 − |10〉) 0 5π/8 π/4 9/48δ 9/8δ
Table 2.3: Values of φ and d for Bell state preparation using the pulse sequenceshown in Fig. 2.7(a).
the non unitary operations needed to convert a mixed state to a pure/pseudo pure
state.
For easier optimization and experimental implementation, we fixed all the pulses to
be (π/2) and optimized only the delays and pulse phases. The resulting sequence
consists of six (π/2) pulses and one (π) pulse for refocusing the chemical shift (Fig.
2.6(a)). The phase of the (π) pulse can be controlled to achieve either the |00〉PPS or the |11〉 PPS. The other PPS are obtained using a combination sequence
of PPS and a SQR π pulse. The experimental results are shown Fig. 2.6(b). An
average experimental fidelity greater than 99% is obtained for various PPS.
The theoretical fidelity (using Eq. 2.3) of the PPS preparation pulse sequence is
also dependent on the ratio J/δ. Matlab simulation of this fidelity as a function
of J/δ is shown in Fig. 2.6(c). We observe that fidelity is > 99.9% for J/δ < 0.1.
2.3.2.2 Creation of Bell States Directly from the Thermal State
Bell states are maximally entangled two-qubit states (also known as the Einstein-
Podolsky-Rosen states) [48]. They play a crucial role in several applications of
quantum information theory. They have been used for teleportation, dense coding
and entanglement swapping [49–51]. Creation of Bell states using NMR conven-
tionally requires PPS creation + Hadamard gate + CNOT gate, and hence is
demanding [52]. We integrated all these steps in a single pulse sequence, and op-
timized that with GA. Again, we kept all pulse amplitudes to be hard (π/2) and
optimized the pulse phases, delay durations and positioning of PFG pulses. The
optimized pulse sequence (Fig. 2.7(a)) has ten non-selective pulses. The final Bell
state can be selected by controlling the phase of the pulses and delay durations
according to Table 2.3.
Chapter 2. Genetic Algorithm Optimization in NMR Computing 33
(a)
(b)
(c)
Figure 2.6: (a) Pulse sequence for PPS creation. All filled pulses are (π/2),and non-filled one is (π), with phases written above them. ± in phase of (π)pulse determines PPS to be created (|00〉 or |11〉). Shaped pulse along Gz rep-resents a PFG pulse, which defocuses all transverse magnetization componentswhile retaining longitudinal magnetization components [46]. (b) Tomography ofdiagonal elements after preparing (i) |00〉 PPS, (ii) |01〉 PPS, (iii) |10〉, and (iv)|11〉 PPS. The labels 1, 2, 3, 4 represent the states |00〉, |01〉, |10〉, |11〉. Anaverage experimental fidelity greater than 99% is observed (calculated accordingto Ref. [43] from diagonal elements of density matrix). (c) Theoretical fidelity(F ) vs. (J/δ) plot of PPS generation pulse sequence.
Chapter 2. Genetic Algorithm Optimization in NMR Computing 34
(a)
(b) (c)
Figure 2.7: (a) Pulse sequence for creating the Bell states directly form thethermal equilibrium state. All filled pulses are (π/2) and non-filled pulses are(π), with the phases written above them. The shaped pulse along Gz representsa PFG pulse which defocuses all the transverse magnetization components, re-taining longitudinal magnetization components [46]. The values of φ’s and ‘d’sare listed in Table 2.3. (b) Observation of the singlet state after applying theoperator U of Eq. 2.9. (c) Density matrix tomography of the created singletstate . The labels 1, 2, 3, 4 represent the states |00〉, |01〉, |10〉, |11〉.
Experimental preparation of the Bell state |φ−〉 = 1√2(|01〉 − |10〉), also known as
the long-lived singlet state (LLSS), is carried out in 5-Bromofuroic acid with the
Hamiltonian H = I1 · I2 [53, 54]:
ρφ− = (0.25I − Ix1Ix2 − Iy1Iy2 − Iz1Iz2). (2.8)
The experimental results are shown in Fig. 2.7(b) and 2.7(c). Since ρφ− is not
directly observable, we convert the created singlet state into observable single
quantum coherence by applying
U = e−i(π/2)Ix1 × e−i(π/2)Ix2 × e−i(π/4)(Iz1−Iz2). (2.9)
An experimental fidelity of more than of 99% is achieved. The pulse sequence
given in Fig. 2.7(a) is the shortest known pulse sequence for creating pure singlet
state in a two-qubit homonuclear NMR system [55].
The singlet state life time (Ts) was measured by applying WALTZ-16 spin-lock
sequence [56] for a variable time period (0− 20 s). We find it to be 11.2 s (Fig.
2.8), which is longer than T1 = 8.7 s and T2 = 3.8 s. Mahesh et al. have found Ts
Chapter 2. Genetic Algorithm Optimization in NMR Computing 35
Figure 2.8: Anti-phase signal decay as a function of interval and fits to asingle exponential decay. The initial intensity of singlet state is normalizedto one. Observed Singlet state life time is Ts = 11.2 Sec. The system has aT1 = 8.7 Sec and T2 = 3.8 Sec.
is rather sensitive to decoupling sequence used [57] and hence Ts can be improved
with different decoupling sequences.
The SQR and CNOT gates implemented using hard pulses (Sections 2.3.1.1 and
2.3.1.2) are valid for larger qubit systems with homonuclear spin pairs (for example
the 1H − 1H − 19F − 19F system in 2,3-Difluro-6-nitrophenol [47]).
2.4 Conclusion
We have used the global optimization power of Genetic Algorithm for,
(i) efficiently implementing SQR and CNOT gates, and
(ii) creating PPS in homonuclear two-qubit system using only hard pulses.
That demonstrates a method for performing universal quantum computation in
such systems. We also demonstrated the creation of LLSS and Bell states directly
from thermal equilibrium state, with the shortest known pulse sequence. It should
be noted that all the pulse sequences discussed here are generic, independent of
the system and the spectrometer.
Chapter 3
Quantum Simulation Using
Fidelity Profile Optimization
“We can’t solve problems by using the same kind of thinking we used when we
created them”
- Albert Einstein
3.1 Introduction
Algorithms with exponential speedups over classical counterparts [58, 59], simu-
lation of quantum systems [23, 60, 61] and testing basic principles of quantum
mechanics [25, 26] makes quantum Information processing (QIP) and quantum
computation intensively investigated fields of physics over last decade. The idea
of simulating quantum systems in a quantum computer was proposed by Feynman
[62] in 1982 and is one of the most important practical application of the quan-
tum computer. Quantum simulation has the potential to revolutionize physics and
chemistry as demonstrated recently by solving problems like – molecular Hydrogen
simulation [23, 63, 64], calculations of thermal rate constants of chemical reactions
[61] and quantum chemical dynamics [65].
37
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 38
Experimental implementation of a quantum unitary operation / gate (U) requires
decomposition of U in terms of experimental system unitaries (such as exter-
nal control operators and natural Hamiltonian evolution of the experimental sys-
tem). We have explained this Unitary Operator Decomposition in previous chap-
ter (Chapter 2) for NMR quantum information processor. Henceforth we will
use the abbreviation UOD for Unitary Operator Decomposition. Fidelity Pro-
file Optimization (FPO) is a generic version of UOD with variable target unitary
operator. Consider a Hamiltonian, H(x1, x2...xn), with time evolution operator,
U(x1, x2...xn, t). FPO is a systematic way of evaluating the most generic sequence
of experimental unitary operators with x1, x2...xn, t dependence. We classify
FPO on the basis of number of independent variables in the functional representa-
tion of the unitary operator (‘nth order FPO’ – for n independent variables). The
UOD optimization for constant target unitaries (zero independent variable) such
as C-NOT, CC-NOT, C-Hadamard etc are zeroth order FPO, which will give a
point in fidelity space (spanned over the range of Hamiltonian parameters). One
independent variable will give a line in the fidelity space, while a two indepen-
dent variable optimization will give a plane. Thus the order of FPO decides the
dimension of fidelity space.
Long-lasting coherences and high fidelity controls in nuclear magnetic resonance
(NMR) are ideal for quantum information processing. Experimental implementa-
tion of quantum algorithms (Deutsch-Jozsa algorithm, Grover’s search algorithm
and Shor’s algorithm of factorization ), testing basic principles of quantum me-
chanics (no-hiding theorem [25], Leggett-Garg inequality [26]) and quantum sim-
ulation (hydrogen molecule [23] and system with competing two and three body
interactions [27]) have been performed using Liquid state NMR (LS-NMR). LS-
NMR with weak coupling approximation has ‘zz’ interaction between spins (Eqn.
3.1),
Hzz = Jzz(σ1zσ2z), (3.1)
(where Jzz is the interaction strength). Hzz with single qubit rotations (SQR)
Rn(θ, φ) (θ angle along φ axis on nth spin),
Rn(θ, φ) = exp(−iθ/2× [Cosφ σnx + Sinφ σny]), (3.2)
forms universal set of unitary operators for quantum information processing.
Chapter 2, describes the use of GA for optimal design of various one and two qubit
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 39
quantum gates by NMR. Genetic algorithm has been used in NMR for designing
new experiments [40], for improving excitation and inversion accuracy of RF pulses
[41] and pulse sequence optimization [42, 66]. Here we use Genetic algorithm
optimization for solving UOD for generic DM Hamiltonian with Heisenberg-XY
interaction.
Section 3.2 deals with theoretical discussion of FPO followed by experimental
implementation in Section 3.3.
3.2 Theory - Fidelity Profile Optimization (FPO)
Fidelity Profile Optimization is performed by a systematic evaluation of Hamilto-
nian parameter dependence in the decomposed experimental unitaries. We have
used multiple optimizations using Genetic Algorithm [33], where complexity of
optimization depends on Hilbert space dimension as well as order of the FPO.
Zeroth order FPOs – Controlled NOT and Hadamard gates – were explained in
Chapter 2. The details of the specific FPO - GA optimization used here are,
Representation Scheme: A valid UOD is represented as a matrix of size (n+1)×2m
(given in Chapter 2). Here ‘n’ is the number of qubits and ‘m’ varies with the
complexity of the operator. So the problem is to find an optimized matrix where
optimality condition is posed by a fitness function.
Fitness Function: We have selected the fitness function Fuod to be the projection
of Uuod onto Utar,
Fuod = Tr(Uuod × U †tar). (3.3)
GA optimization procedure of a first order FPO, Controlled-φ gate (Eqn. 3.7), is
explained in following steps as an illustrative example.
Step.i : Perform zeroth order FPO of Controlled-φ gate for a typical value of φ
(say φ = π/2) (as explained in Chapter 2).
Step.ii : Starting from the individual with maximum fidelity, from Step.i, we will
identify the φ dependent UOD parameters (individual array elements –
amplitudes, phases and delays) in this step. This is accomplished by iter-
atively checking φ dependence of each UOD parameters. For Controlled-φ
gate, there are three φ dependent UOD parameters.
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 40
Step.iii : Identify functional dependence of UOD parameters, by substituting each
UOD parameters by basic functions such as polynomials, exponential
functions, Fourier functions etc. and solve for the functional constants by
performing GA optimization again. Complexity of functional dependence
vary systematically in such a way that functions with minimum number
constants can be obtained.
For Controlled-φ gate, substituting each UOD parameters with degree one poly-
nomial, aiφ+bi and performing GA optimization, constants (ai, bi) are obtained
as (1/2, 0), (1/4, π/2), (-1/4, π).
Second order FPO is also performed by the same procedure described above by
independently checking each FPO parameters for its Hamiltonian variable depen-
dency. Nature of the functional dependence (single variable or two variable) of
FPO parameters is also important in this case. Single variable dependent UOD
parameters can be identified by substituting basic functions as described above for
Controlled-φ case. Two variable dependent UOD parameters functions are iden-
tified by substituting product of basic functions. A direct generalization to this is
possible for n variable dependent unitary operators (or nth order FPO).
All FPOs discussed here are based on basic two qubit interaction as Hsys = Hzz
(weak coupling in nuclear magnetic resonance; (Eqn. 3.1) with evolution unitary,
Uzz(θ) = exp(−i θ/2 σzσz). (3.4)
However it can be generalized to almost any system interaction by term isolation
procedure by Bremner et al. [67] and Hill et al. [68]. As an example consider the
case,
Hsys = J(σxσx + σyσy + σzσz). (3.5)
The Hzz terms can be isolated from Eqn. 3.5 by (as shown in [68]),
exp(−iJzzσzσzt) = R1(π, z)exp(−iHsyst)R1(π, z)
exp(−iHsyst).(3.6)
All the programs are written in Matlab in combination with Matlab’s optimization
toolbox. The typical time of FPO varies exponentially with number of qubits (since
Hilbert space dimension grows exponentially). A typical first order FPO for three
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 41
spin system (CC-φ gate in Sec. 3.2.1.2) took nearly three hours on a computer
with 3.6 GHz processor (Matlab® 7.14; 32 bit Ubuntu 12.04 OS).
3.2.1 First Order FPOs
Experimental unitary decomposition of single variable dependent quantum oper-
ations such as time evolution unitary of a constant Hamiltonian, controlled phase
gates (Cn-φ) and adiabatic evolution operators can be solved by performing first
order FPO. They yield a single line in the Fidelity profile. Here we investigate
how efficiently we can simulate these operations.
3.2.1.1 Controlled-Phase gate (C-φ)
Phase gate operation on target qubit, controlled by the state of control qubit is
an important operation in basic quantum information circuit. The unitary matrix
for this operation is given by (Eqn. 3.7),
C-φ =
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 e−iφ
. (3.7)
The unitary operator decomposition by first order FPO is obtained as,
C-φ = R2(π, 5π/4) Uzz(φ/2) R2(π, (2π + φ)/4)
R1(π, π − φ/4) R1(π, 0).(3.8)
where Rn(θ1, θ2) and Uzz(φ) are given in Eqn. 3.2 and 3.4 respectively. The unitary
operator decomposition (Eqn. 3.8) has three φ dependent UOD parameters and
is exact (100% fidelity for all values of φ). We have performed experimental
verification of this in a two spin NMR system (Sec. 3.3).
3.2.1.2 Controlled-Controlled-Phase gate (CC-φ)
Controlled-Controlled-Phase gate applies a phase φ to the target when both control
qubits are in |1〉 state. The experimental unitary decomposition by first order FPO
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 42
is obtained as,
CC-φ = R1(π, 0) R2(π, φ/8) R3(π, φ/8) U13zz (φ/4)
R1(π/2, π − φ/4) U12zz (π/2) R1(π/2, 3π/2− φ/4)
U13zz (π − φ/4) R1(π/2, π/2) R3(π, π/2)U12
zz (π/2)
U23zz (2π − φ/4) R1(π/2, 0) R2(π, 0).
(3.9)
The decomposition (Eqn. 3.9) is exact (100% fidelity for all values of φ) has seven
φ dependent UOD parameters.
For C-φ and CC-φ gates, the φ dependence of the experimental unitary decom-
position are obtained as first order polynomials.
3.2.1.3 Quantum Simulation of three spin Heisenberg XY Hamiltonian
Heisenberg XY interaction (Eqn. 3.10) is a naturally occurring Hamiltonian in
various spin systems [69–71].
HXY = J(I1xI
2x + I1
yI2y + I2
xI3x + I2
yI3y ). (3.10)
Where J is the coupling strength. Spin product operators (Iα) and Pauli matrices
(σα) are related by Iα = σα/2 (where α = x, y, z). The time (t) evolution
operator of Heisenberg XY Hamiltonian is given by,
UXY = exp(−itHXY ) = exp(−iτ(I1xI
2x + I1
yI2y + I2
xI3x + I2
yI3y )). (3.11)
UXY is a single variable (τ = Jt) dependent three qubit unitary operator. The
UOD using first order FPO is obtained as (Eqn. 3.12),
UXY = U1 Uzz U2 Uzz U3 Uzz U2 Uzz U1,
U1 = R1(π/2, 0) R2(π/2, π/2) R3(π/2, 0),
U2 = R1(π/2, 0) R2(π/2, 0) R3(π/2, 0),
U3 = R1(θ1, θ2) R2(π, π/2) R3(π/2, 0),
Uzz = exp(−i π/2 (2I1z I
2z + 2I2
z I3z )),
(3.12)
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 43
0 4 8 1299.8
99.9
100
τ F
ide
lity
(%)
Figure 3.1: Fidelity profile of FPO (Eqn. 3.12) of Heisenberg XY interaction(Eqn. 3.11). Fidelities are calculated using Eqn. 3.3.
where,
θ1 = 5.171× 10−5τ 6 − 0.001392τ 5 + 0.01259τ 4
−0.03942τ 3 + 0.07208τ 2 − 0.5488τ + 3.152,
θ2 = 0.0002799τ 5 − 0.006301τ 4 + 0.04469τ 3
−0.09354τ 2 + 0.3353τ − 0.02239.
(3.13)
The fidelity profile (Fig. 3.1) for the decomposition Eqn. 3.12 has a least fidelity
> 99.95% for one complete period (τ=0 to 8.9) of UXY . While the Cn-φ FPOs
discussed in Sec. 3.2.1.1 and 3.2.1.2 are exact (100% fidelity), the FPO in the
present case is not exact, the fidelity here can be improved (if needed) by sub-
stituting more complex polynomial functions for θ1 and θ2. The decomposition
shown here is time symmetric – can be applied in the order left to right (increasing
time) or right to left – and uses less number of pulses (around 20 hard pulses) and
delays compared to the method by Zhang et al (around 40 hard pulses) [72].
3.2.1.4 Three Spin DM interaction
Dzyaloshinskii-Moriya (DM) interaction is an anisotropic antisymmetric exchange
interaction arising from spin-orbit coupling [73, 74]. Interaction of this type was
proposed by Dzyaloshinskii to explain the weak ferromagnetism of antiferromag-
netic crystals (α-Fe2O3, MnCO3)[73]. DM interaction is crucial in the description
of many antiferromagnetic systems [75–77].
Three spin DM interaction is given by (Eqn. 3.14) [74] ,
HDM =D
4(σ1
xσ2y − σ1
yσ2x + σ2
xσ3y − σ2
yσ3x), (3.14)
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 44
Figure 3.2: Pulse sequence for experimental simulation of three spin DMinteraction.
where D is the strength of the DM interaction. The time (t) evolution operator
of Dzyaloshinskii-Moriya Hamiltonian is given by,
UXY = exp(−itHDM) = exp(−iτ4
(σ1xσ
2y − σ1
yσ2x + σ2
xσ3y − σ2
yσ3x)). (3.15)
UDM is a single variable (τ = Jt) dependent three qubit unitary operator. The
UOD using first order FPO is shown in Fig. 3.2. The functional dependence of τ
on the rotation angles and phases (θ1 and θ2) of the pulse sequence (Fig. 3.2) is
obtained as,
θ1 = 2.527× 10−4τ 5 − 5.616× 10−3τ 4 + 0.03838τ 3
−0.068τ 2 + 0.2928τ + 1.568,
θ2 = −4.701× 10−5τ 6 + 1.253× 10−3τ 5 − 0.01101τ 4
+0.03076τ 3 − 0.04865τ 2 + 0.5207τ.
(3.16)
The decomposition shown here is time symmetric and uses only nearest neighbor
couplings.
3.2.1.5 Adiabatic Quantum Evolution
A typical example for first order FPO is the operators for Adiabatic evolution,
where the system is evolved under a slowly time varying Hamiltonian [32]. Adia-
batic Evolution Hamiltonian is given by,
H(S) = (1− S)Hi + SHf ,
where 0 6 S 6 1, Hi is the initial Hamiltonian and Hf is the final Hamiltonian.
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 45
Figure 3.3: Pulse Sequence decomposition of Eqn. 3.18.
0 0.25 0.5 0.75 199.99
99.995
100
S
Fid
elit
y (%
)
Figure 3.4: Fidelity profile of the Hamiltonian (Eqn. 3.18).
The unitary operator – U(S), Eqn. 3.17 – is a single variable unitary and hence
first order FPO can be used for solving adiabatic unitary decomposition.
U(S) = e−iH(S) (3.17)
Consider a case with Hi = (I1x + I2
x) and Hf = I1z I
2z , the adiabatic Hamiltonian
(HA) is,
HA = S(I1x + I2
x) + (1− S)I1z I
2z . (3.18)
The S dependent pulse sequence obtained using first order FPO is given in Fig.
3.3. The S dependent rotation angles and phases (θ1 to 4) are obtained as (Eqn.
3.19),
θ1 = 0.5904 S3 − 1.606 S2 + 1.544 S + 0.2609
θ2 = 0.1569 S3 − 0.1376 S2 + 0.7638 S + 0.002335
θ3 = 1.026 S3 − 2.839 S2 + 4.482 S + 0.4788
θ4 = −0.274 S3 + 0.1716 S2 + 0.1059 S − 0.003411
(3.19)
The decomposition is time symmetric and the fidelity is more than 99.99% for all
values of S.
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 46
3.2.2 Second order FPO
Second order FPO deals with two variable dependent quantum Unitary operations
and hence the fidelity profile is a plane. Here we explain this with an example of
simulation of Dzyaloshinskii-Moriya (DM) interaction (Sec. 3.2.1.4) in the pres-
ence of Heisenberg XY interaction, where the relative interaction strength (γ) and
time (τ) are the system parameters (Eqn. 3.20).
HDM = D(σ1xσ2y − σ1yσ2x), (3.20)
where D represents the strength of DM interaction. Here we present a generic
unitary operator decomposition to simulate the Hamiltonian – DM interaction in
the presence of Heisenberg XY interaction – in a two qubit system (Eqn. 3.21).
H(J,D) = J(σ1xσ2x + σ1yσ2y) +D(σ1xσ2y − σ1yσ2x), (3.21)
The unitary evolution operator of H(J,D) (Eqn. 3.21) is given by,
U(D, J, t) = exp(−iH(J,D)× t), (3.22)
Without losing generality, Eqn. 3.22 can be written as,
U(γ, τ) = exp(−i[(σ1xσ2x + σ1yσ2y)+
γ(σ1xσ2y − σ1yσ2x)] τ),(3.23)
where γ represents the relative ratio of interaction strengths (γ = D/J) and a
dimensionless quantity τ = J × t.As the first step, we performed FPO of U(γ, τ) with following assumptions – (a).
the range of τ is from 0 to 15 (this includes one complete period of 12.59; please
see Eqn. 3.27) , (b). the range of γ is from 0 to 1 and (c). The system Hamiltonian
is Hzz (Eqn. 3.1).
The optimized UOD (Eqn. 3.24) has seven SQRs and two system Hamiltonian
evolutions.
U(γ, τ) = R1(π2,−π
2) R1(π
2, θ2) R2(π, π) Uzz(
π4)
R1(θ1, θ2 + π2) R2(π − θ1, 0) Uzz(
π4)
R1(π2, θ2 + π) R2(π
2, π
2),
(3.24)
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 47
Figure 3.5: Fidelity profile of UOD given in Eqn. 3.24. (Fidelities are calcu-lated using Eqn. 3.3.)
where θ1 and θ2 (given in Eqn. 3.25) impart γ and τ dependence to UOD and Uzz
is given by Eqn. 3.4.
θ1 = [0.8423− 0.3455 Cos(1.117 γ)+
0.01806 Sin(1.117γ)] τ,
θ2 = 1.345 exp(−0.8731γ) + 1.796.
(3.25)
The fidelity profile of UOD is shown in Fig. 3.5. The minimum point in fidelity
profile is greater than 99.99 %. It should be noted that, the total length of UOD
(Eqn. 3.24) is invariant under γ or τ (with the assumption that all the SQRs are
instantaneous).
For generalizing the assumption on τ , we solved Eqn. 3.26 numerically and find
the period P(γ) of U(γ,τ) (Eqn. 3.27).
H(γ, τ + n× P (γ)) = H(γ, τ). (3.26)
where n is any positive integer.
P (γ) = 3.008 γ3 − 6.627 γ2 − 0.1498 γ + 12.59. (3.27)
Eqn. 3.27 has a maximum value of 12.59 at γ=0. Since the maximum value of
period is less than 15 (FOP performed till τ=15), UOD (Eqn. 3.24) can be used
for any value of τ . Same argument can be used for extending the range of τ to
−∞.
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 48
In order to incorporate the range of γ from 0 to ∞, we performed FPO for the
following operator (Eqn. 3.28),
U(Υ,∆) = exp(−i[Υ (σ1xσ2x + σ1yσ2y)+
(σ1xσ2y − σ1yσ2x)] ∆),(3.28)
where Υ and ∆ are dimensionless variables with range 0 to 1 (for Υ ) and 0 to 15
(for ∆). The optimized unitary decomposition using second order FPO is obtained
as,
U(Υ,∆) = R1(π2, π
2) R2(π
2, θ3) Uzz(
π4) R1(θ2 + θ3, 0)
R2(θ1, θ4) Uzz(π4) R1(π
2, π
2) R2(π
2, θ3),
(3.29)
where θ1 · · · θ4 (Eqn. 3.30) impart Υ and ∆ dependence to UOD.
θ = [0.09812 exp(−2.42Υ ) + 0.4023 exp(0.5524Υ )]∆,
θ1 = −θ + 3.142,
θ2 = θ − [1.242 exp(−0.9617Υ ) + 0.3546 exp(−0.1145Υ )],
θ3 = 1.259 exp(−0.957Υ ) + 3.479 exp(−0.0087Υ ),
θ4 = 1.256 exp(−0.959Υ ) + 1.912 exp(−0.0166Υ ).
(3.30)
Eqn. 3.28 satisfy the same periodicity relation as shown in Eqn. 3.27 and hence
can use the same reasoning for extending ∆ range from 0 to +∞.
For γ > 1, Eqn. 3.23 can be written as,
U(γ′, τ ′) = exp(−i[γ′(σ1xσ2x + σ1yσ2y)+
(σ1xσ2y − σ1yσ2x)] τ′),
(3.31)
where γ′ = 1/γ and τ ′ = γ × τ .
Eqn. 3.28 and Eqn. 3.31 are equivalent and hence UOD for Eqn. 3.23 can be
shown as,
U(γ, τ) =
Eqn. 3.23 if γ 6 1
Eqn. 3.28 if γ > 1(3.32)
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 49
Figure 3.6: 13C labeled Chloroform, two qubit NMR system, used for thequantum simulation of DM interaction (Eqn. 3.32) and controlled phase gate.13C and 1H act as qubits with zz interaction (Jzz=215.1 Hz) between them.The longitudinal and transverse relaxation time – T1, T2 (in seconds)– for 1Hand 13C are 21, 3.4 and 16, 0.29. The observed FID decay rates (T ∗2 ) are720 ms (1H) and 175 ms (13C).
Figure 3.7: 13C labeled CHFBr2 used for experimental verification of FPOof Heisenberg XY interaction (Eqn. 3.12). 1H , 13C and 19F forms a threequbit system with Hzz interaction. The longitudinal and transverse relaxationtime – T1, T2 (in seconds)– for 1H, 13C and 19F are 6.7, 1.4, 1.9, 0.71 and4.0, 0.7.
Combining all above steps forms most generic UOD of the Hamiltonian – DM
interaction in the presence of Heisenberg XY interaction.
3.3 Experimental
We have performed experimental verification of the FPOs – Controlled Phase gate,
simulation of Heisenberg XY interaction and DM interaction – in two and three
spin LS-NMR quantum information processor. Two spin LS-NMR system – 13C
labelled 13CHCl3 (dissolved in Acetone-D6) (Fig. 3.6) with 13C and 1H spins as
qubits (coupling interaction - Hzz; Eqn. 3.1) is used for simulating Controlled
phase gate (Eqn. 3.8) and DM interaction (Eqn. 3.32). Three spin LS-NMR
system – 13C labeled CHFBr2 (Fig. 3.7) is used for simulating Heisenberg XY
interaction (Eqn. 3.12). All experiments have been performed using Bruker AV-
500 NMR spectrometer at an ambient temperature of 300 K.
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 50
Experimental procedure for quantum computing in NMR is as follows,
(i) Initialization of quantum system – initial state creation is necessary for im-
plementing any quantum algorithm. In this chapter, we are initializing the
quantum system to computational basis state (pseudo pure state – [28–30])
and pure transverse magnetization. DM interaction simulation (Sec. 3.3.3)
used two spin (13C labelled CHCl3) pseudo pure state,
(ii) processing the initial state by evolving under different average Hamiltonian
[14] and
(iii) read-out by quantum state tomography [32].
There are several methods for creating pseudo pure states (PPS) in NMR from
equilibrium state [28–30]. We have utilized the spatial averaging technique [28] for
creating |0〉⊗n pseudo pure states as described in [78]. The pulse sequence for two
spin PPS creation is given in Fig. 3.8. This two spin PPS pulse sequence is used
for the experimental quantum simulation of DM interaction and entanglement
preservation (Sec. 3.3.3). A basic introduction to PPS and quantum state tomog-
raphy is explained in Introduction chapter (Chapter 1). Experimental verification
of controlled phase (Sec. 3.3.1) and simulation of Heisenberg XY interaction (Sec.
3.3.2) is performed on pure Y magnetization as the initial state. The quantum
state tomography is performed for the final state [78] read-out (Chapter 1).
External rf pulses of specific frequency, amplitude(A), phase and duration(t) are
used for implementing Rn(θ, φ) in NMR. Here A and t controls θ, rotation angle
of the pulse. zz unitary operator, Uzz(Θ) is implemented using time evolution (for
a time, t = Θ/πJ) of the spin states in NMR weakly coupling Hamiltonian (Eqn.
3.1) [14].
For comparing a series of experimental Ei=1...n and the theoretical Ti=1...nvalues, we have used ‘Average Experimental Deviation’ (AED), defined as (Eqn.
3.33),
AED =n∑i=1
|Ei − Ti||Ti|
. (3.33)
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 51
Figure 3.8: Pseudo pure state creation in a proton and carbon two spin sys-tem. Starting from thermal equilibrium, ρeq ∝ 4IHz + ICz , first (π/3)x pulseon proton covert 4IHz to 2(IHz −
√3IHy ). The first gradient (G1) kills the
transverse magnetization 2√
3 IHy and the remaining H magnetization, 2IHzis converted to
√2(IHz − IHy ) by a (π/4)x pulse. Evolution under J coupling
for a time 1/(2J) converts it to√
2(IHz − 2IHx ICz ), which again converted to
(IHz − IHx ) + (2IHx ICz + 2IHz I
Cz ) by (π/4)y pulse. The final gradient pulse (G2)
is applied to kill the transverse magnetization, yeilding the total magentizationas IHz + ICz + 2IHx I
Cz and is |00〉 pps. All other computational basis pps states
can be geneated by using π inversion pulses.
3.3.1 Controlled Phase (C-φ) gate
We have performed experimental verification of Controlled Phase (C-φ) FPO in13C labelled Chloroform CHCl3 (Fig. 3.6). The unitary operator decompositions
for Controlled phase gate is given in Eqn. 3.8.
Starting from initial state, ρin = I1y , controlled phase C-φ gate will modulate the
‘y’ magnetization as shown in Eqn. 3.34.
⟨I1y
⟩= Trace(C-φ× ρin × C-φ†) =
1
2(1 + Cos(φ)). (3.34)
We have experimentally measured this modulation and found it to match well
with theoretical result. Experimental and simulated plots with normalized⟨I1y
⟩intensity are shown in Fig. 3.9. We have observed an AED of 1.2 %, which
confirms the FPO.
3.3.2 Quantum Simulation of three spin Heisenberg XY
Hamiltonian
Experimental verification of the first order XY Hamiltonian FPO (Eqn. 3.12) is
performed in 13C labeled CHFBr2 three spin NMR system (Fig. 3.7). The spins
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 52
Figure 3.9: Experimental implementation of Controlled Phase (φ) gate forvarious phase angles (decompositions in Eqn. 3.8). ‘y’ magnetization –
⟨I1y
⟩(measured by partial tomograph[32] of the final state) – shows a φ modulationas shown. Experiments are performed in 13C labelled Chloroform (CHCl3; Fig.3.6) dissolved in CDCl3.
Figure 3.10: Experimental quantum simulation of Heisenberg XY Hamilto-nian (decompositions in Eqn. 3.12). Experiments are performed in 13C labeledCHFBr2 (Fig. 3.7). Second spin ‘y’ magnetization –
⟨I2y
⟩(measured by par-
tial tomograph[32] of the final state) – shows a periodic modulation with τ asshown.
1H, 13C and 19F forms a linear spin chain with NN coupling values JH,Czz = 224.5
Hz and JC,Fzz = −310.5 Hz (Fig. 3.7), if the two bond JH,Fzz = 49.7 Hz is refocused
during the entire experimental time.
Experimental verification of FPO is performed by using following scheme,
I1y + I2
y + I3y (initial state)
UXY (τ)−−−−→ Measure
τ modulated ‘y’ magnetization⟨I2y
⟩.
Second spin ‘y’ magnetization –⟨I2y
⟩– is modulated by τ as shown in Fig. 3.10.
Experimental results matches the theoretically simulated curve with an AED
(Eqn. 3.33) of 2.2%.
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 53
3.3.3 Quantum simulation of DM Hamiltonian and entan-
glement preservation
Here we study the entanglement dynamics (quantified by concurrence, Eqn. 3.36)
of a Bell state (Eqn. 3.35), evolving in DM + XY Hamiltonian (Eqn. 3.21).
|φ〉− =1√2
(|01〉 − |10〉). (3.35)
For a two qubit system (of qubits A and B), with density matrix ρAB, concurrence
(CAB) is defined as [79],
CAB = maxλ1 − λ2 − λ3 − λ4, 0, (3.36)
where λ1....4 are eigenvalue of the matrix ρABρAB with ρAB = (σy⊗σy)ρ∗AB(σy⊗σy).
Using the unitary operator decompositions shown in Eqn. 3.32, we have simulated
the Hamiltonian H(γ, τ) for γ=0.33, 0.66, 0.99. The experiments are performed
with 13C labelled Chloroform (Fig. 3.6). We have utilized spatial averaging tech-
nique [28] for creating |00〉 pseudo pure states [78]. Singlet state is created from
|00〉 state by using a Hadamard gate and controlled NOT gate (singlet state be-
longs to NOON state class and its creation is explained in Chapter 4).
Starting from singlet state (Eqn. 3.35), we have simulated DM + XY Hamiltonian
using the decomposition given in Eqn. 3.32. The τ modulated concurrence is
oscillating and is shown in Fig. Fig. 3.12(a). The experimental results (Fig.
3.12(a)) show a good agreement with the theoretical simulations. It is observed
that the value of concurrence oscillates for all values of γ and stays high for high
γ values.
Average experimental deviation (Eqn. 3.33) is observed to be 4.5%, 3.9 % and
3.1% for γ=0.33, 0.66, and 0.99 respectively. This experimental deviation can be
due to decoherence, static and rf inhomogeneities. Total experimental time was
9 ms, which is much shorter than the shortest T2 (290 ms) as well as T ∗2 (175 ms).
This rules out decoherence and static field inhomogeneity as the sources of error.
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 54
Figure 3.11: Experimental scheme used for entanglement preservation. τ isvaried from 0 to 3 in sixteen steps.
Entanglement Preservation
Entanglement plays a crucial role in several applications of quantum computation
and quantum information theory with best known applications in superdense cod-
ing [80] and quantum teleportation[81]. Preservation of entanglement is a crucial
element in the realization of quantum information processing. Hou et al. [82]
demonstrated a mechanism for entanglement preservation of a quantum state in
a Hamiltonian of the type given in Eqn. 3.22. Preservation of initial entangle-
ment of a quantum state is performed by free evolution interrupted with a certain
operator O, which makes the state to go back to its initial state. The operator
sequence for preservation is given in Eqn. 3.37.
O U(γ, τ) O U(γ, τ) ≡ I, (3.37)
where O = I1 ⊗ σ2z.
We have validated this entanglement preservation procedure using the experimen-
tal scheme as shown in Fig. 3.11. Starting from singlet state (|01〉−|10〉), we have
applied the preservation procedure as shown in Eqn. 3.37. Experiments have been
performed for three different values of relative interaction strength (γ), which are
γ=0.33, 0.66, 0.99. Varying τ (Fig. 3.11) from 0 to 3 in sixteen steps. The ex-
perimental results (Fig. 3.12(b)) shows excellent entanglement preservation. The
average experimental deviation of concurrence is less than 3%.
Conclusion
We have performed Fidelity Profile Optimization by Genetic algorithm and solved
single parameter and two parameters dependent quantum gates with experimental
verification in NMR. Exact φ dependent experimental unitary decompositions of
C-φ and CC-φ are solved using first order FPO. Unitary operator decomposition
Chapter 3. Quantum Simulation using Fidelity Profile Optimization 55
(a)
(b)
Figure 3.12: (a).Entanglement (concurrence – Eqn. 3.36) dynamics of 13C−Hsystem under the Hamiltonian Eqn. 3.21, (b). Entanglement preservation ex-periment using Eqn. 3.37. Starting from singlet state, the concurrence sustainsat unity with the preservation procedure. (Symbols are same as in 3.12(a)).The inlet shows a blowup of the concurrence on a scale of 0.96 to 1 and τ from0 to 3. The entanglement varies between 0.97 and 0.995 for all the three valuesof γ.
for experimental quantum simulation of DM interaction in the presence of Heisen-
berg XY interaction is shown with second order FPO (for any relative strengths
of interactions γ and evolution time τ). Experimental gate time for this decom-
position is invariant under γ or τ , which can be used for relaxation independent
studies of the system dynamics. Using these decompositions, we have experimen-
tally verified the entanglement preservation mechanism suggested by Hou et al.
Entanglement being the most important resource, its preservation is crucial in
quantum information processing.
Chapter 4
Efficient Creation of NOON
States in NMR
4.1 Introduction
NOON state or Schrodinger cat state is a maximally entangled N qubit state with
superposition of all individual qubits being at |0〉 and being at |1〉(Eqn. 4.1),
|ψNOON〉 =|N|0〉, 0|1〉〉+ e−iφ |0|0〉, N|1〉〉√
2. (4.1)
NOON states [83, 84] – introduced by Barry Sanders [85] in 1989 – have received
much attention recently for its high precession phase measurements, which enables
the design of high sensitivity sensors in optical interferometry [86] and NMR [87].
Ability to make high precesion phase measurements of NOON state is very impor-
tant in quantum metrology and is exploited recently in NMR by Jones et al. [87].
Here we show an efficient and practical method for entangling and disentangling
spin-NOON states in NMR, which is,
(i) a minimal unitary operator decomposition in terms of experimental unitaries
– radio frequency pulses (single qubit rotation (SQR)) and delays (J evolu-
tion, basic two spin operation in NMR) – and has
(ii) good experimental fidelity by Genetic algorithm (GA) optimized composite
unitary operations (composite pulses and composite evolutions).
57
Chapter 4. Efficient creation of NOON states in NMR 58
This high fidelity decomposition effectively enables NOON state creation (NSC)
with minimal knowledge about the system parameters (allowed variation of rf
control field strength and coupling strength upto ±50%). We have addressed this
Efficient NOON state creation problem in spin chain with nearest neighbor (NN)
couplings and spin star topology (SST).
Even though our study is based on basic NMR weak coupling approximation
Hamiltonian (Eqn. 4.2), it can be extended to any system Hamiltonian (Hsys)
by ‘term isolation’ procedure by Bremner et al. [67, 68].
Hsys = 2πJIzIz (4.2)
Section 4.2 discuss the entangling and disentangling operator decomposition fol-
lowed by experimental implementation in three spin NMR quantum information
processor (Sec. 4.3).
4.2 Theory
4.2.1 NOON State Creation (NSC) operators
Starting from |0〉⊗N state, NOON state can be created by a Hadamard gate, UH
(Eqn. 4.3) and (N − 1) Controlled-NOT (UCNOT ; Eqn. 4.4) gates as shown in
Fig. 4.1[88].
UH =1√2
[1 1
1 −1
](4.3)
UCNOT =
1 0 0 0
0 1 0 0
0 0 0 1
0 0 1 0
(4.4)
The experimental unitary decomposition of UH and U i,jCNOT in a weakly coupled
Liquid State NMR (LS-NMR) system are given by,
UH = R(π/2, π/2)×R(π, 0), (4.5)
Chapter 4. Efficient creation of NOON states in NMR 59
Figure 4.1: Quantum circuit for NOON state creation. N qubit system re-quires (N − 1) Controlled-NOT gates and a Hadamard gate.
and
U i,jCNOT = Rj(π/2, 0)×Rj(π/2, π/2)× U i,j
zz (π/2)
×Rj(π/2,−π/2)×Ri(π/2, π/2)
×Ri(π/2, 0)×Ri(π/2,−π/2),
(4.6)
where,
Rn(θ, φ) = exp(−iθ × [Cosφ Inx + Sinφ Iny ]), (4.7)
is single qubit rotation (SQR) operation on nth spin and,
Uzz(Θ) = exp(−i Θ 2IzIz). (4.8)
External rf pulses of specific frequency, amplitude(A), phase and duration(t) are
used for implementing Rn(θ, φ) in NMR, where A and t controls θ, rotation angle
of the pulse. Uzz(Θ) is implemented using, time evolution (τ = Θ/πJ) of the
NMR weakly coupling Hamiltonian (Eqn. 4.2).
The general NOON state circuit (Fig. 4.1) in a N spin NMR system with weak
coupling Hamiltonian requires,
Nr = 2 + 6× (N − 1), (4.9)
number of SQR rf pulses which cause experimental errors due to rf inhomogeneity
and imperfect calibration [89].
Using GA optimization, a minimal unitary operator decomposition for NSC is per-
formed for small spin systems (N < 10) by state to state optimization (explained
in Chapter 2) with |0〉⊗N as initial and NOON state as final state. Using these
decompositions we have identified the pattern of the SQRs and evolutions in NSC
and generalized to N spin cases – spin chain with NN couplings (Fig. 4.3(a))
and spin star topology (Fig. 4.3(b)). Most general spin system – spin cloud (Fig.
Chapter 4. Efficient creation of NOON states in NMR 60
4.3(c)) – with random coupling between any random qubit pairs can be solved
as a spin chain model with NN coupling using interaction selection procedure by
Leung et al [90].
4.2.1.1 Spin chain with nearest neighbor couplings
Spin chain with nearest neighbor couplings (Fig. 4.3(a)) is one of the most studied
qubit network in quantum information processing [91, 92]. We studied the spin
chain NSC in following two cases,
(i) Spin chain with NN equal couplings (J i,i+1 = constant): The experimental
unitary operator decomposition in this case is obtained as,
Uchain = R(π/2, φ)×N−1∏n=1
[Un,n+1zz (π/2)R(π/2, φ)], (4.10)
All the (N + 1) pulses here are same (R(π/2, φ)) and global to all the spins
in the chain. This is the only case where NSC can be performed with global
external controls.
(ii) Spin chain with NN non-equal couplings : The unitary operator decomposi-
tions in this case is obtained as,
U ′chain =N∏n=2
[Rn(π/2, 0)× Un−1,nzz (π/2)]×
N∏m=2
[Rm(π/2, π/2)]×R1(θ, φ).
(4.11)
The decomposition has one global and (N − 1) local control pulses. This
corresponds to three pulses for Bell state (|00〉 + |11〉) creation (Eqn. 4.12,
Fig. 4.2),
U2002 = R2(π/2, 0)× U1,2zz (π/2)×R2(π/2, π/2)×R1(θ, φ), (4.12)
and five pulses for GHZ state (|000〉 + |111〉) creation in a heteronuclear
NMR system. Only one NN spin pair coupling will evolve during each delay
and is performed by interaction selection procedure by Leung et al. [90].
Chapter 4. Efficient creation of NOON states in NMR 61
Figure 4.2: Pulse sequence for for two spin NOON state creation.
(a) (b) (c)
Figure 4.3: (a). Spin ising chain with nearest neighbor couplings. J i,i+1 isthe coupling between ith and (i+ 1)th spin, (b). Spin Star network. All B spinsare coupled with spin A with equal strength and there is no coupling betweenB spins, (c). A general multi-spin system (spin cloud).
4.2.1.2 Spin star topology
Spin star topology (Fig. 4.3(b)) has a central spin A connected to (N − 1) num-
ber of B spins with equal coupling and there is no coupling between any two B
spins. This is the case with more practical importance because of the availabil-
ity of experimental systems with sufficient size (N) such as trimethylphosphite
(N = 1 (31P ) + 9 (1H) = 10) and tetramethylsilane (TMS) (N = 1 (29Si) +
12 (1H) = 13). Experimental implementation of magnetic field sensing is per-
formed in trimethylphosphite by Jones et al [87, 93].
Using GA, the optimized unitary operator decomposition for spin star topology is
obtained as,
Ustar = RB(π/2, 0)× UA,Bzz (π/2)×RB(π/2, π/2)×RA(θ, φ). (4.13)
The system requires two global control pulses on spin-B and a single pulse on
spin-A for NSC.
Chapter 4. Efficient creation of NOON states in NMR 62
4.2.2 Disentangling NOON states
Operators for disentangling NOON states are evaluated using the fact that product
of entangling operator (UE) and disentangling operator (UD) is identity (Eqn.
4.14).
UE × UD = I ⇒ UD = (UE)−1, (4.14)
where UE NSC operators.
Unitary decompositions for NSC operators (Eqn. 4.10, 4.11 and 4.13) combined
with inverse unitaries of SQRs (Eqn. 4.15) and the evolutions (Eqn. 4.16) can
be used for evaluating experimental unitary decomposition (UD) for disentangling
NOON states.
Rn(θ, φ)−1 = Rn(θ, φ+ π). (4.15)
U i,jzz (π/2)−1 = U i,j
zz (π/2)×Ri or j(π, 0). (4.16)
4.2.3 Robust NOON state creation
For an N qubit system, a complete knowledge about system parameters – control
field strengths (Fi), coupling strengths (Jj) – are necessary for the high fidelity
quantum information processing. Possible errors – uncalibrated control fields or
error in coupling value evaluations – and incomplete knowledge on system param-
eters can be avoided by finding unitary operator decompositions which has a self
compensatory mechanism for these errors.
The basic unitaries in an NMR pulse sequence are pulses (or SQRs) and delays
(or J evolutions). Robust unitary operator decomposition of NSC operators (UE
or UD) in terms of these SQRs and J evolutions are performed by replacing the
SQRs with composite SQRs and evolutions with composite evolutions. Composite
SQR (or composite pulse) [94, 95] is a sequence of adjacent pulses with optimal
amplitudes and phases whereas composite evolution operator [96] is a sequence
of adjacent evolution operators (here delays) interrupted by pulses. The opti-
mal delay values, amplitudes and phases for these composite unitaries (Ucom) are
evaluated using Genetic algorithm.
Chapter 4. Efficient creation of NOON states in NMR 63
Composite unitary (Ucom) optimization using Genetic algorithm is performed by
maximizing (to unity) the fitness function,
F = Trace(Ucom(θp;φp;Dd)× Utar), (4.17)
where p varies from 1 toNp (total number of pulses) and d varies from 1 toNd (total
number of delays). Utar is the target unitary operator (SQR or evolution operator).
Now the optimization problem is to find the optimal values for 2×Np +Nd pulse
sequence parameters, which will maximize the function F. GA optimization of π/2
composite unitary (five pulse – requires optimal values for 10 parameters) took
around 15 minutes in a 3 GHz Intel processor.
Composite five pulse sequence for 90 (Eqn. 4.18) and 180 (Eqn. 4.19) SQR with
phase φ are obtained as (in the format Rotation Angle(Phase)),
90(φ) ≡ 91.60(45.00 + φ) : 153.67(30.00 + φ) : 177.20(235.81 + φ) :
183.19(342.47 + φ) : 128.72(120.00 + φ).(4.18)
and
180(φ) ≡ 178.81(102.57 + φ) : 171.90(334.93 + φ) : 186.35(291.44
+φ) : 175.12(76.76 + φ) : 186.70(17.39 + φ).(4.19)
The response of Eqn. 4.18 and Eqn. 4.19 against pulse angle error is shown in Fig.
4.4(a) and Fig. 4.4(b) respectively. Near 100% fidelity is observed for an error (in
pulse angle) upto ± 50 %.
The composite evolution (U czz(90)) is obtained as,
U czz(π/2) = P1(64.28)× Uzz(169)× P2(244.26)× Uzz(359.26)×
P2(64.26)× Uzz(179.4)× P1(244.29)(4.20)
P1(δ) = 108.99(298.79 + δ) : 172.05(77.49 + δ) : 108.99(298.79 + δ),
P2(δ) = 317.22(319.06 + δ) : 174.45(105.64 + δ) : 175.19(9.85 + δ) :
153.89(261.95 + δ).(4.21)
Chapter 4. Efficient creation of NOON states in NMR 64
−100 −50 0 50 1000
50
100
Error in Flip Angle (%)
Fid
elity
(%
)
(a)
−100 −50 0 50 1000
50
100
Error in Flip Angle (%)
Fid
elity
(%
)
(b)
Figure 4.4: (a). 90 composite pulse (five pulse) response for flip angle error,(b). 180 composite pulse (five pulse) response for flip angle error.
0
80
8080
8090
9090
90 95
95
95
95
99
9999
99
99
90
Error in J (%)
Err
or in
flip
angl
e (%
)
−100 −50 0 50 100−100
−50
0
50
100
Figure 4.5: Flip angle and coupling error response of composite evolution(Eqn. 4.20).
P1 and P2 (Eqn.4.21) are in same format as shown in Eqn.4.18.
The response of composite evolution (Eqn. 4.20) is shown in Fig. 4.5 and has
robust (> 99%) response against coupling value and flip angle errors upto ±50%.
This gives an added advantage such that, complete knowledge on exact coupling
values or control field (Hrf ) strengths are not required for NSC. This advantage is
Chapter 4. Efficient creation of NOON states in NMR 65
Figure 4.6: 13C labeled CHFBr2 used for experimental NSC. The spins 1H,13C and 19F forms a linear spin chain with NN coupling values JH,Czz = 224.5Hz and JC,Fzz = −310.5 Hz, if the two bond JH,Fzz = 49.7 Hz is refocused duringthe entire experimental time. The longitudinal and transverse relaxation time– T1, T2 (in seconds)– for 1H, 13C and 19F are 6.7, 1.4, 1.9, 0.71 and4.0, 0.7.
exploited in our experimental demonstration of three spin NOON state creation.
This idea can be used for cases like perfect state transfer [97], where engineered
spin chains are required.
4.3 Experimental
Experiments have been carried out in the three spin NMR system 13CHFBr2 (13C
labeled) (Fig. 4.6). The spins 1H , 13C and 19F forms a linear spin chain with NN
coupling values JH,Czz = 224.5 Hz and JC,Fzz = −310.5 Hz. The coupling between1H and 19F (JH,Fzz = 49.7 Hz) can be refocused during JH,F evolution [14]. All
experiments are performed in Bruker AV-500 NMR spectrometer at an ambient
temperature of 300 K.
Quantum information processing experiments in NMR starts with, (a). prepara-
tion of pseudo pure states [28–30], (b). processing the state by evolving under
different average Hamiltonians [14] and (c). read-out by quantum state tomogra-
phy [32]. We have utilized the spatial averaging technique [28] for creating |0〉⊗n
pseudo pure states as described in [78]. Pulse sequence for three spin PPS (|000〉)is shown in Fig. 4.7.
We have demonstrated NOON state creation using the operator decomposition for
spin chain (Eqn. 4.11) and spin star topology (Eqn. 4.13) in combination with
composite evolution sequence shown in Eqn. 4.20. Composite evolution (Eqn.
4.20) doesn’t require any prior knowledge on exact system variables (coupling
Chapter 4. Efficient creation of NOON states in NMR 66
( . , )
, ,
,
,
,
,
Figure 4.7: Pulse sequece for three spin pesudo pure state creation.
(a) (b)
Figure 4.8: Quantum state tomography of three spin NOON state createdin (a). spin chain (Eqn. 4.11) and (b). spin star topology (Eqn. 4.13) incombination with composite evolution sequence shown in Eqn. 4.20. The labelsin XY plane 1,2...8 correponds to the computational basis states |000〉, |001〉,...... |111〉. The observed experimental fidelity is 95.9 % for spin chain and96.4 % for spin star topology.
values and control field strength). For the 1H −13 C −19 F linear spin chain with
NN coupling values JH,C = 224.5Hz and JC,F = 310.5Hz (sign flipped with an
extra π pulse), we have used the composite evolution tuned for 267 Hz, which can
give near 100% evolution response fidelity over a coupling range of 133 Hz to 401
Hz. For spin star topology, we have used 13C as the A spin and 1H and 19F as B
spins. The 1H −19 F coupling is refocused using a spin echo sequence [98].
The quantum state tomography is performed for the final state [78] and is shown
in Fig. 4.8(a) and 4.8(b). The observed experimental fidelity is 95.9% and 96.4%.
The remaining error is likely due to rf inhomogeneity. The contribution of deco-
herence to loss of fidelity is likely to be negligible, since the total experimental
time (≈ 30ms) is much less than the smallest T2 (= 700ms). A slightly lighter
Chapter 4. Efficient creation of NOON states in NMR 67
fidelity in star topology is likely to be due to smaller number of pulses compared
to spin chain.
Conclusion
An efficient method for NOON state creation in NMR is demonstrated. The
unitary operator decomposition shown here is experimental friendly with a self
compensation mechanism on common experimental errors. Over a 50 % variation
in coupling value or spin control field strength is allowed, which effectively rules
out any strict constraints on system parameters of the spin system.
Chapter 5
Composite Operator Design
Using Genetic Algorithm
5.1 Introduction
A composite pulse is a cluster of nearly connected rf pulses which emulate the
effect of a simple spin operator with robust response over common experimental
imperfections [95]. Composite pulses are mainly used for improving broadband
decoupling [99–101], population inversion [102, 103], coherence transfer [104] and
in nuclear overhauser effect [105] experiments. Composite operator is a general-
ized idea where a basic operator (such as rotation or evolution of zz coupling)
is made robust against common experimental errors (such as miscalibration of rf
power or error in evaluation of zz coupling strength) by using a sequence of ba-
sic operators available for the system. Here using Genetic Algorithm, a global
optimization method based on the natural genetic evolution, we have designed
following composite operators,
(i) broadband rotation pulses,
(ii) rf inhomogeneity compensated rotation pulses and
(iii) zz evolution operator with robust response over a range of zz coupling
strengths.
We also performed rf inhomogeneity compensated Controlled NOT gate.
69
Chapter 5. Composite Operator design using Genetic Algorithm 70
5.2 Theory and Experimental
Single spin NMR Hamiltonian in presence of an external rf field is given by,
H = B1(Cos(φ)Ix + Sin(φ)Iy) + δIz, (5.1)
where B1 → strength of external rf in Hz, φ → Phase and δ → offset. The
evolution operator for this Hamiltonian (Eqn. 5.1) for a time τ is given by,
U(θ, φ,Ω) = exp(−iHτ),
= exp(−i(θ(Cos(φ)Ix + Sin(φ)Iy) + ΩIz)),(5.2)
where θ = B1τ and Ω = δτ .
We have solved U(θ, φ,Ω) (Eqn. 5.2) using Mathematica[106] yielding,
U(θ, φ,Ω) =
[Cos Θ− iΛΩ −iθΛe−iφ
−iθΛeiφ Cos Θ + iΛΩ
](5.3)
where Θ→√θ2 + Ω2
2and Λ→ Sin(Θ)
2Θ.
From the basic property of this operator matrix Eqn. 5.3, can be written as,
U(θ, φ,Ω) =
[α β
−β∗ α∗
](5.4)
where ‘*’ represents complex conjugate and α, β = Cos Θ − iΛΩ,−iθΛe−iφ.Hence α, β array is enough to represent matrix elements of a single spin unitary
operator.
U(θ, φ,Ω) ≡ α, β. (5.5)
The matrix product in this new representation can be shown as,
α1, β1 × α2, β2 = α1α2 − β1β∗2 , α1β2 + β1α
∗2 (5.6)
This reduction in computation of effective single spin operator can be used for
enhancing the optimization procedure. A Matlab function is written – Compos-
itePulseOperator (Code. 1) – for evaluating effective operation for N single spin
operations. Inputs of the function are arrays of N composite single spin operation
parameters – ‘Amplitude’, ‘Phase’ and ‘Offset’. Computational time advantage
Chapter 5. Composite Operator design using Genetic Algorithm 71
1 f unc t i on E f f e c t i v eOpera to r=CompositePulseOperator (A1 , P1 , cs )E1=0.5*(A1.ˆ2 + cs . ˆ 2 ) . ˆ 0 . 5 ; E2=s i n (E1) . / ( 2*E1) ;
3 M=[( cos (E1)−1 i .* cs .*E2) (−1 i *exp(−1 i *P1) .*E2 .*A1) ] ;L=length (M) /2 ;
5 pdt=[M(1) M(L+1) ] ;f o r k=2:L ;
7 pdt=[pdt (1 ) *M( k )−pdt (2 ) * conj (M(L+k ) ) pdt (1 ) *M(L+k )+pdt (2 ) *conj (M( k ) ) ] ;
end9 Ef f e c t i v eOpera to r=pdt ;
CODE 1: Matlab Program for finding the effective operation of N compositeSQRs.
0 50 100 150 2000
20
40
60
Number of SQRs/Pulses
Tim
e A
dvan
tage
(γ)
Figure 5.1: Computational time advantage (γ, Eqn. 5.7) efficiency of effectiveoperator evaluation using ‘CompositePulseOperator ’ and Matlab builtin func-tion ‘expm’ [1]. The response is evaluated for a set of randomnly created Npulse composite sequence. (Matlab 7.14, 32 bit Ubuntu 12.04 OS with core i53.6 GHz Intel processor)
study (γ, Eqn. 5.7) of CompositePulseOperator over Matlab’s built in function for
matrix exponential – expm – is performed and is shown in Fig. 5.1.
γ = τ1/τ2, (5.7)
where τ1 → time for evaluating expm [1] and τ2 → time for evaluating Compos-
itePulseOperator).
In optimization point of view composite operator optimization can be categorized
as state to state (SS) and operator optimizations.
(i) State to state (SS) Optimization: Optimizing composite operator pa-
rameters for high fidelity final state generation from a specified initial state.
We have performed the following SS optimisations
Chapter 5. Composite Operator design using Genetic Algorithm 72
(a) Excitation – SS optimization with thermal equilibrium (I iz) state as
initial state and I ix or I iy as the final state.
(b) Inversion – SS optimization with I iz state as initial state and −I iz as
the final state.
(c) Polarization Transfer – SS optimization with initial state → I1y and
final state → 2I1z I
2y
(ii) Operator Optimization : Optimizing composite operator parameters for
high fidelity spin operator. We have performed operator optimization for,
(a) π and π/2 pulses,
(b) zz - evolution operator (Uzz),
Uzz = exp(−iπ2σzσz), (5.8)
where σz is z - Pauli matrix,
(c) Robust Controlled-NOT [3] gate (UCNOT ),
UCNOT =
1 0 0 0
0 1 0 0
0 0 0 1
0 0 1 0
. (5.9)
GA optimization details of the Composite Pulse generation are given in the fol-
lowing,
5.2.1 Representation Scheme
Encoding solutions of the problem to individuals in genetic evolution is performed
using representation scheme. Here in composite pulse generation, we have used a
double array of length twice that of number of pulses (n) in the composite pulse.
5.2.2 Fitness Function
A fitness function is a particular type of objective function that describes the
optimality of a solution or individual. Flow of genetic evolution process of the
Chapter 5. Composite Operator design using Genetic Algorithm 73
population is mainly controlled by fitness function and is crucial for any GA op-
timization. The effective operator of a composite pulse with an ‘n’ on-resonance
(Ω = 0) rf pulses is given by,
UCP = u1(θ1, φ1)× u2(θ2, φ2)....un(θn, φn), (5.10)
where uk(θk, φk) is the kth (k = 1, 2....n) rf pulse operator. Composite pulse
operator with robust response against rf inhomogeneity are optimized with the
fitness function (Frf ),
Frf =1
m
m∑p=1
Trace(UpCPrf × U
′tar), (5.11)
where,
UpCPrf = u1(θ1× (1 + rp), φ1)×u2(θ2× (1 + rp), φ2)....un(θn× (1 + rp), φn), (5.12)
where rp varies from −r to r in m steps and r mainly depends on number of rf
pulses in the composite pulse.
Composite pulse with broadband operator is performed with the fitness function
(Fbb),
Fbb =1
m
m∑p=1
Trace(UpCPbb × U
′tar), (5.13)
where UpCPbb is given by,
UpCPbb = u1(θ1, φ1,Ωp)× u2(θ2, φ2,Ωp)....un(θn, φn,Ωp), (5.14)
where Ωp is varies from −Ω to +Ω in m steps.
Composite pulse state to state optimizations (excitation and inversion) with robust
rf inhomogeneity are optimized with the fitness function (Frf ),
Frf =1
m
m∑p=1
Trace(UpCPrf × ρin × (Up
CPrf )† × ρ′tar), (5.15)
where UpCPrf is given in Eqn. 5.12.
Chapter 5. Composite Operator design using Genetic Algorithm 74
Broadband state to state composite pulses (excitation and inversion) are optimized
with the fitness function (Fbb),
Fbb =1
m
m∑p=1
Trace(UpCPbb × ρin × (Up
CPbb)† × ρ′tar), (5.16)
where UpCPbb is given in Eqn. 5.14.
We also performed composite zz or evolution optimization for robust response
over a range of zz coupling values
Optimized composite pulses – in the format θ(φ) – and its experimetal verifi-
cation results are given below. The experimental verifications are performed in
H2O (1% H2O in D2O) . All experiments are performed in Bruker AV-500 NMR
spectrometer at an ambient temperature of 300 K. The strength of rf power used
is 50 KHz.
5.2.3 Broadband excitation pulses
We have performed optimization for three pulse, five pulse and seven pulse broad-
band excitation composite pulses by state to state optimization and are,
Three Pulse →274.6 (270.01 ): 206.23(90.01): 19.01(269.91),
Five Pulse →78.05(244.7): 198.22(2.58): 120.33(136.72): 130.74(17.35):
227.85(140.07) and
Seven Pulse →88.7(275.28): 112.66(294.07): 129.02(54.34): 61.24(4.17):
144.85(315.02): 128.37(66.65): 40.4(259.96).
The Matlab simulation excitation profiles of these composite pulses are shown in
Fig. 5.2.
The experimental verifications are performed by finding the proton spectral peak
intensity variation with change in offset and are shown in Fig. 5.3.
Chapter 5. Composite Operator design using Genetic Algorithm 75
−2 −1 0 1 20
50
100
Relative Resonance OffsetFi
delit
y(a) Three pulse excitation
−2 −1 0 1 20
50
100
Relative Resonance Offset
Fide
lity
(b) Five pulse excitation
−2 −1 0 1 20
50
100
Relative Resonance Offset
Fide
lity
(c) Seven pulse excitation
Figure 5.2: Matlab simulation of three pulse, five pulse and seven pulse broad-band excitation sequences. (State to state optimization with Iz as initial stateand Ix as final state.)
5.2.4 Broadband Inversion pulses
We have performed optimization for three pulse, five pulse and seven pulse broad-
band excitation composite pulses by state to state optimization and are,
Three Pulse → 288.8(0): 197.6(180): 123(0): 34.2(180),
Five Pulse →89.56(58.28): 118.57(324.64): 330.87(162.59): 118.57(324.64):
89.56(58.28) and
Chapter 5. Composite Operator design using Genetic Algorithm 76
−80 −40 0 40 800
50
100
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(a) Three pulse
−80 −40 0 40 800
50
100
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(b) Five pulse
−80 −40 0 40 800
50
100
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(c) Seven pulse
Figure 5.3: Experimental proton spectral intensity variation with differentoffset values of three pulse, five pulse and seven pulse broadband excitationsequences.
Seven Pulse →228.39(80.43): 67.28(74.67): 143.09(254.21): 61.94(65.14):
147.44(249.08): 203.67(66.17): 94.54(241.04).
The Matlab simulation inversion profiles of these composite pulses are given in
Fig. 5.4.
The experimental verifications are performed by finding the proton spectral peak
intensity variation with change in offset value and are given in Fig. 5.5.
Chapter 5. Composite Operator design using Genetic Algorithm 77
−2 −1 0 1 2−100
0
80
Relative Resonance OffsetFi
delit
y
(a) Three pulse
−2 −1 0 1 2−100
0
80
Relative Resonance Offset
Fide
lity
(b) Five pulse
−2 −1 0 1 2−100
0
80
Relative Resonance Offset
Fide
lity
(c) Seven pulse
Figure 5.4: Matlab simulation of three pulse, five pulse and seven pulse broad-band inversion sequences. (State to state optimization with Iz as initial stateand −Iz as final state.)
5.2.5 Broadband π/2 pulses
We have performed optimization for three pulse, five pulse and seven pulse broad-
band π/2 composite pulses by operator optimization. Unlike excitation pulses (SS
optimization which maximizes the transverse magnetization components starting
from thermal equilibrium state), broadband π/2 pulses act as π/2 rotation oper-
ator and doesn’t require any initial state conditions. The optimized broadband
π/2 pulses are,
Three Pulse → 328.86(2.22): 260.98(181.3): 22.51(354.8),
Chapter 5. Composite Operator design using Genetic Algorithm 78
−80 −40 0 40 80−100
−50
0
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(a) Three pulse
−80 −40 0 40 80−100
−50
0
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(b) Five pulse
−80 −40 0 40 80−100
−50
0
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(c) Seven pulse
Figure 5.5: Experimental proton spectral intensity variation with differentoffset values of three pulse, five pulse and seven pulse broadband inversionsequences.
Five Pulse →120.35(180): 334.61(360): 161.65(180): 260.24(360):
122.11(360) and
Seven Pulse →139.65(4.7): 314.71(183.14): 192.47(3.66): 110.79(182.52):
311.19(1.54): 149.91(180.83): 20.6(347.33).
The Matlab simulation excitation profiles of these composite pulses are given in
Fig. 5.6.
The experimental verifications are performed by finding the proton spectral peak
intensity variation with change in offset value and are given in Fig. 5.7.
Chapter 5. Composite Operator design using Genetic Algorithm 79
−2 −1 0 1 2
0
50
100
Relative Resonance OffsetFi
delit
y(a) Three pulse
−2 −1 0 1 20
50
100
Relative Resonance Offset
Fide
lity
(b) Five pulse
−2 −1 0 1 20
50
100
Relative Resonance Offset
Fide
lity
(c) Seven pulse
Figure 5.6: Matlab simulation of three pulse, five pulse and seven pulse broad-band π/2 sequences.
5.2.6 Broadband π pulses
We have performed optimization for three pulse, five pulse and seven pulse broad-
band π composite pulses by operator optimization. Unlike inversion pulses (SS
optimization which maximizes the −Iz magnetization starting from thermal equi-
librium state), broadband π pulses act as π rotation operator and doesn’t require
any initial state conditions. The GA optimized broadband π pulses are,
Three Pulse → 58.08(0.42): 238.18(180.09): 360(0.06) ,
Chapter 5. Composite Operator design using Genetic Algorithm 80
−80 −40 0 40 800
50
100
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(a) Three pulse
−80 −40 0 40 800
50
100
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(b) Five pulse
−80 −40 0 40 800
50
100
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(c) Seven pulse
Figure 5.7: Experimental proton spectral intensity variation with differentoffset values of three pulse, five pulse and seven pulse broadband π/2 sequences.
Five Pulse →136.15(61.39): 298.76(356.71): 138.12(141.33): 298.83(356.83):
136.1(61.54) and
Seven Pulse → 231.54(138.67): 239.31(366.12): 140.63(138.87):
97.51(41.98): 142.07(137.25): 237.87(364.3): 231.31(134.41).
The Matlab simulation inversion profiles of these composite pulses are given in Fig.
5.8. The experimental verifications are performed by finding the proton spectral
peak intensity variation with change in offset value and are given in Fig. 5.9.
Chapter 5. Composite Operator design using Genetic Algorithm 81
−2 −1 0 1 2−100
0
Relative Resonance OffsetFi
delit
y
(a) Three pulse
−2 −1 0 1 2−100
0
Relative Resonance Offset
Fide
lity
(b) Five pulse
−2 −1 0 1 2−100
0
80
Relative Resonance Offset
Fide
lity
(c) Seven pulse
Figure 5.8: Matlab simulation of three pulse, five pulse and seven pulse broad-band π sequences.
5.2.7 rf inhomogeneity compensated composite pulses
We have performed rf inhomogeneity compensated composite pulse optimization
for seven pulse excitation, nine pulse excitation, five pulse π/2 and five pulse π
composite sequences and are obtained as,
Seven pulse excitation →185.94(323.68): 349.1(287.64): 367.24(22.65):
289.92(344.5): 21.3(115.32): 637.28(153.15): 712(35.48),
Chapter 5. Composite Operator design using Genetic Algorithm 82
−80 −40 0 40 80−100
−50
0
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(a) Three pulse
−80 −40 0 40 80−100
−50
0
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(b) Five pulse
−80 −40 0 40 80−100
−50
0
Offset ∆ω (KHz)
Nor
mal
ised
Inte
nsity
(c) Seven pulse
Figure 5.9: Experimental proton spectral intensity variation with offset valuesof broadband three pulse, five pulse and seven pulse π composite pulses.
Nine pulse excitation →885.06(232.85): 228.48(300.87): 128.53(157.7):
301.86(97.7): 759.06(213.3): 884.46(42.83): 819.03(120.31): 702.6(45.25):
1077.55(217.84),
Five pulse π/2→228.61(105): 383.51(90): 442.24(295.81): 457.19(402.47):
321.25(180) and
Five pulse π→465.94(17.39): 437.05(76.76): 465.06(291.44): 429.01(334.93):
446.26(102.57).
Chapter 5. Composite Operator design using Genetic Algorithm 83
−1 −0.5 0 0.5 10
50
100
Error in RFFi
delit
y(a) seven pulse
−1 −0.5 0 0.5 10
50
100
Error in RF
Fide
lity
(b) nine pulse
Figure 5.10: Matlab simulation of seven pulse and nine pulse composite exci-tation sequences with robust rf inhomogeneity profile.
The Matlab simulation of these composite pulses are given in Fig. 5.10 and Fig.
5.11.
The experimental verifications are performed by finding the proton spectral peak
intensity variation with change in rf and are given in Fig. 5.12.
5.2.8 Composite Polarization transfer and Evolution op-
erator
We have performed composite polarization transfer optimization (state to state
transfer with initial state I1y and final state 2I1
z I2x) as well as composite zz evolution
operator (Uzz, Eqn. 5.8). Unlike composite pulses discussed above, composite
delay operator optimization is performed over zz evolution operators (or delays)
and hard pulses for near 100% fidelity for a range of zz coupling strength.
The GA optimized composite operator sequence is given below (in the format
θ(φ) − d ), where d (in bold) represents the zz evolution delay in terms of
angle.
Chapter 5. Composite Operator design using Genetic Algorithm 84
−1 −0.5 0 0.5 150
75
100
Error in RF
Fide
lity
(a) Five pulse π/2
−1 −0.5 0 0.5 10
50
100
Error in RF
Fide
lity
(b) Five pulse π
Figure 5.11: Matlab simulation of five pulse π/2 and π composite sequenceswith robust rf inhomogeneity profile.
−0.8 −0.4 0 0.4 0.8
50
100
Error in rf
Nor
mal
ised
Inte
nsity
(a) Five pulse π/2
−0.8 −0.4 0 0.4 0.8
50
100
Error in rf
Nor
mal
ised
Inte
nsity
(b) Seven pulse excitation
Figure 5.12: Experimental proton spectral intensity variation with rf valuesof five pulse π/2 and seven pulse excitation.
Chapter 5. Composite Operator design using Genetic Algorithm 85
0 2 4 60
50
100
Error in JFi
delit
y(a) Polarization transfer (initial state: I1y and final state: 2I1z I
2x)
−1 −0.5 0 0.5 150
75
100
Error in J
Fide
lity
(%)
(b) Uzz (Eqn. 5.8) operator
Figure 5.13: Composite operator sequence for for J compensation.
The optimized operator sequence obtained for polarization transfer is,
Polarization Transfer →177.5 - 358.17(77.11) - 88.99 - 161.95(72)
- 177.52 - 296.4(57.73) - 177.52 - 73.44(348.41) - 177.51 - 97.67(32.24)
- 90(90).
The Matlab simulation of this polarization transfer profile with change in J cou-
pling strength is shown in Fig. 5.13(a). The composite polarization transfer
operator has a near 100% fidelity for error in J upto ±50% and shows a periodic
repetition as shown.
The optimized composite Uzz operator (Eqn. 5.8) is obtained as,
evolution operator →85.52(64.28) - 168.97 - 203.25(64.26) - 359.26
- 200.7(244.26) - 179.4 - 83.2(244.29) - 91.53 - 349.57(64.29).
The Matlab simulation of Uzz operator profile with change in J coupling strength
is shown in Fig. 5.13(b). The composite polarization transfer operator has a near
100% fidelity for error in J upto ±50%.
Chapter 5. Composite Operator design using Genetic Algorithm 86
Figure 5.14: Pulse Sequence for Controlled NOT gate.
−1 −0.5 0 0.5 125
50
75
100
Error in RF
Fide
lity
(%)
Figure 5.15: Inhomogeneity response profiles of controlled NOT gate.
5.2.9 RF inhomogeneity compensated Controlled NOT gate
Controlled NOT (CNOT) gate is a basic two qubit gate. CNOT applies a NOT
operation on target qubit based on the state of the control qubit. Any quantum
circuit can be simulated to an arbitrary degree of precision using a combination of
CNOT gates and single qubit rotations [3]. The pulse sequence used in NMR to
perform CNOT gate is shown in Fig. 5.14. We have performed GA optimization
for a robust CNOT gate operation in a range of rf intensity by replacing all rf
pulse with composite rotation pulses.
The rf compensated Controlled NOT gate operator (UCNOT – Eqn. 5.9) is obtained
as,
Controlled-NOT gate →349.63(343.94)1: 349.63(318.43)1:
173.51(127.44)1: 352.33(106.18)1: 173.51(84.92)1: 349.63(343.94)2:
349.63(318.43)2: 173.51(127.44)2 : 352.33(106.18)2: 173.51(84.92)2
- 90 - 349.63(343.94)2: 349.63(318.43)2: 173.51(127.44)2:
352.33(106.18)2: 173.51(84.92)2.
Chapter 5. Composite Operator design using Genetic Algorithm 87
The composite sequence has 15 hard pulses and single delay. The rf response of
composite pulse is shown in Fig. 5.15. The fidelity stays near 100% for rf deviation
upto ±50%.
5.3 Conclusion
Using Genetic Algorithm Optimization, we have generated composite operators
– composite pulses and composite evolution operators – which is robust against
common experimental errors. We experimentally verified all the composite pulses
in H2O sample. We also optimized controlled NOT gate, which is robust against
±50% error in control field strength.
The theory of robust composite pulses in NMR starts by the article “NMR Pop-
ulation Inversion Using a Composite Pulse” by Malcolm Levitt and Freeman[95]
(1979). Ever since, fine tuning of composite pulses are performed for different
applications in NMR using numerical and analytical methods [99–105, 107–112].
The composite pulse sequences for brodband and rf inhomogeneity compensation
shown in this chapter may not show a good improvement over existing sequences,
still GA optimization can be used for making robust complex operators as shown
in the case of polarization transfer operator and Controlled NOT gate.
Chapter 6
Fast and Accurate Quantification
Using GA Optimized 1H −13 C
Refocused Constant-Time INEPT
6.1 Introduction
NMR spectroscopy provides a powerful and convenient tool for quantitative analy-
sis of newly synthesized compounds and natural products [113–115]. NMR spectra
used for quantification must have (i) well resolved peaks with unambiguous spec-
tral assignment and (ii) non-modulated peak intensities. In general spectral peak
intensities are modulated by system parameters such as coupling strength, relax-
ation, mobility of molecule and number of protons attached. These quantification
criteria naturally select 1H decoupled 13C NMR for complex mixture quantifica-
tion. Large experimental time, because of the less sensitive 13C nuclei, is the major
limitation of traditional 13C quantification experiment. To overcome this, experi-
ments utilizing polarization transfer from abundant high sensitive protons to 13C
(APT, NOE, INEPT) [116–118] are used, but they are sensitive to system param-
eters. By averaging over a number (n) of experiments with different polarization
transfer response, experiments like INEPT or refocused INEPT (R-INEPT) can
be made less sensitive to coupling strength and the number of attached protons
[119, 120]. This forms an optimization problem, where uniformity of the transfer
response over J coupling and average polarization transfer decides the convergence
criterion (or quantification efficiency).
89
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 90
This problem was recently addressed by Makela et al. [121], who devised refocused
constant time INEPT (Q-INEPT-CT) experiment. Makela et al. optimized a
delay of eight scans, and obtained a nearly uniform polarization transfer for CH1,
CH2 and CH3 carbons over a coupling range of 115-170 Hz, allowing accurate
quantification of the relative concentration of various compounds in a mixture.
Makela et al. used quasi-Newton non-linear regression algorithm in Microsoft
Excel solver routine for optimization of the values of the delays in a delay set.
Since experimental time is the major concern, we have addressed the problem
with various optimization constraints, which can improve the average polarization
transfer and hence reduce the experimental time. We have used Genetic Algorithm
(GA) optimization (Matlab®) for finding the optimized delay sets. We name these
experiments as GA-(Q-INEPT-CT) or in short GAQIC.
There are several potential solutions for J compensation in literature [122–124].
Broad-Band INEPT (BB-INEPT) by Bodenhausen et al. [122] and 90/45 INEPT
by A.M Torres et al. [123] uses composite evolution (sequence of evolution delays
interrupted by pulses) for J compensation. In optimization point of view, both
GAQIC and BB-INEPT controls the pulse sequence parameters to obtain an opti-
mized transfer response. BB-INEPT controls the pulse sequence parameters from
single experiment for J compensation, whereas GAQIC perform this by control-
ling pulse sequence parameters from a set of experiments, which provides a bigger
search space for GAQIC.
For a given range of J coupling, the response of BB-INEPT is not uniform for
CH1, CH2 and CH3, which rules out the possibility of using all the carbon signals
for quantification (averaging over many peaks is always an advantage). The pulse
sequence length as well as number of pulses also limit BB-INEPT for quantifica-
tion. Spectral edited quantitative delay sets (S1 and S2) are optimal in terms of
number of pulses and length, hence it is advantageous over Nyberg et al. [125] for
quantification.
Sec. 6.2 describes Genetic Algorithm optimization of delay sets followed by ex-
perimental implementations in Sec. 6.3. We also performed quantitative spectral
editing technique based on average polarization transfer and discuss it in Sec.
6.2.4.
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 91
6.2 Theory
Proton-carbon polarization transfer functions for methine (CH1), methylene (CH2)
and methyl (CH3) groups in refocused INEPT experiment [14, 126] are given by,
PCH1 =Sin(πJCH∆T )× Sin(πJCH∆R),
PCH2 =Sin(πJCH∆T )× 2Sin(πJCH∆R)× Cos(πJCH∆R),
PCH3 =Sin(πJCH∆T )× 3Sin(πJCH∆R)× Cos2(πJCH∆R),
(6.1)
where ∆T and ∆R are polarization transfer and refocusing delays. By averaging
over a number of experiments, n, with optimal delay set ∆T , ∆R, J dependence
of the polarization transfer can be approximated to a constant function (Eqn. 6.2).
n∑m=1
Pmi (J115Hz−170Hz) ≈ Ki, a constant,
where i = CH1, CH2, CH3.(6.2)
Only those delay sets which satisfies the constant function approximation (Eqn.
6.2) (or having a uniform polarization transfer response) can be used for quantifi-
cation.
The GAQIC pulse sequence is shown in Fig. 6.1. The delays ∆1, ∆2 and ∆3 (in
pulse sequence) are related in such a way that the chemical shift evolution during
transfer and refocusing delays are refocused (Eqn. 6.3). Using Eqn. 6.3, it can
be shown that the effective J-coupling evolution during INEPT and refocusing
segments are 2∆T1 and 2∆R1 respectively.
∆i1 + ∆i2 = ∆i3, (6.3)
Where i = T,R. The delays ∆i1, ∆i2 and ∆i3 are given by,
∆i1 = ∆/2,
∆i2 = ∆max/2−∆/2,
∆i3 = ∆max/2.
(6.4)
Where ∆ is the delay value from optimized delay set and ∆max is the maximum
delay in the optimized delay set. These choices of delay values will make sure the
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 92
Figure 6.1: GAQIC pulse sequence. This pulse sequence is same as in Makelaet al. [121]. Narrow black and wide gray bars represents 90o and 180o pulsesrespectively. Multiple narrow white bars in the 13C channel is 180o compositepulse, which reduce the off-resonance effects as shown in [121]. The first two 90o
pulses on carbon channel (purge) are to remove the native 13C magnetization.Gradient strengths are 45.4 G/cm (for G1 and G2) and 53 G/cm (for G3)with 4ms (for G1 and G2) and 200 us (for G3). We have used WALTZ-16decoupling during acquisition [56]. Phase cycling is as follows: φ1 = y, -y; φ2 =x, x, -x, -x; φrec = x, -x, -x, x.
total experimental evolution time (Tevol) is constant (Eqn. 6.5).
Tevol = (∆T1 + ∆T2 + ∆T3) + (∆R1 + ∆R2 + ∆R3)
= ∆Tmax + ∆Rmax = Constant.(6.5)
Keeping Eqn. 6.2 as converging criterion for optimization, we used Genetic Algo-
rithm for finding the optimal delay sets for (n =) 8 and 16 scan experiments. The
range of J coupling values for delay set optimization is set to (115 Hz - 170 Hz)
(C-H direct bond coupling of most of the chemical structures falls in this range).
The first step in GA optimization is defining a proper representation scheme for
encoding the solution of the problem to individuals in genetic evolution. In our
representation scheme, we have selected a floating point array of length 2n (n-
number of scans) for representing an individual. We have used a population of
size of 100 individuals for 1000 generations. Initial population is created by a
random array generator with a restriction on max value (posed by maximum
experimental time limit). Individual selection for next generation is performed by
Roulette wheel selection method [127]. We have used uniform mutation with rate
0.05 and crossover function ‘intermediate’ with ratio 1. ‘Elite count’ is set to 2
and ‘crossover fraction’ to 0.8 [127]. In a 3.6 GHz Intel-i5 processor with Ubuntu
12.04 OS, optimization takes around 2 hrs to converge for 8 scan experiment.
A fitness function is a particular type of objective function that prescribes the
optimality of a solution or individual. The basic fitness function of GAQIC (or
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 93
‘quantification efficiency’ of a delay set) is a function of average min-max differ-
ence (MMD) (difference between minimum and maximum points in the polariza-
tion transfer response) and average polarization transfer (APT) (mean value of
the transfer response). Min-max difference is a direct measure of uniformity (∝accuracy of quantification) of the transfer response whereas average polarization
transfer is inversely proportional to total experimental time.
Derived from Eqn. 6.2, we have used three different fitness functions with various
constraint levels. (A) near uniform and equal polarization transfer response collec-
tively for all the three types of carbons (B) near uniform and separate polarization
transfer response individually for each carbon and (C) near uniform polarization
transfer response for CH2 and CH3 carbons taken together.
6.2.1 Near uniform and equal response for CH1, CH2 and
CH3 transfer functions
The GA optimization fitness function in this case searches for a n scan experiment
delay set with near uniform and equal polarization transfer response for CH1, CH2
and CH3. Eqn. 6.2 with,
KCH1 = KCH2 = KCH2 , (6.6)
forms the fitness function constraint of the optimization.
We have optimized case A - GAQIC delay sets for (n=)8 and (n=)16 scan ex-
periments. The experiments are named as A8 (eight scan), A8´ (eight scan time
restricted (5.5 ms)) and A16 (sixteen scan). The optimized delay sets (∆T , ∆R)(in ms) are obtained as,
A8 → (2.1955, 2.7088) (1.9759, 1.235) (1.9657, 1.2291) (2.0504, 1.2121)
(4.7903, 3.7006) (8.9549, 4.4513) (4.0376, 3.7339) (1.9959, 1.1984),
A8´ → (5.5, 3.5836) (1.5283, 1.7226) (1.4814, 1.8069) (1.5419, 1.7993)
(1.4037, 1.7661) (1.5982, 1.8304) (5.5, 3.5999) (1.4943, 1.839) and
A16 → (2.0019, 1.4796) (8.9623, 4.6409) (1.8833, 1.6614) (5.8929, 3.9368)
(2.3277, 1.6185) (1.5878, 1.7082) (1.916, 1.0237) (5.6642, 3.9257) (3.9001,
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 94
3.7054) (1.7713, 1.0008) (0.54152, 9) (1.7067, 2.2225) (1.8876, 2.4176) (1.9443,
1.6098) (3.8368, 3.7403) (1.9736, 1.6823).
Average polarization transfer obtained for case A is ≈ 45%.
6.2.2 Near uniform and separate response for CH1, CH2
and CH3 transfer functions
The delay set optimization in this case search for near uniform polarization transfer
responses separately for each CH1, CH2 and CH3 carbons. The constraints of the
fitness function is given by Eqn. 6.2 with,
KCH1 6= KCH2 6= KCH3 . (6.7)
Compared with case A optimization, above constraints forms a bigger search space
and hence improved delay sets (with better MMD and APT). The optimized delay
sets in this case are named as B8 (eight scan) and B8´ (eight scan time restricted
(6 ms)). The optimized delay sets (in ms) are obtained as,
B8 → (2.943, 1.1792) (8.9563, 2.7485) (4.6173, 2.0081) (2.9618, 1.1761)
(2.9845, 1.1799) (4.2151, 3.0045) (2.9557, 1.1698) (2.9543, 1.1797) and
B8´ → (2.4107, 1.266) (2.8095, 0.9906) (1.7982, 1.2918) (5.9998, 2.6305)
(1.707, 1.5899) (5.6002, 3.1036) (1.958, 1.3282) (3.8098, 1.4394).
Average polarization transfer for case B delay set is ≈ 60%. Group wise aver-
age polarization transfer for B1 and B2 are 41.37CH1 , 62.73CH2 , 76.89CH3 and
45.23CH1 , 63.38CH2 , 73.349CH3 respectively.
6.2.3 Near uniform and separate response for CH2 and CH3
only
The delay set optimization in this case is a special case of case B (Sec. 6.2.2)
optimization. Here we loosen up the constraint level (Eqn. 6.2 and Eqn. 6.8) to
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 95
(a) (b)
(c) (d)
Figure 6.2: Comparison of polarization transfer response of refocused INEPTand GAQIC experiments. (a) refocused INEPT with J=145 Hz, GAQIC ex-periments; (b) case A (delay set A8 ), (c) case B (delay set B8 ) and (d) case C(delay set C8 ). Case C optimization has the best quantification efficiency with‘min-max difference’ of < 1 %. Average polarization transfer (which effectivelydetermines the experimental time) for case A, case B and case C delay sets are≈ 45%, 60% and 70% respectively.
improve the search space and hence results in delay sets with better quantification
efficiency.
KCH2 6= KCH3 . (6.8)
There is no constraint on CH1 polarization transfer. Compared to case A and
case B delay sets, we have obtained case C delay set with (i) average polarization
transfer > 70%, (ii) max-min difference < 1% and (iii) maximum delay of 5.5 ms.
However case C delay set is not suggested for those cases with more number of
CH1 groups in the compound mixture. The optimized delay set (in ms) obtained
is,
C8 → (2.2947, 1.2272) (2.9719, 1.7409) (3.0456, 1.7166) (5.5, 2.9666)
(2.403, 1.1468) (5.5, 3.0401) (1.7251, 1.0497) (2.4688, 1.0348).
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 96
Method Delay Set MMD (%) APT (%)
GAQIC
A8 2.98 44.47A8´ 8.45 47.46A16 2.13 44.23B8 1.25 60.33B8´ 2.77 60.70C8 0.83 70.28
QINEPT-CTH8 3.64 44.65H8´ 12.73 46.83
R-INEPT 49.88 78.03
Table 6.1: Min-Max Difference (MMD) and Average Polarization Transfer(APT) of all optimized delay sets and R-INEPT.
Average polarization transfer for case C delay set is ≈ 70%. Group wise average
polarization transfer for C8 is 66.1CH2 , 74.46CH3.
A comparative study of polarization transfer responses of refocused INEPT, case
A, case B and case C are shown in Fig. 6.2(a), 6.2(b), 6.2(c) and 6.2(d) respectively.
Average polarization transfer, which effectively determines the experimental time
for case A, case B and case C delay sets are ≈ 45%, 60% and 70% respectively
(detailed quantification efficiency of all the optimization delay sets are shown in
Tab. 6.1). The non-optimized R-INEPT (Fig. 6.2(a)) has MMD of ≈ 50% with
APT 78%, whereas case C optimized delay sets has MMD < 1% with APT 70%.
We have also compared our optimized delay sets with ‘original modulation’ (H8 )
and ‘time restricted modulation’ (H8´) of Makela et al. (Tab. 6.1).
6.2.4 Spectral editing by delay sets
A uniform, near zero polarization transfer response (Eqn. 6.9) for any of the
chemical group (CH1 or CH2 or CH3), will effectively remove those peaks in
refocused-INEPT spectrum. In the quantitative case, we have used the following
constraint,
KCH2 = 0 & KCH1 = KCH3 , (6.9)
which simplifies the spectrum (by removing the signals from CH2 carbons) without
losing the quantitative information. Compared to DEPT, the polarization transfer
response here is averaged over a number of experiments and has a near uniform
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 97
(a) (b)
Figure 6.3: Polarization transfer responses for quantitative spectral editingdelay sets (a). four scan (S4 ) (3.1187, 4.988) (5.8421, 6) (2.9747, 1.3583)(2.24, 3.0151), (b). sixteen scan (S16 ) (1.2241, 3.2027) (2.5979, 1.4173)(5.9911, 6.8728) (9.7327, 8.9775) (1.7114, 2.4693) (6.1213, 7.6599) (2.6226,4.9325) (2.8015, 1.5952) (2.1121, 4.9178) (5.3706, 7.9835) (2.17, 2.9369) (1.0491,3.0062) (2.7459, 1.0893) (3.4169, 5.5314) (1.7431, 4.403) (1.7297, 5.4263) (inms).
behavior over a J coupling range 115 Hz to 170 Hz. Using the constraints given in
Eqn. 6.9, we have optimized two delay sets, which are S4 (four scan) (Fig. 6.3(a))
and S16 (sixteen scan) (Fig. 6.3(b)).
It may be mentioned that the optimized delay sets given here will also work for
basic refocused INEPT sequence [118, 126]. However, the sequence of Fig. 6.1 as
suggested by Makela et al. also takes care of offset and inhomogeneity effects and
is a constant time experiment.
Additionally we have generated angle sets (θ = Jτ) for quantification in a mixture
with only one type of chemical group (SI1 or SI2 or SI3). We have named these
experiments as SI1-GAQIC (Fig. 6.4(a)), SI2-GAQIC (Fig. 6.4(b)) and SI3-
GAQIC (Fig. 6.4(c)). Optimization is performed over a parameter set of eight scan
experiment (8×2 delays) and has average polarization transfer ≈ 80%. The delay
sets can be used for any types of spin system such as CHn or NHn (n = 1, 2, 3).
The width of uniform profile can be increased by compromising over the maximum
polarization transfer.
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 98
0 0.5 1 1.5 20
0.4
0.8
J/JSI
Res
pons
e
(a) SI1-GAQIC
0 0.5 1 1.5 20
0.4
0.8
J/JSI
Res
pons
e
(b) SI2-GAQIC
0 0.5 1 1.5 20
0.4
0.8
J/JSI
Res
pons
e
(c) SI3-GAQIC
Figure 6.4: Angle sets (θ = Jτ) for (a) SI1 (1.5340, 1.5325), (2.7269,2.7205), (3.3392, 3.3238), (1.5250, 1.5267), (1.5269, 1.5298), (1.5244, 1.5271),(1.5344, 1.5318), (1.5152, 1.5238), (b) SI2 (3.3346, 1.6642), (1.5229,0.7640), (1.5270, 0.7624), (2.7182, 1.3651), (1.5273, 0.7633), (1.5343, 0.7671),(1.5338, 0.7643), (1.5213, 0.7618) and (c) SI3 (1.3839, 0.4803), (2.4315,1.3210), (1.4751, 0.6842), (1.2293, 1.7333), (1.3692, 0.6453), (1.0413, 0.6375),(2.0486, 0.9171), (0.8760, 0.5637).
6.3 Experimental
Experimental implementation (Fig. 6.1) of GAQIC - case A, B and C delay
sets have been performed in a mixture of 4-Penten 2 ol (P) (Fig. 6.5(a)), α-
methoxy phenyl acetic acid (M) (Fig. 6.5(b)) and 1-2-dimethyl cyclohexane (D)
(Fig. 6.5(c)) dissolved in CDCl3. All the experiments are performed in Bruker
AV-500 spectrometer at 300 K. Proton decoupled 13C spectrum of the mixture
with peak assignments is shown in Fig. 6.6. The GAQIC pulse sequence (Fig.
6.1) is identical to the pulse sequence given by Makela et al. [121]. The composite
pulses effectively takes care of rf inhomogeneity effects and detailed analysis of
offset effects are contained in Fig. 3 of [121] and the same is applicable here. All
FIDs are acquired for 1 second and apodized using an exponential function (0.3
Hz line broadening). All the spectra were analyzed using custom written matlab
program. The relative quantities of compound-M (40 mg) and compound-D (20
µl) with respect to compound-P (10 µl) are obtained as,
RQ21 =moles of compound-M
moles of compound-P= 2.47,
RQ31 =moles of compound-D
moles of compound-P= 1.42.
(6.10)
Experimentally measured relative quantities (RQ21 and RQ31) using case A (three
experiments: A8, A8´, A16 ), case B (two experiments: B8, B8´) and case C (A8 )
optimized delay sets are shown in Tab. 6.2. Inverse gated 1H decoupled 13C
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 99
(a) (b) (c)
Figure 6.5: Samples used for GAQIC experiment (a) 4-Penten 2 ol (P), (b) α-methoxy phenyl acetic acid (M) and (c) 1-2-dimethyl cyclohexane (D).
Figure 6.6: Equilibrium spectrum of the compound mixture (Fig. 6.5). Theassignments of various peaks (derived from DEPT experiments) are indicatedon the top of each resonance.
experiment with 45 excitation pulse is used as traditional 13C quantitative (Q)
measurement [121]. Average over all signals was taken since it will reduce the
random errors in quantification. The quantification procedure is as follows. For
compound P,
(i) identify the chemical group of all the carbons (CH1 or CH2 or CH3),
(ii) find the mean integration value for each group (I1, I2 and I3),
(iii) normalize I1, I2 and I3 using the average polarization transfer from each
group and find average over all the three (QP), which is proportional to the
concentration of the compound in the mixture,
(iv) repeat the same for all the compounds and find QP, QM and QD. Relative
quantities are the ratios of QP, QM and QD.
Since reduction in experimental time is the major concern, we performed the
GAQIC experiments to obtain minimum experimental time without significant
compromise on the quantification accuracy. The different GAQIC delay sets run
for different number of scans according to their average polarization transfer values.
Case A optimization has the minimum average polarization transfer efficiency (≈45%) and requires the highest number of scans (256) whereas case C optimization
(average polarization transfer ≈ 70%) needs only 128 scans. The experimentally
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 100
Delay Set Scans RQ21 RQ31
Q 512 2.42 1.39A8
256
2.33 1.35A8´ 2.35 1.31A16 2.33 1.31H8 2.40 1.35H8´ 2.38 1.30B8
1602.43 1.38
B8´ 2.37 1.35C8 128 2.43 1.37
Table 6.2: Experimentally measured relative quantities of compound 2 (RQ21)(α-methoxy phenyl acetic acid) and compound 3 (RQ22) (1-2-dimethyl cyclo-hexane) with respect to compound 1 (4-penten 2 ol). Relative quantity byweight is shown in Eqn. 6.10.
(a)
(b)
Figure 6.7: Experimental implementation of GAQIC spectral editing delayset. (a). 1H decoupled 13C spectra and (b). using the delay set (S4) shownin Fig. 6.3(a), CH2 peaks (indicated by ↓ mark above) are missing in the fourscan averaged INEPT spectrum.
measured relative quantities (RQ21 and RQ31) are compared with the traditional13C quantitative experiment (Q) in Tab. 6.2. Among all GAQIC delay sets, C
shows the most accurate quantification and needs only half the time compared
to Q-INEPT-CT of Makela et al. T1 and T2 relaxation times of all the com-
pounds are measured and is well within the limits of quantification (T1, T2)= (9.8, 2.5)P , (6.7, 1.7)M , (10.4, 3.9)D seconds. Experimental implementation of
spectral editing using S4 (four scan delay set) is shown in Fig. 6.7. It is seen that
CH2 peaks are almost absent in Fig. 6.7(b) and hence S4 delay set can be used
for spectral simplification and quantification in complex mixtures. Delay set S16
(sixteen scan) should be used for high accuracy of quantification.
Chapter 6. GA Optimized Quantitative INEPT (GAQIC) 101
Conclusion
Using Genetic algorithm optimization with various fitness functions, we have gen-
erated optimized delay sets, which can perform faster and accurate quantification.
Experimental time advantage of these delay sets are shown by performing GAQIC
experiments in a mixture of α-methoxy phenyl acetic acid, 1-2-dimethyl cyclohex-
ane and 4-penten 2 ol. We also discussed quantitative spectral editing technique
(spectral simplification and quantification) for complex mixtures.
Appendix A
Unitary Decomposition of
Quantum Gates Using GA
Optimization
Here we discuss optimal experimental unitary decomposition of some of the impor-
tant quantum gates as well as state preparations using Genetic Algorithm. The
theory of UOD using GA is explained in Chapter 2.
The representation format of all the UODs in this appendix is as follows,
(i) θ(φ)n represents a θ length pulse with phase φ on nth spin (Eqn. A.1),
Rn(θ, φ) = exp(−iθ × [Cosφ Inx + Sinφ Iny ]). (A.1)
(ii) Uzz(Θ) represents spin system evolution under scalar coupling (Eqn. A.2).
Uzz(Θ) = exp(−i Θ 2IzIz). (A.2)
A.1 Controlled Hadamard (C-H)
Controlled Hadamard is one of the basic two qubit controlled operation [128].
The action of the gate is as follows - “If the Control qubit state is |1〉, apply
103
Appendix A. Quantum Gates using GA Optimization 104
Figure A.1: Pulse sequence for Controlled Hadamard.
a Hadamard operation on the target qubit.”. The UOD of C-H using GA is an
operator optimization (discussed in Chapter 2) with target unitary (Eqn. A.3),
C-H =
1 0 0 0
0 1 0 0
0 0 1/√
2 −1/√
2
0 0 1/√
2 1/√
2
. (A.3)
The optimized decomposition / pulse sequence is obtained as (Eqn. A.4 & Fig.
A.1),
C-H = 135(90)2 : Uzz(90) : 128.63(143.10)2 :
180(135)1 : 90(118)2.(A.4)
The pulse sequence has four SQRs and single 1/2J evolution under spin scalar
coupling.
A.2 Controlled-Controlled-NOT (CCNOT) Gate
Controlled-Controlled-NOT or Toffoli gate (Eqn. A.5) [128, 129] is a three qubit
universal reversible gate proposed by Tommaso Toffoli in 1980. Toffoli gate has two
control qubits and a target qubit. The action of Toffoli gate can be defined as (Eqn.
A.5) - “If the two control qubits are at |11〉, apply ‘NOT’ operation to the target”.
The variants of Toffoli gate based on the different states – |00〉, |01〉, |10〉, |11〉– of the control qubits are CCNOT(1,2) 3, CCNOT(1,2) 3, CCNOT(1,2) 3,
Appendix A. Quantum Gates using GA Optimization 105
Figure A.2: Pulse sequence for Toffoli Gate.
Operator φ1 φ2 φ3 φ4 φ5
CCNOT(1,2) 3 3π/8 3π/4 7π/8 π/4 π/2CCNOT(1,2) 3 5π/8 5π/4 15π/8 3π/4 π/2CCNOT(1,2) 3 11π/8 3π/4 π/8 5π/4 3π/2CCNOT(1,2) 3 π/2 π/4 π/8 3π/4 3π/2
Table A.1: Phase modulation values for different Toffoli gates
CCNOT(1,2) 3.
CCNOT(1,2) 3 =
1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0
0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0
0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 1
0 0 0 0 0 0 1 0
. (A.5)
Using GA optimization, we have obtained the CCNOT gate decomposition and is
shown in Fig. A.2. Different variants can be obtained by modulating the phases
φi=1...5 as shown in Tab. A.1.
A.3 Controlled SWAP (C-S)
Controlled SWAP gate or Fredkin gate (Eqn. A.3) [128, 129] is a three qubit
reversible operation invented by Ed Fredkin. The action of Controlled SWAP
Appendix A. Quantum Gates using GA Optimization 106
Figure A.3: Controlled-SWAP decomposition using C-NOT gates.
(Eqn. A.6) is as follows - “If the control qubit state is |1〉, SWAP (Eqn. A.7) the
states of two target qubits.”.
C-S =
[I4×4 0
0 USWAP
], (A.6)
where I4×4 is the identity matrix of dimension 4 and
USWAP =
1 0 0 0
0 0 1 0
0 1 0 0
0 0 0 1
(A.7)
Using the C-NOT decomposition of SWAP gate (Fig. A.3), Controlled- SWAP
can be built using three Controlled-Controlled-NOT gates. The heavy resource
requirement (3×CCNOT; Fig. A.2) of this decomposition can lead to experimental
errors. The optimized decomposition using GA is obtained as(Eqn. A.8 & Fig.
A.4) ,
C-S = 90(90)2 : U13zz (90) : 180(22.5)1 : U23
zz (90) : 90(180)2 :
45(0)3 : U13zz (90) : U23
zz (90) : U12zz (90) : 90(90)3 : U23
zz (45) :
U13zz (45) : 90(90)3 : 180(112.5)2 : 90(270)2 : 45(180)3 :
U23zz (45) : 90(180)2 : 180(0)1.
(A.8)
A.4 Quantum Counter circuits
Quantum counter operator transforms an nth quantum state in the computational
basis to (n+ s)th state, where s is the step size. The basic counter circuit for an n
qubit can be made using Controlled NOT Gates – three qubit example is shown
in Fig. A.5.
Appendix A. Quantum Gates using GA Optimization 107
Figure A.4: Pulse sequence for Controlled SWAP gate.
Figure A.5: Three qubit quantum counter using CC-NOT and C-NOT Gates.
Initial State Final State|00〉 |01〉|01〉 |10〉|10〉 |11〉|11〉 |00〉
Table A.2: Truth table for two qubit quantum counter.
The action of two qubit counter operator with step 1 (U2C1) is shown in Tab. A.2.
The operator matrix for two qubit quantum counter (Tab. A.2) is given by,
U2C1 =
0 0 0 1
1 0 0 0
0 1 0 0
0 0 1 0
The optimized unitary decomposition for the two qubit counter is obtained as
(Eqn. A.9),
U2C1 = 180(45)2 : 90(90)1 : Uzz(90) : 90(0)1 : 180(45)1. (A.9)
Appendix A. Quantum Gates using GA Optimization 108
Figure A.6: Pulse sequence for Two Qubit Counter.
The pulse sequence (Fig. A.6) has four SQRs and a single delay of length 1/2J .
The operator matrix for three qubit quantum counter (U3C1 – counting one at a
time) is given by (Eqn.A.10),
U3C1 =
0 0 0 0 0 0 0 1
1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0
0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0
0 0 0 0 0 1 0 0
0 0 0 0 0 0 1 0
(A.10)
The optimized decomposition is obtained as (Eqn. A.11),
U3C1 = 90(90)1 : U12zz (45) : 90(45)2 : U13
zz (45) : U23zz (90) :
180(112.5)1 : 180(90)2 : 90(45)3 : 180(112.5)3 :
U12zz (45) : 90(90)3.
(A.11)
The pulse sequence (Fig. A.7) has seven SQRs and four scalar coupling evolutions.
The decomposition using Controlled NOT gates as shown in Fig. A.5 requires more
resources.
Counting two at time (U3C2) can be easily implemented using the following rela-
tion,
U3C2 = (U2C1 ⊗ I2×2), (A.12)
Appendix A. Quantum Gates using GA Optimization 109
Figure A.7: Pulse sequence for Three Qubit Counter.
where I2×2 is the identity matrix and U2C1 is given in Eqn. A.9. Likewise counting
three at a time can implemented by,
U3C3 = U3C1 × (U2C1 ⊗ I2×2). (A.13)
A.5 W State Creation
W state is a three qubit maximally entangled state [128] and is given by (Eqn.A.14),
|W 〉 =1√3
(|001〉+ |101〉+ |100〉). (A.14)
W state is robust against particle loss error and is a good candidate for ensemble
based quantum memories [130].
Starting from |000〉 state, the optimized unitary decomposition for W state cre-
ation using GA is obtained as (Eqn. A.15),
UW = U23zz (90) : U12
zz (90) : 90(0)1 : 90(180)2 : U23zz (45) :
U12zz (90) : 90(90)2 : 70.53(90)3 : 90(315)2 : 90(90)1.
(A.15)
The pulse sequence (Fig. A.8) has six SQRs and four neighbor scalar coupling
evolutions.
Appendix A. Quantum Gates using GA Optimization 110
Figure A.8: Pulse sequence for W state creation.
Figure A.9: Quantum circuit for probability splitter.
Figure A.10: Pulse sequence for Quantum Probability Splitter.
A.6 Quantum Probability Splitter (QPS)
Quantum Probability Splitter (QPS) modifies the probability amplitudes of a qubit
so that the probability on the selected basis state is halved [131]. Quantum state
creation is the potential application of QPS. The circuit for quantum probability
splitter is shown in Fig. A.9 [131].
Appendix A. Quantum Gates using GA Optimization 111
The optimized decomposition using GA is obtained as (A.16),
UQPS = 90(45)2 : U23zz (90) : 90(315)2 : 90(225)3 : U12
zz (45) :
U23zz (90) : U13
zz (90) : 60(22.5)1 : 225(45)2 : 90(45)3 :
U23zz (90) : 120(22.5)1 : 135(135)2 : U12
zz (90) :
U23zz (135) : 180(0)1.
(A.16)
The pulse sequence (Fig. A.10) uses nine SQRs and six scalar coupling evolution
delays.
Appendix B
Matlab UOD Package
Experimental implementation of a quantum gate or an unitary operator (U) re-
quires decomposition of U , in terms of experimentally available system unitaries
– such as rf pulses and evolution of zz interaction (in NMR). This decomposition
should be optimal in terms of available resources to reduce experimental errors.
A Pulse sequence in Nuclear Magnetic Resonance can be considered as a decom-
posed unitary representation, where main unitary (U) is given by the equivalent
average Hamiltonian. Unitary operator decomposition or Pulse sequence gener-
ation of a unitary operator, U can be treated as an optimization problem by
defining a function (F) in such a way that the function profile of F defined by
pulse sequence parameters must have a minimum or maximum point for best pos-
sible pulse sequence or unitary decomposition of U . Here using Genetic Algorithm
optimization, we describe a scheme for generating optimal pulse sequence or uni-
tary decomposition for any unitary operator. Genetic algorithm optimization is a
global search method inspired by nature’s evolutionary process [34]. Genetic al-
gorithm has been used in NMR for designing new experiments [40], for improving
excitation and inversion accuracy of rf pulses [41] and pulse sequence optimization
[42].
The details of GA optimization such as representation scheme and fitness function
are defined in Chapter 2. The Matlab package for Unitary Operator Decomposition
(UOD) is explained here with a typical example of three qubit target operator -
Controlled SWAP (C-S). Controlled SWAP applies SWAP operation between two
113
Appendix B. Matlab UOD package 114
target qubits if the control qubit is in |1〉 state (Eqn. B.1).
Target operator, C-S =
1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0
0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0
0 0 0 0 0 0 1 0
0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 1
(B.1)
This target operator can be replaced by any three qubit unitary operator for
performing its UOD. The output of the program should be decoded using the
representation scheme (explained in Chapter 2). The program package given here
is a three qubit variant of UOD. The generalization of the program to any qubit
is in development stage. The UOD Matlab package has 11 subroutines working
together with Matlab’s optimization toolbox optimtool [132, 133]. All programs
are developed in Matlab® 7.14, 32 bit Ubuntu 12.04 OS with core i5 3.6 GHz
Intel processor.
UOD optimization package performs multiple GA optimizations in a loop with
population size and number of generations are defined in uod.m. The convergence
checking of the optimization and applying constraints on the population are out-
side these GA segments. Matlab subroutine list used in UOD package is explained
below,
1. uod.m: This is the main user interface program. Target operator and other
problem specific Genetic algorithm optimization details – such as population
size, generation, initial population etc. – are specified here.
Problem specific variables used in ‘uod.m’ are,
utar : Target unitary operator.
utarName : Name of the target unitary operator.
Population-
Size Data
: number of individuals in the population.
Generation-
s Data
: Number of generations.
Appendix B. Matlab UOD package 115
segno : number of segments or basic unit of pulse sequence (see
Chapter 2).
ctoffx : used for checking the convergence and termination. If the
fitness value difference between two consecutive run is less
than ctoffx, the program will count it into the termination
counter.
lpno : number of GA optimization segments.
stopat : Maximum fitness value (in %) to be satisfied for basic op-
timizations. Refining the solution segment of the program
will make near 100% fidelity.
initialmutaion-
ratio
: Initial value for mutation ratio. This value will change
according to the convergence during the Optimization.
initial-
population
: Specify the initial population here. GA will generate a
random population if initialpopulation field is blank.
EliteCount-
Number
: Specifies the number of elite individuals. Elite generation
is explained in Chapter 1.
MutationSwit-
ching flag
: Switch the subroutine MutationSwitching on. This will
change the basic mutation function of basic GA optimiza-
tion alternatively between uniform and adaptivefeasible
inside main loop.
ApplyGaxAppr-
oximations flag
: Switch the subroutine ApplyGaxApproximations on. This
will round each pulse parameters in the individual with
0, π/2, π.
GaxKicking-
flag
: Switch the subroutine GaxKicking on. This applies con-
stant perturbation to the final individual (gax), by scaling
up or down. ApplyGaxApproximations and GaxKicking
helps to keep the optimization away from local minimas.
2. uodmain.m: The main program which controls the optimization flow. Set-
ting up of initial population opening new data file for logging the optimiza-
tion are performed here. The basic Optimization flow is as follows,
(i) Direct initial optimization with all resources available. This usually
results in a pulse sequence with more number of delays and pulses than
optimal value required.
Appendix B. Matlab UOD package 116
(ii) ‘Delay Zeroing ’ procedure: to reduce the length of the pulse sequence
obtained from Step.(i).
(iii) ‘Pulse Zeroing ’ procedure: to reduce the number of pulses in the pulse
sequence obtained from Step.(ii).
(iv) Final refining of the solution obtained from Step. (iii). This will make
the fidelity near to 100%
3. SpinOperators.m: Basic spin product operators defined here.
4. GADefaults.m: Sets the default values of Genetic Algorithm optimization.
The main default values set in UOD are,
(i) ‘StallGenLimit Data’ is kept as number of generations. (The algorithm
stops if the weighted average relative change in the best fitness function
value over Stall generations is less than or equal to Function tolerance).
(ii) ‘TolFun Data=0’. (The algorithm stops if the weighted average relative
change in the best fitness function value over Stall generations is less
than or equal to Function tolerance[133]).
(iii) TolCon Data=0. (It is used to determine the feasibility with respect to
nonlinear constraints [133]).
(iv) Cross over function is set as crossoverheuristic (explained in Chapter
1).
(v) Parental selection function is set as selectionroulette (explained in Chap-
ter 1).
5. uodFitness.m: Fitness function of the GA optimization. The fitness function
used for UOD is explained in Chapter 2.
6. uodFitnessPulseO.m: The fitness function used during pulse zeroing.
7. ApplyGaxConstraints.m: Sets the conditions on pulse parameters such as
maximum value of delay length and rotation angle of pulse.
8. optimize delay.m: Subroutine for delay zeroing – to reduce the pulse se-
quence length.
9. MutationSwitching.m: This change the basic mutation function of GA opti-
mizations inside the main loop alternatively between uniform and adaptive-
feasible(Chapter 1). This action found to be converging faster than single
mutation function.
Appendix B. Matlab UOD package 117
10. GaxKicking.m: Applies a small perturbation in the form of scaling up or
down or adding small random number to best individual in the population.
This action helps to keep optimization out of local minimas.
11. uniform.m: Applies the mutation function uniform to the population. Func-
tion uniform is explained in Chapter 1.
The individual pulse sequences has ‘segno’ of pulse-delay segments, which makes
it ‘9 × segno’ number of pulse sequence parameters and is represented by an
array ‘gax ’ (as given in matlab code ‘uodFitness.m’). The first 6 × segno of gax
elements represents the pulse parameters (odd numbers pulse angles and even
numbers pulses phases) and remaining elements represents the delay segments,
which is again devided into groups of three elements with first element in a group
represents the coupling evolution of spins 1 and 2, second element represents the
coupling evolution of spins 2 and 3 and third element represents coupling evolution
of spins 1 and 3.
Appendix B. Matlab UOD package 118
uod.m
c l e a r a l l2 c l c
g l o b a l segno utar4 g l o b a l i n i t i a l m u t a i o n r a t i o dt in
g l o b a l Populat ionSize Data Generat ions Data c t o f f x lpno6 g l o b a l r e s e t count s topat i n i t i a l p o p u l a t i o n EliteCountNumber
utarNameg l o b a l ApplyGaxApproximations f lag Mutat ionSwi tch ing f l ag
GaxKick ing f lag8 g l o b a l x y z I I
t i c10 x=[0 1 ;1 0 ] / 2 ; y=[ 0 −1 i ; 1 i 0 ] / 2 ; z =[1 0 ; 0 −1]/2; I I=eye (2 )
;
12 % Def ine Target operator −− utarutarName = ’ cswap ’ ; % name o f gate
14 utar = [ . . .1 0 0 0 0 0 0 0
16 0 1 0 0 0 0 0 00 0 1 0 0 0 0 0
18 0 0 0 1 0 0 0 00 0 0 0 1 0 0 0
20 0 0 0 0 0 0 1 00 0 0 0 0 1 0 0
22 0 0 0 0 0 0 0 1 ] ;
24 % D e t a i l s o f Genet ic Algorithm ParametersPopulat ionSize Data =40;
26 Generat ions Data =60;segno =9; % number o f segments in pu l s e sequence
28 c t o f f x =0.0125; % Zero approximation c u t o f f f o r parameter ini n d i v i d u a l s
lpno =999999; % I t e r a t i o n s no o f above (Populat ionSize Data , Generat ions Data )
30 r e s e t count =5; % Reset to i n i t i a l i f f i t n e s s cont inues f o rr e s e t count t imes
s topat =96.0 ;32 i n i t i a l m u t a i o n r a t i o =0.11;
i n i t i a l p o p u l a t i o n=pi /2* ones (1 , segno *9) ;34 EliteCountNumber=2;
Mutat ionSwi tch ing f l ag = 1 ;36 ApplyGaxApproximations f lag = 1 ;
GaxKick ing f lag =1;38 dt in = d a t e s t r (now , ’mmmm dd , yyyy HH:MM: SS .FFF AM’ ) ;
d i sp ( [ ’ Optimizat ion Started : ’ dt in ] ) ;40 di sp ( ’ Target Unitary Operator : ’ ) ;
d i sp ( num2str ( utar ) ) ;42 di sp ( [ ’ Populat ion S i z e : ’ num2str ( Populat ionSize Data ) ] ) ;
Appendix B. Matlab UOD package 119
di sp ( [ ’ Generat ions : ’ num2str ( Generat ions Data ) ] ) ;44 di sp ( [ ’ E l i t e Count : ’ num2str ( EliteCountNumber ) ] ) ;
i f Mutat ionSwi tch ing f l ag ; d i sp ( ’ Mutation Switching ON’ ) ; end ;46 i f ApplyGaxApproximations f lag ; d i sp ( ’Gax Approximations ON’ ) ;
end ;uodmain ; % main program with d e f a u l t va lue s o f GA
opt imiza t i on48 dt = d a t e s t r (now , ’mmmm dd , yyyy HH:MM: SS .FFF AM’ ) ;
toc
uodmain.m
f unc t i on [ gax , f v a l ] = uodmain2 g l o b a l segno utar utarName
g l o b a l lpno r e s e t count s topat opt ions4 g l o b a l d l z f l g gaxsave i d x d l y z e r o s idxd lynonzeros d lyar ray
i n i t i a l p o p u l a t i o ng l o b a l ApplyGaxApproximations f lag Mutat ionSwi tch ing f l ag
GaxKick ing f lag6 g l o b a l p l z f l g p l s a r r a y d e l a y s t a r t delayend dt in i d x p l s z e r o s
g a x f i n a l delaysum
8 % c r e a t i n g d a t a f i l e f o r output wr i t i ngf i l e I D = fopen ( [ utarName ’ . txt ’ ] , ’ a ’ ) ;
10
f p r i n t f ( f i l e I D , [ ’ \n\n−−−−−−−−New Optimisation−−−−−−−−−−\nStarted On: ’ dt in ’ \n ’ ] ) ;
12
SpinOperators % Three sp in Product ope ra to r s14
utar=utar ’ ; % us ing in f i t n e s s eva lua t i on16 nvars=9*segno ;
18 f v a l 1 =0;d e l a y s t a r t =6*( segno ) +1; delayend=nvars ;
20 dlyar ray=d e l a y s t a r t : delayend ;p l s a r r a y =1:2 : de l ay s ta r t −1;
22
GADefaults % Defau l t va lue s f o r GA parameters24
% I n i t i a l Populat ion S e t t i n g s . . . . . . .26 i f isempty ( i n i t i a l p o p u l a t i o n )
opt ions = gaopt imset ( opt ions , ’ I n i t i a l P o p u l a t i o n ’ , 2* pi *rand (1 , nvars ) ) ; gax=2*pi * rand (1 , nvars ) ;
28 e l s eopt ions = gaopt imset ( opt ions , ’ I n i t i a l P o p u l a t i o n ’ ,
i n i t i a l p o p u l a t i o n ) ; gax=i n i t i a l p o p u l a t i o n ;30 end
Appendix B. Matlab UOD package 120
g a x f i n a l=gax ;32 r e s e t chk ( 1 : 1 0 0 ) =0;
d l z f l g =0; endde layze ro ing =0; endpu l s e z e ro ing =0; p l z f l g =0;34 f p r i n t f ( [ ’ \n\nStep (/ ’ num2str ( lpno ) ’ ) \ t F i tne s s \n ’ ] ) ;
f p r i n t f ( ’−−−−−−−−−−−−−−−−−−−− \n ’ ) ;36 f o r cnt =1: lpno
i f endde layzero ing && endpu l s e z e ro ing38 break
e l s e40 [ gax , f va l , ˜ , ˜ , ˜ , ˜ ] = . . .
ga ( @uodFitness , nvars , [ ] , [ ] , [ ] , [ ] , [ ] , [ ] , [ ] , [ ] , opt i ons ) ;42 gax=ApplyGaxConstraints ( gax ) ;
f v a l =12.5* abs ( f v a l ) ;44 f p r i n t f ( [ num2str ( cnt ) ’ \ t \ t \ t ’ num2str ( f v a l ) ’ \n ’ ] ) ;
f p r i n t f ( f i l e I D , ’ \nloop no : %d/%d , f i t n e s s va lue : %5.2 f’ , cnt , lpno , f v a l ) ;
46 end
48 % Stopping C r i t e r i o ni f f va l> s topat
50 di sp ( [ ’ f i t n e s s va lue : ’ num2str ( f v a l ) ] )f p r i n t f ( f i l e I D , ’ \ nFi tnes s Value : %5.2 f \nBest
I n d i v i d u a l : ’ , f v a l ) ;52 f p r i n t f ( f i l e I D , ’%f ’ , gax ) ;
i f ˜ p l z f l g54 opt im i z e de l ay % f o r z e ro ing o f de lays
end56
i f endde layzero ing58 t imetaken= toc ;
f p r i n t f ( f i l e I D , [ ’ \n\nTime Taken f o rOptimisat ion : ’ num2str ( t imetaken ) ’ Seconds ’ ] ) ;
60 d t f i = d a t e s t r (now , ’mmmm dd , yyyy HH:MM: SS .FFFAM’ ) ;
d i sp ( d t f i )62 f p r i n t f ( f i l e I D , [ ’ \nFinished at : ’ d t f i ’ \n
−−−−−−−−End o f Optimisation−−−−−−−−−−\n ’ ] ) ;break
64 endend
66
i f ApplyGaxApproximations f lag68 gax=ApplyGaxApproximations ( gax ) ; % Aproximations f o r 0 ,
p i /2 and pi parametersend
70
% Checking Reset c o n d i t i o n s72 i f abs ( fva l1−f v a l ) < 1
f i d e l i t y n o=round ( f v a l ) ;
Appendix B. Matlab UOD package 121
74 r e s e t chk ( f i d e l i t y n o )=re s e t chk ( f i d e l i t y n o ) +1;i f r e s e t chk ( f i d e l i t y n o ) > ( r e s e t count ) | | f v a l > s topat
76 i f ( d l z f l g && re s e t chk ( f i d e l i t y n o ) > 2* c e i l ( f v a l/15) ) | | f v a l > s topat % ” | | f v a l > s topat ” i s nece s sa ryf o r the case o f s p e c i f i e d f u l l s e t o f i d x d l y z e r o s andidxd lynonzeros
i f f v a l >= stopat78 di sp ( ’ Delay ze ro ing cont inu ing . ! ! ! ’ ) ;
gax=gaxsave ;80 g a x f i n a l=gax ;
e l s e82 di sp ( ’ Delay ze ro ing f a i l e d ! ! ! . ’ ) ;
idxd lynonzeros =[ idxd lynonzeros i d x d l y z e r o s (l ength ( i d x d l y z e r o s ) ) ] ;
84 gax=gaxsave ; i d x d l y z e r o s=i d x d l y z e r o s ( 1 :l ength ( i d x d l y z e r o s )−1) ;
end86
i f l ength ( [ i d x d l y z e r o s idxd lynonzeros ] ) ==delayend − d e l a y s t a r t + 1
88 d i s p l a y ( [ ’ F i d e l i t y : ’ num2str ( abs (uodFitness ( gax ) ) *12 .5 ) ] ) ;
gax ( i d x d l y z e r o s )=ze ro s (1 , l ength ( i d x d l y z e r o s) ) ;
90 d i s p l a y ( [ ’ Zeroed i n d i c e s = ’ num2str (i d x d l y z e r o s ) ] )
endde layze ro ing =1;92 p l z f l g =1;
end94
e l s e i f (˜ d l z f l g && f i d e l i t y n o >= 89 && res e t chk (f i d e l i t y n o ) > ( r e s e t count +3*( f i d e l i t y n o −89)+0* d l z f l g *5) ) | |( ˜ d l z f l g && f i d e l i t y n o < 89 && re s e t chk ( f i d e l i t y n o ) >r e s e t count ) % ”+ d l z f l g *10” added f o r g i v i ng morei t e r a t i o n s in z e ro ing proce s s . . .
96 di sp ( ’ r e s e t i n g . . . . . . ’ ) ;f p r i n t f ( f i l e I D , ’ \nReset ing . . . . . \ n Best
I n d i v i d u a l : ’ ) ;98 f p r i n t f ( f i l e I D , ’%f ’ , gax ) ;
gax=pi * rand (1 , nvars ) ; r e s e t chk ( 1 : 1 0 0 ) =0;d l z f l g =0;
100 endend
102
% a l t e r n a t e l y g i v ing d i f f e r e n t mutation func t i on with d i f f e r e n tparameters
104 i f Mutat ionSwi tch ing f l agMutationSwitching ( fva l , cnt )
106 end
Appendix B. Matlab UOD package 122
108 % f o r s t a b l e i n d i v i d u a li f GaxKick ing f lag
110 gax=GaxKicking ( gax , f v a l ) ;end
112
e l s e i f f v a l 1 > f v a l114 di sp ( ’NOT A STABLE POINT . . . . Ro l l i ng back with Mutation
Guassian . . . . . . ’ )opt i ons = gaopt imset ( opt ions , ’ MutationFcn ’ ,
@mutationgaussian 1 .23 1 ) ;116 end
118 f v a l 1=f v a l ;opt i ons = gaoptimset ( opt ions , ’ I n i t i a l P o p u l a t i o n ’ , gax ) ;
120 endgax=g a x f i n a l ;
122 g a x f i n a l = ApplyGaxConstraints ( g a x f i n a l ) ;f p r i n t f ( [ ’DELAY ZEROING FINISHED . . . . . . . . . . \n Fina l F i d e l i t y
After Delay Zero ing i s : ’ num2str (−12.5* uodFitness ( g a x f i n a l )) ] ) ;
124 f p r i n t f ( [ ’ \n Fina l GAX i s : ’ num2str ( g a x f i n a l ) ] ) ;
126 delaysum = sum( g a x f i n a l ( d e l a y s t a r t : delayend ) ) ;
128 f p r i n t f ( ’ \n\n\n\n\n %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%\n Pulse Zero ing Started . . . . \ n ’ ) ;
c n t p l s =1;130 numof i ter =0; f l g p l s c h a n g e =1;
i d x p l s z e r o s = [ ] ; numof i ter ( 1 : 1 0 0 ) =0;132 opt ions = gaopt imset ( opt ions , ’ MutationFcn ’ ,
@mutat ionadapt feas ib le ) ;f o r k=cnt : lpno
134 i f f l g p l s c h a n g ed i sp ( [ ’ Zeroed Pulse I n d i c e s : ’ num2str ( i d x p l s z e r o s ) ] ) ;
136
% S e l e c t i n g best pu l s e parameter from p l s a r r a y to makeze ro ing
138 PLS=[ g a x f i n a l ( p l s a r r a y ) ; p l s a r r a y ] ;PLS=sort rows (PLS . ’ , 1 ) . ’ ;
140 p l s a r r a y=PLS ( 2 , : ) ;i d x p l s z e r o s =[ i d x p l s z e r o s p l s a r r a y ( c n t p l s ) ] ;
142
c n t p l s=c n t p l s +1; f l g p l s c h a n g e =0;144 end
[ gax , fva l , ˜ , ˜ , ˜ , ˜ ] = . . .146 ga ( @uodFitnessPulseO , nvars , [ ] , [ ] , [ ] , [ ] , [ ] , [ ] , [ ] , [ ] ,
opt i ons ) ;gax=ApplyGaxConstraints ( gax ) ;
148 f v a l =12.5* abs ( f v a l ) ;numof i ter ( round ( f v a l ) )=numof i ter ( round ( f v a l ) ) +1;
Appendix B. Matlab UOD package 123
150 f p r i n t f ( [ num2str ( k ) ’ \ t \ t \ t ’ num2str ( f v a l ) ’ \n ’ ] ) ;f p r i n t f ( f i l e I D , ’ \nloop no : %d/%d , f i t n e s s va lue : %5.2 f
’ , cnt , lpno , f v a l ) ;152 opt ions = gaopt imset ( opt ions , ’ I n i t i a l P o p u l a t i o n ’ , gax ) ;
i f f v a l > 97 % stopat154 g a x f i n a l=gax ;
opt ions = gaopt imset ( opt ions , ’ I n i t i a l P o p u l a t i o n ’ ,g a x f i n a l ) ;
156 f l g p l s c h a n g e =1;numof i ter ( 1 : 1 0 0 ) =0;
158 i f c n t p l s > l ength ( p l s a r r a y ) ; break ; ende l s e
160 i f numof i ter ( round ( f v a l ) ) > 5f l g p l s c h a n g e =1;
162 opt ions = gaopt imset ( opt ions , ’ I n i t i a l P o p u l a t i o n ’ ,g a x f i n a l ) ;
d i sp ( ’ Pulse z e ro ing f a i l e d ! ! ! ’ ) ;164 i d x p l s z e r o s=i d x p l s z e r o s ( 1 : end−1) ;
numof i ter ( 1 : 1 0 0 ) =0;166 i f c n t p l s > l ength ( p l s a r r a y ) ; break ; end
end168 end
end170 i d x p l s z e r o s=s o r t ( [ i d x p l s z e r o s i d x p l s z e r o s +1]) ;
g a x f i n a l = ApplyGaxConstraints ( g a x f i n a l ) ;172
%%%%%%%%%% FINAL REFINING THE SOLUTION %%%%%%%%%%%%%%174 di sp ( ’ **** REFINING THE SOLUTION **** ’ )
c1=k ; c2 =1;176 whi le ( f v a l < 9 8 . 5 ) && ( c2 < 100)
[ gax , fva l , ˜ , ˜ , ˜ , ˜ ] = . . .178 ga ( @uodFitness , nvars , [ ] , [ ] , [ ] , [ ] , [ ] , [ ] , [ ] , [ ] , opt ions ) ;
gax=ApplyGaxConstraints ( gax ) ;180 f v a l =12.5* abs ( f v a l ) ;
f p r i n t f ( [ num2str ( c1 ) ’ \ t \ t \ t ’ num2str ( f v a l ) ’ \n ’ ] ) ;182 f p r i n t f ( f i l e I D , ’ \nloop no : %d/%d , f i t n e s s va lue : %5.2 f
’ , cnt , lpno , f v a l ) ;opt i ons = gaopt imset ( opt ions , ’ I n i t i a l P o p u l a t i o n ’ , gax ) ;
184 c1=c1 +1; c2=c2 +1;end
186
%%%%%%%%%%% PRINTING THE SOLUTION %%%%%%%%%%%%%%188 g a x f i n a l = ApplyGaxConstraints ( gax ) ;
d i s p l a y ( [ ’ Zeroed p u l s e s are : ’ num2str ( i d x p l s z e r o s ) ] ) ;190 d i s p l a y ( [ ’ F ina l F i d e l i t y i s : ’ num2str ( abs ( uodFitness ( g a x f i n a l
) *12 .5 ) ) ] ) ;d i s p l a y ( [ ’ F ina l GAX i s : ’ num2str ( g a x f i n a l ) ] ) ;
Appendix B. Matlab UOD package 124
SpinOperators.m
f unc t i on SpinOperators2 g l o b a l x y z I I x i i y i i i x i i y i i i x i i y z z i i z z z i z z i i i z i
i i zx i i=mkron (x , I I , I I ) ; y i i=mkron (y , I I , I I ) ;
4 i x i=mkron ( I I , x , I I ) ; i y i=mkron ( I I , y , I I ) ;i i x=mkron ( I I , I I , x ) ; i i y=mkron ( I I , I I , y ) ;
6 z i i=diag (mkron ( z , I I , I I ) ) ; i z i=diag (mkron ( I I , z , I I ) ) ; i i z=diag (mkron ( I I , I I , z ) ) ;
z z i=diag (mkron ( z , z , I I ) ) ; z i z=diag (mkron ( z , I I , z ) ) ; i z z=diag (mkron ( I I , z , z ) ) ;
8 end
GADefaults.m
1 f unc t i on GADefaultsg l o b a l opt ions
3 g l o b a l Populat ionSize Data Generat ions Datag l o b a l i n i t i a l m u t a i o n r a t i o EliteCountNumber
5 Stal lGenLimit Data=Generat ions Data ;TolFun Data=0;TolCon Data=0; opt ions = gaopt imset ;
7 opt ions = gaopt imset ( opt ions , ’ TolFun ’ , TolFun Data ) ;opt i ons = gaopt imset ( opt ions , ’ TolCon ’ , TolCon Data ) ;
9 opt ions = gaopt imset ( opt ions , ’ CrossoverFcn ’ , @ c r o s s o v e r h e u r i s t i c [ ] ) ;
opt i ons = gaopt imset ( opt ions , ’ Popu lat ionS ize ’ ,Populat ionSize Data ) ;
11 opt ions = gaopt imset ( opt ions , ’ Generat ions ’ ,Generat ions Data ) ;
opt i ons = gaopt imset ( opt ions , ’ Sta l lGenLimit ’ ,Stal lGenLimit Data ) ;
13 opt ions = gaopt imset ( opt ions , ’ Se l e c t i onFcn ’ ,@ s e l e c t i o n r o u l e t t e ) ;opt i ons = gaopt imset ( opt ions , ’ Display ’ , ’ o f f ’ ) ;
15 opt ions = gaopt imset ( opt ions , ’ El i teCount ’ , EliteCountNumber) ;
opt i ons = gaopt imset ( opt ions , ’ Vector i zed ’ , ’ o f f ’ ) ;17 opt ions = gaopt imset ( opt ions , ’ Us eP ar a l l e l ’ , ’ always ’ ) ;
opt i ons = gaopt imset ( opt ions , ’ MutationFcn ’ ,@mutat ionadapt feas ib le ) ;
19 opt ions = gaopt imset ( opt ions , ’ El i teCount ’ , EliteCountNumber) ;
end
Appendix B. Matlab UOD package 125
uodFitness.m
1 f unc t i on f t n s = uodFitness ( gax )g l o b a l segno utar
3 g l o b a l x i i y i i i x i i y i i i x i i y z z i i z z z i z %z i i i z i i i zgax=ApplyGaxConstraints ( gax ) ;
5 u e f f=eye (8 ) ;f o r j =0: segno−1
7 count1=j *6 ;RfMat=−1i *( gax ( count1+1)*( cos ( gax ( count1+2) ) * x i i+s i n ( gax (
count1+2) ) * y i i ) + . . .9 gax ( count1+3)*( cos ( gax ( count1+4) ) * i x i+s i n ( gax ( count1+4) ) * i y i )
+ . . .gax ( count1+5)*( cos ( gax ( count1+6) ) * i i x+s i n ( gax ( count1+6) ) * i i y ) ) ;
11 [ v1 , d1]= e i g (RfMat) ;cc =6*( segno )+3* j ;
13 ZMat=exp(−1 i *( gax ( cc+1)* z z i+gax ( cc+2)* i z z+gax ( cc+3)* z i z ) ) ;u e f f =( u e f f *v1*diag ( exp ( diag ( d1 ) ) ) /v1 ) *diag (ZMat) ;
15 endf t n s=−(abs ( t r a c e ( utar * u e f f ) ) ) ; % utar here i s uatr ’
uodFitnessPulseO.m
f unc t i on f t n s = uodFitnessPulseO ( gax )2 g l o b a l segno utar d e l a y s t a r t delayend delaysum
g l o b a l x i i y i i i x i i y i i i x i i y z z i i z z z i z %z i i i z i i i z4 gax=ApplyGaxConstraints ( gax ) ;
u e f f=eye (8 ) ;6 f o r j =0: segno−1
count1=j *6 ;8 RfMat=−1i *( gax ( count1+1)*( cos ( gax ( count1+2) ) * x i i+s i n ( gax (
count1+2) ) * y i i ) + . . .gax ( count1+3)*( cos ( gax ( count1+4) ) * i x i+s i n ( gax ( count1+4) ) * i y i )
+ . . .10 gax ( count1+5)*( cos ( gax ( count1+6) ) * i i x+s i n ( gax ( count1+6) ) * i i y ) ) ;
[ v1 , d1]= e i g (RfMat) ;12 cc =6*( segno )+3* j ;
ZMat=exp(−1 i *( gax ( cc+1)* z z i+gax ( cc+2)* i z z+gax ( cc+3)* z i z ) ) ;14 u e f f =( u e f f *v1*diag ( exp ( diag ( d1 ) ) ) /v1 ) *diag (ZMat) ;
end16 i f (sum( gax ( d e l a y s t a r t : delayend ) ) ) <= ( delaysum+1)
f t n s=−(abs ( t r a c e ( utar * u e f f ) ) ) ;18 e l s e
f t n s =−1;20 end
Appendix B. Matlab UOD package 126
ApplyGaxConstraints.m
2 f unc t i on [ gaxout ] = ApplyGaxConstraints ( gaxin )g l o b a l d e l a y s t a r t delayend i d x d l y z e r o s i d x p l s z e r o s
4 gaxout ( 1 : ( de l ay s ta r t −1) )=mod( gaxin ( 1 : ( de l ay s ta r t −1) ) ,2* pi ) ;gaxout ( d e l a y s t a r t : delayend )=mod( gaxin ( d e l a y s t a r t : delayend ) , p i ) ;
6 gaxout ( [ i d x d l y z e r o s i d x p l s z e r o s ] ) =0;end
optimize delay.m
2 g l o b a l i d x d l y z e r o s d l z f l g idxd lynonzeros d lyar rayc l e a r z e ro idx gaxsave
4 ze ro idx = [ ] ;gaxsave=gax ;
6 l ngz ro=length ( d lyar ray ) ;f zp =0;
8 memberchk=˜ismember ( dlyarray , [ i d x d l y z e r o s idxd lynonzeros ] ) ;f o r k=1: lngz ro
10 i f memberchk ( k )g a x t r i a l=gax ;
12 g a x t r i a l ( d lyar ray ( k ) ) =0;f z=abs ( uodFitness ( g a x t r i a l ) ) ;
14 i f f z > f zpfzp=f z ;
16 ze ro idx=dlyar ray ( k ) ;end
18 endend
20 i d x d l y z e r o s =[ i d x d l y z e r o s z e ro idx ] ;d i sp ( ’ Delay Zero ing Started . . . Zeroed I n d i c e s are : ’ ) ;
22 di sp ( i d x d l y z e r o s )d l z f l g =1;
24 r e s e t chk ( 1 : 1 0 0 ) =0;
Appendix B. Matlab UOD package 127
MutationSwitching.m
1 f unc t i on MutationSwitching ( fva l , cnt )g l o b a l opt ions
3 switch (1+mod( cnt , 2 ) )case 1
5 i f f v a l > 90opt ions = gaopt imset ( opt ions , ’ MutationFcn ’ ,
@uniform (100− f v a l ) /80 ) ;7 e l s e
opt ions = gaopt imset ( opt ions , ’ MutationFcn ’ , @uniform rand (1) /5 ) ;
9 endcase 2
11 opt ions = gaopt imset ( opt ions , ’ MutationFcn ’ ,@mutat ionadapt feas ib le ) ;
end13 end
GaxKicking.m
f unc t i on [ gaxout ] = GaxKicking ( gaxin , f v a l )2 i f f v a l < 95
gaxout=gaxin *0 . 9 7 ;4 e l s e
gaxout=gaxin *0 . 9 9 ;6 end
end
Appendix B. Matlab UOD package 128
uniform.m
f unc t i on mutationChi ldren = uniform ( parents , opt ions ,GenomeLength , ˜ , ˜ , ˜ , th i sPopu lat ion , mutationRate )
2 mutationChi ldren = ze ro s ( l ength ( parents ) , GenomeLength ) ;f o r i =1: l ength ( parents )
4 c h i l d = th i sPopu la t i on ( parents ( i ) , : ) ;mutationPoints = f i n d ( rand (1 , l ength ( c h i l d ) ) <
mutationRate ) ;6 range = opt ions . PopInitRange ;
[ ˜ , c ] = s i z e ( range ) ;8 i f ( c ˜= 1)
range = range ( : , mutationPoints ) ;10 end
lower = range ( 1 , : ) ;12 upper = range ( 2 , : ) ;
span = upper − lower ;14 c h i l d ( mutationPoints ) = lower + rand (1 , l ength (
mutationPoints ) ) .* span ;mutationChi ldren ( i , : ) = c h i l d ;
16 end
Appendix B. Matlab UOD package 129
OUTPUT
Optimizat ion Started : September 18 , 2013 1 0 : 0 5 : 4 2 . 1 5 3 PM2 Target Unitary Operator :
1 0 0 0 0 0 0 04 0 1 0 0 0 0 0 0
0 0 1 0 0 0 0 06 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 08 0 0 0 0 0 0 1 0
0 0 0 0 0 1 0 010 0 0 0 0 0 0 0 1
Populat ion S i z e : 4012 Generat ions : 60
E l i t e Count : 214 Mutation Switching ON
Gax Approximations ON16
18 Step (/999999) F i tne s s−−−−−−−−−−−−−−−−−−−−
20 1 66 .3592 69.9054
22 3 72.04964 73.3215
24 5 73 .60714 74 .713
26 15 74.6462r e s e t i n g . . . . . .
28 16 64.583317 72.1024
30 18 73.651324 76.4163
32 25 77 .31626 78.3563
34 29 82.029431 87.4276
36 32 89.628434 93.6552
38 35 95.414236 96.5928
40 f i t n e s s va lue : 96 .5928Delay Zero ing Started . . . Zeroed I n d i c e s are :
42 73
44 37 97.5006f i t n e s s va lue : 97 .5006
46 Delay Zero ing Started . . . Zeroed I n d i c e s are :
Appendix B. Matlab UOD package 130
73 5748 Delay ze ro ing cont inu ing . ! ! !
38 97 .771150 f i t n e s s va lue : 97 .7711
Delay Zero ing Started . . . Zeroed I n d i c e s are :52 73 57 63
Delay ze ro ing cont inu ing . ! ! !54 39 98.4392
f i t n e s s va lue : 98 .439256 Delay Zero ing Started . . . Zeroed I n d i c e s are :
73 57 63 6658 Delay ze ro ing cont inu ing . ! ! !
40 98 .472660 f i t n e s s va lue : 98 .4726
Delay Zero ing Started . . . Zeroed I n d i c e s are :62 73 57 63 66 75
64 Delay ze ro ing cont inu ing . ! ! !41 98 .7079
66 f i t n e s s va lue : 98 .7079Delay Zero ing Started . . . Zeroed I n d i c e s are :
68 73 57 63 66 75 74Delay ze ro ing cont inu ing . ! ! !
70 42 98.9116f i t n e s s va lue : 98 .9116
72 Delay Zero ing Started . . . Zeroed I n d i c e s are :73 57 63 66 75 74 61
74 Delay ze ro ing cont inu ing . ! ! !43 99 .0491
76 f i t n e s s va lue : 99 .0491Delay Zero ing Started . . . Zeroed I n d i c e s are :
78 73 57 63 66 75 74 61 78Delay ze ro ing cont inu ing . ! ! !
80 44 99.1897f i t n e s s va lue : 99 .1897
82 Delay Zero ing Started . . . Zeroed I n d i c e s are :73 57 63 66 75 74 61 78 59
84 Delay ze ro ing cont inu ing . ! ! !45 99 .2914
86 f i t n e s s va lue : 99 .2914Delay Zero ing Started . . . Zeroed I n d i c e s are :
88 73 57 63 66 75 74 61 78 59 76Delay ze ro ing cont inu ing . ! ! !
90 46 99.3085f i t n e s s va lue : 99 .3085
92 Delay Zero ing Started . . . Zeroed I n d i c e s are :73 57 63 66 75 74 61 78 59 7679
94 Delay ze ro ing cont inu ing . ! ! !47 99 .4054
Appendix B. Matlab UOD package 131
96 f i t n e s s va lue : 99 .4054Delay Zero ing Started . . . Zeroed I n d i c e s are :
98 73 57 63 66 75 74 61 78 59 7679 64
Delay ze ro ing cont inu ing . ! ! !100 48 99.4561
f i t n e s s va lue : 99 .4561102 Delay Zero ing Started . . . Zeroed I n d i c e s are :
73 57 63 66 75 74 61 78 59 7679 64 60
104 Delay ze ro ing cont inu ing . ! ! !49 99 .4098
106 f i t n e s s va lue : 99 .4098Delay Zero ing Started . . . Zeroed I n d i c e s are :
108 73 57 63 66 75 74 61 78 59 7679 64 60 81
110 Delay ze ro ing cont inu ing . ! ! !50 99 .4813
112 f i t n e s s va lue : 99 .4813Delay Zero ing Started . . . Zeroed I n d i c e s are :
114 73 57 63 66 75 74 61 78 59 7679 64 60 81 68
Delay ze ro ing cont inu ing . ! ! !116 51 99.4404
f i t n e s s va lue : 99 .4404118 Delay Zero ing Started . . . Zeroed I n d i c e s are :
73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55
120 Delay ze ro ing cont inu ing . ! ! !52 99 .3623
122 f i t n e s s va lue : 99 .3623Delay Zero ing Started . . . Zeroed I n d i c e s are :
124 73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69
Delay ze ro ing cont inu ing . ! ! !126 53 98.0176
f i t n e s s va lue : 98 .0176128 Delay Zero ing Started . . . Zeroed I n d i c e s are :
73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69 67
130 54 97.5145f i t n e s s va lue : 97 .5145
132 Delay Zero ing Started . . . Zeroed I n d i c e s are :73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69 67 80
134 Delay ze ro ing cont inu ing . ! ! !55 98 .874
136 f i t n e s s va lue : 98 .874Delay Zero ing Started . . . Zeroed I n d i c e s are :
Appendix B. Matlab UOD package 132
138 73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69 67 80 71
140 56 98.9964f i t n e s s va lue : 98 .9964
142 Delay Zero ing Started . . . Zeroed I n d i c e s are :73 57 63 66 75 74 61 78 59 76
79 64 60 81 68 55 69 67 80 7162
144 Delay ze ro ing cont inu ing . ! ! !57 98 .4442
146 f i t n e s s va lue : 98 .4442Delay Zero ing Started . . . Zeroed I n d i c e s are :
148 73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69 67 80 7162 72
Delay ze ro ing cont inu ing . ! ! !150 58 91.5406
59 91.8223152 60 91.9674
70 92.3325154 71 92.2263
72 92.2929156 73 92 .191
Delay ze ro ing f a i l e d ! ! ! .158 74 98.9949
f i t n e s s va lue : 98 .9949160 Delay Zero ing Started . . . Zeroed I n d i c e s are :
73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69 67 80 7162 58
162 75 90.571276 91.4515
164 77 91.777384 92.2361
166 85 92 .12188 92.2745
168 90 92.240391 92.2614
170 Delay ze ro ing f a i l e d ! ! ! .92 99 .4279
172 f i t n e s s va lue : 99 .4279Delay Zero ing Started . . . Zeroed I n d i c e s are :
174 73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69 67 80 7162 70
176 93 89.130394 90.0468
178 95 90 .407110 92 .133
Appendix B. Matlab UOD package 133
180 111 92.0422112 92 .145
182 Delay ze ro ing f a i l e d ! ! ! .113 99.5394
184 f i t n e s s va lue : 99 .5394Delay Zero ing Started . . . Zeroed I n d i c e s are :
186 73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69 67 80 7162 65
114 80.4524188 115 82.9635
116 84.1121190 127 85.0483
128 85.1719192 129 85.0664
Delay ze ro ing f a i l e d ! ! ! .194 130 99.5974
f i t n e s s va lue : 99 .5974196 Delay Zero ing Started . . . Zeroed I n d i c e s are :
73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69 67 80 7162 77
198 131 77.5263132 78.5236
200 133 78.6551143 78.9574
202 144 78.8042Delay ze ro ing f a i l e d ! ! ! .
204 145 99.7167f i t n e s s va lue : 99 .7167
206 Delay Zero ing Started . . . Zeroed I n d i c e s are :
208 73 57 63 66 75 74 61 78 59 7679 64 60 81 68 55 69 67 80 7162 56
146 82.8367210 147 84.1324
148 84.9284212 158 84.8303
159 85.0946214 160 85.1853
Delay ze ro ing f a i l e d ! ! ! .216 F i d e l i t y : 99 .7167
Zeroed i n d i c e s = 73 57 63 66 75 74 61 78 59 76 79 6460 81 68 55 69 67 80 71 62
218 161 99.7602f i t n e s s va lue : 99 .7602
220 September 18 , 2013 1 0 : 1 3 : 2 0 . 7 1 7 PMDELAY ZEROING FINISHED . . . . . . . . . .
222 Fina l F i d e l i t y After Delay Zero ing i s : 98 .4442
Appendix B. Matlab UOD package 134
Fina l GAX i s : 0 .052313 1 .9609 0.20021 5 .5731 .4956 1 .7704 6 .2615 3 .0848 1 .1086
0.99784 5 .4168 4 .4772 1 .6643 0.712645 .4304 6 .1377 0.49897 0.80231 0.81889
0.58919 5 .9486 0.67844 0.72385 6 .26896 .2621 0.40098 1 .2504 1 .4099 6 .28124 .1682 0.71421 0.97346 6 .0656 1 .82296 .1095 5 .7344 0.94346 0.92946 0.871985 .4689 0.71925 1 .7229 1 .4546 1 .1623
0.087632 0.13403 5 .4307 5 .1024 0.713191 .1218 0.17422 2 .7155 1 .4414 0.45867
0 2 .8471 0 1 .9202 00 0 0 0 0
2 .709 0 0 0 01 .8733 0 1 .483 0 0
0 0 2 .8545 0 00 0
224
226
228 %%%%%%%%%%%%%%%%%%%%%Pulse Zero ing Started . . . .
230 Zeroed Pulse I n d i c e s :161 99.7847
232 Zeroed Pulse I n d i c e s : 1162 99.8482
234 Zeroed Pulse I n d i c e s : 1 25163 99.8845
236 Zeroed Pulse I n d i c e s : 1 25 7164 99.8804
238 Zeroed Pulse I n d i c e s : 1 25 7 45165 99.6779
240 Zeroed Pulse I n d i c e s : 1 25 7 45 51166 99.7928
242 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3167 98.7937
244 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17168 99.2784
246 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31169 99.5468
248 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29170 98.4426
250 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 41171 98.6473
252 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 4137
172 97.0637254 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 41
37 23173 89.9886
Appendix B. Matlab UOD package 135
256 174 90 .529178 91.1387
258 179 91 .192Pulse z e ro ing f a i l e d ! ! !
260 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 4137 23
180 94.9882262 181 97.2162
Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 4137 23 19
264 182 89.4439183 90.7552
266 184 91.0245185 91.1873
268 186 91.2343187 91.2565
270 188 91.2705Pulse z e ro ing f a i l e d ! ! !
272 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 4137 23 19
189 90.5005274 190 96.4813
191 97.6454276 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 41
37 23 19 49192 89.9519
278 193 90.8465196 91.1577
280 197 91.1757198 91.1851
282 Pulse z e ro ing f a i l e d ! ! !Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 41
37 23 19 49284 199 83.3913
200 84 .288286 201 84 .797
204 85.2254288 205 85.2591
206 85 .284290 Pulse z e ro ing f a i l e d ! ! !
Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 4137 23 19 49
292 207 78.3271208 78.8203
294 209 78.9421211 79.0401
296 212 79.0447213 79.0495
298 Pulse z e ro ing f a i l e d ! ! !Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 41
37 23 19 49
Appendix B. Matlab UOD package 136
300 214 93.7353215 97.1219
302 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 4137 23 19 49 27
216 82.3724304 217 84 .107
221 85.0711306 222 85.1314
223 85.1613308 Pulse z e ro ing f a i l e d ! ! !
Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 4137 23 19 49 27
310 224 73.9409225 74 .653
312 229 74.9167230 74.9581
314 Pulse z e ro ing f a i l e d ! ! !Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 41
37 23 19 49 27316 231 74.4074
232 74.8193318 233 74.9427
236 74.9886320 237 74.9911
Pulse z e ro ing f a i l e d ! ! !322 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 41
37 23 19 49 27238 68.2949
324 239 69.3932244 69.9106
326 245 69.9213Pulse z e ro ing f a i l e d ! ! !
328 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 4137 23 19 49 27
246 84.0321330 247 86.0858
248 88.8257332 253 90.9588
254 90.9858334 255 91.0307
Pulse z e ro ing f a i l e d ! ! !336 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 41
37 23 19 49 27256 94.0572
338 257 96.8046258 97.8087
340 Zeroed Pulse I n d i c e s : 1 25 7 45 51 3 17 31 29 4137 23 19 49 27 15
259 88.8588342 260 89.5749
Appendix B. Matlab UOD package 137
266 90.9354344 267 90.9555
Pulse z e ro ing f a i l e d ! ! !346 **** REFINING THE SOLUTION ****
267 98.1254348 268 98.2823
269 98.3255350 270 98.4202
271 98.5255352 Zeroed p u l s e s are : 1 2 3 4 7 8 15 16 17 18 19
20 23 24 25 26 27 28 29 30 31 32 37 38 41 4245 46 49 50 51 52
Fina l F i d e l i t y i s : 98 .5255354 Fina l GAX i s : 0 0 0 0
1 .4911 2 .6772 0 0 0.979671 .6372 4 .6514 4 .4006 3 .1373 1 .2645
0 0 0 0 00 5 .332 0.61489 0 00 0 0 0 00 0 0 1 .0635 2 .285
6 .2102 1 .3274 0 0 1 .04280.12714 0 0 3 .1448 1 .6622
0 0 4 .6146 5 .3583 00 0 0 1 .4973 0 .85770 2 .976 0 2 .065 00 0 0 0 0
3 .1412 0 0 0 01 .931 0 1 .5619 0 0
0 0 2 .9985 0 00 0
356 Elapsed time i s 708.051272 seconds .
Appendix B. Matlab UOD package 138
Figure B.1: Controlled SWAP pulse sequence.
The output shown here can read using the representation scheme used for encoding
the problem solutions in GA optimization. The pulse sequence matrix constructed
from the array ‘gax’ (as given in matlab code ‘uodFitness.m’) is shown in Eqn.
B.2,
0 0 0 0 85.43 153.39 0 170.51 0
0 0 56.13 93.80 266.50 252.14 118.32 0 0
179.75 72.45 0 0 0 0 0 0 0
0 0 305.50 35.23 0 0 0 179.98 0
0 0 0 0 0 0 0 0 0
0 0 60.93 130.92 355.82 76.05 110.64 0 89.49
0 0 59.75 7.28 0 0 0 0 0
180.18 95.24 0 0 264.40 307.01 0 171.80 0
0 0 0 0 85.79 49.14 0 0 0
(B.2)
Here each row represents a pulse-delay segment with column values specifies it
parameters values. First six columns represents pulse parameters as shown in
the matlab code ‘uodFitness.m’) and last three columns represents the coupling
evolution of 1-2 coupling, 2-3 coupling and 1-3 coupling. The equivalent pulse
sequence can be shown as,
85.43 (153.39)3 - U23(170.51) - 56.13(93.80)2 : 266.50(252.14)3 - U12(118.32)
- 179.75(72.45)1 : 305.50 (35.23)2 - U23(179.98) - 60.93(130.92)2 : 355.82(76.05)3
- U12(110.64) - U13(89.49) - 59.75(7.28)2 : 180.18(95.24)1 : 264.40(307.01)3
- U23(171.80) - 85.79(49.14)3.
Graphical representation of the this pulse sequence is shown in Fig. B.1.
Appendix C
Non-Destructive Discrimination
of Arbitrary Set of Orthogonal
Quantum States by NMR using
Quantum Phase Estimation∗
C.1 Introduction
There are several theoretical protocols available for orthogonal state discrimina-
tion [134, 135]. Walgate et al. showed that, using local operations and classical
communication (LOCC) multipartite orthogonal states can be distinguished per-
fectly [134]. However if only a single copy is provided and only LOCC is allowed,
it cannot discriminate quantum states either deterministically or probabilistically
[136]. Estimation of the phase plays an important role in quantum information
processing and is a key subroutine of many quantum algorithms. When the phase
estimation is combined with other quantum algorithms, it can be employed to
perform certain computational tasks such as quantum counting, order finding and
factorization [137, 138]. Phase Estimation Algorithm has also been utilised in re-
cent important application in which the ground state of the Hydrogen molecule has
been obtained upto 45 bit accuracy in NMR and upto 20 bit accuracy in photonic
systems [63].
∗Published in AIP conference proceedings, 1384 229(2011) (Article number 1 in page vii).First work performed without GA optimization.
139
Appendix C. Non-Destructive Discrimination using Phase Estimation 140
By defining an operator with preferred eigen-values, phase estimation can be used
logically for discrimination of quantum states with certainty [3]. It preserves the
state since local operations on ancilla qubit measurements do not affect the state.
Here we describe an algorithm for non-destructive state discrimination using only
phase estimation alone. The algorithm described here is scalable and can be
used for discriminating any set of orthogonal states (entangled or non-entangled).
Earlier non-destructive Bell state discrimination has been described by Gupta et
al. [139] and verified experimentally in our laboratory by Jharana et al. [52]. Bell
states are specific example of orthogonal entangled states. The circuit used for
Bell state discrimination [139] is based on parity and phase estimation and will
not be able to discriminate a superposition state which has no definite parity. For
example consider a state |ψ〉 = 1√2(|00〉+|01〉), which belongs to a set of orthogonal
states. Here |00〉 has parity 0 and |01〉 has parity 1. Hence the above |ψ〉 does
not have a definite parity and cannot be distinguished from its other members of
the set, by the method of Gupta et al. [139]. Sec. C.2 describes the design of a
circuit for non-destructive state discrimination using phase estimation. Sec. C.2
also contains non-destructive discrimination of special cases such as Bell states and
three qubit GHZ states using phase estimation. Sec. C.3 describes experimental
implementation of the algorithm for two qubit states by NMR quantum computer
and Sec. C.4 describes the Matlab® simulation of non-destructive discrimination
of three qubit GHZ states.
C.2 Theory
For a given eigen-vector |φ〉 of a unitary operator U , phase estimation circuit
with Controlled-U operator can be used for finding the eigen-value of |φ〉 [3].
Conversely the reverse of the algorithm, with defined eigen-values can be used for
discriminating eigen-vectors. By logically defining the operators with preferred
eigen-values, the discrimination, as shown here, can be done with certainty.
C.2.1 The General Procedure (n-qubit case):
For n qubit case the Hilbert space dimension is 2n, having 2n independent orthog-
onal states. Hence we need to design a quantum circuit for state discrimination for
a set of 2n orthogonal quantum states. Consider a set of 2n orthogonal states φi,
Appendix C. Non-Destructive Discrimination using Phase Estimation 141
Figure C.1: The general circuit for non-destructive Quantum State Discrimi-nation. For discriminating n qubit states it uses n number of ancilla qubits withn controlled operations. n ancilla qubits are first prepared in the state |00...0〉.Here H represents Hadamard transform and the meter represents a measure-ment of the qubit state. The original state encoded in n qubits is preserved(notdestroyed).
where i = 1, 2, ....2n. The main aim of the discrimination circuit is to make direct
correlation between the elements of φi and possible product states of ancilla
qubits. As there are 2n states, we need n ancilla qubits for proper discrimination.
The discrimination circuit requires n Controlled Operations. Selecting these n
operators Uj (where j = 1, 2, ...n) is the main task in designing the algorithm.
The set Uj depends on the 2n orthogonal states in such a way that the set of
orthogonal vectors forms the eigen-vector set of the operators, with eigen-values
±1. The sequence of +1 and −1 in eigen-values should be defined in a special way,
as outlined below. Let eij (with i = 1, 2...2n) be the eigen-value array of Uj, and
it should satisfy following conditions.
Condition #1 : Eigen-value arrays eij of all operators Uj should contain equal
number of +1 and -1,
Condition #2 : For the first operator U1, the eigen-value array ei1 can be any
possible sequence of +1 and -1 with Condition #1,
Condition #3 : The restriction on eigen-value arrays starts from Uj=2 onwards.
The eigen-value array (ei2) of operator U2 should not be equal to ei1 or its
complement, while still satisfying the Condition #1.
Condition #4 : By generalizing the Condition #3, the eigen-value array (eik)of operator Uk should not be equal to eim (m = 1, 2, ...k−1) or its complement.
Appendix C. Non-Destructive Discrimination using Phase Estimation 142
Let Mj be the diagonal matrix formed by eigen-value array eij of Uj. The
operator Uj is directly related to Mj by a unitary transformation given by,
Uj = V −1 ×Mj × V, (C.1)
where V is the matrix formed by the column vectors |φi〉, V = [ |φ1〉 |φ2〉|φ3〉 ..... |φn〉].
The circuit diagram for implementation of Phase Estimation Algorithm (PEA) to
discriminate orthogonal states using the Controlled-Uj operations such that the
original state is preserved for further use in any quantum circuit is shown in Fig.
C.1.
As the eigen-values defined are either +1 or −1, the final ancilla qubit states will
be in product state (without superposition), and hence can be measured with
certainty. It can be shown that the selection of specific operator set Uj with the
conditions discussed above makes direct correlation between 2n product states of
ancilla qubit and elements of |φi〉 so that ancilla measurements can discriminate
the state.
C.2.2 Single qubit case:
For a single qubit system, the Hilbert space dimension is 2. So we can discriminate
a state from a set of two orthogonal states. Consider an illustrative example with
the orthonormal set as |φ1〉 = 1√2(|0〉+ |1〉), |φ2〉 = 1√
2(|0〉−|1〉) . The quantum
circuit for this particular case can be designed by following the general procedure
discussed in Sec. C.2. The V matrix for the given states |φ1〉, |φ2〉 is,
V =1√2
(1 1
1 −1
)
Appendix C. Non-Destructive Discrimination using Phase Estimation 143
According to the rules given in Sec. C.2, M can be either
(1 0
0 −1
)or
(−1 0
0 1
).
For M =
(1 0
0 −1
),
U = V −1 ×M × V =
(0 1
1 0
). (C.2)
The circuit diagram for this case is identical to Fig. C.1, having only one work and
one ancilla qubit. It can be easily shown that, the ancilla qubit measurements are
directly correlated with the input states. For the selected M1, if the given state is
|φ1〉 then ancilla will be in the state |0〉 and if the given state is |φ2〉 ancilla will
be in the state |1〉.
For a general set |φ1〉 = (α|0〉+ β|1〉), |φ2〉 = (β|0〉 − α|1〉) (where α and β are
real numbers satisfying, |α|2 + |β|2 = 1), operator U for eigenvalue array 1,−1can be shown as,
U =
(Cos(θ) Sin(θ)
Sin(θ) −Cos(θ)
), (C.3)
where θ = 2× Tan−1(βα
).
C.2.3 Two qubit case:
The Hilbert space dimension of two qubit system of is four. Consider an illustrative
example with a set of orthonormal states
|S(α, β)〉 = (α|00〉+ β|01〉), (α|10〉+ β|11〉), (β|10〉 − α|11〉), (β|00〉 − α|01〉)(C.4)
where α and β are real numbers satisfying, |α|2 + |β|2 = 1. This set is so chosen
that the states are (a)orthogonal, (b)not entangled, (c)different from Bell states,
(d)do not have definite parity and (e)contain single-superposed-qubits (SSQB)
(in this case second qubit is superposed). Using the general procedure discussed
Appendix C. Non-Destructive Discrimination using Phase Estimation 144
states Ancilla-1 Ancilla-2|φ1〉 |0〉 |0〉|φ2〉 |0〉 |1〉|φ3〉 |1〉 |0〉|φ4〉 |1〉 |1〉
Table C.1: State of ancilla qubits for different input states for two qubitorthogonal states.
above, we can select the eigen-value arrays for two operators U1 and U2 as
e1 = 1, 1,−1,−1, e2 = 1,−1, 1,−1. (C.5)
U1 and U2, the unitary transformation of the diagonal matrices formed by e1and e2 are,
U1 =
Cos(θ) Sin(θ) 0 0
Sin(θ) −Cos(θ) 0 0
0 0 Cos(θ) Sin(θ)
0 0 Sin(θ) −Cos(θ)
, (C.6)
U2 =
Cos(θ) Sin(θ) 0 0
Sin(θ) −Cos(θ) 0 0
0 0 −Cos(θ) −Sin(θ)
0 0 −Sin(θ) Cos(θ)
, (C.7)
where, θ = 2× Tan−1(βα
).
The output state of the ancilla qubit run through all possible product states as
input state changes, as listed in Tab. C.1. The quantum circuit for two qubit
state discrimination is shown in Fig. C.2a.
Appendix C. Non-Destructive Discrimination using Phase Estimation 145
C.2.3.1 Special case (α = β = 1√2):
The set of orthogonal states are,
|S(1√2,
1√2
)〉 = |φi〉 = 1√2
(|00〉+ |01〉), 1√2
(|10〉+ |11〉),
1√2
(|10〉 − |11〉), 1√2
(|00〉 − |01〉).(C.8)
The operators U1 and U2 can be found by substituting the value of θ = π2
in (5)
and (6),
U1 =1√2
0 1 0 0
1 0 0 0
0 0 0 1
0 0 1 0
and U2 =1√2
0 1 0 0
1 0 0 0
0 0 0 −1
0 0 −1 0
. (C.9)
The quantum circuit for the set (Eqn. C.8) is same as the general case of any set
of two qubit orthogonal states(Fig. C.2a). Experimental implementation of this
case has been performed using NMR and is described in Sec. C.3
C.2.4 Bell state discrimination:
Bell states are maximally entangled two qubit states (also known as Einstein-
Podolsky-Rosen states) [48]. They play a crucial role in several applications of
quantum computation and quantum information theory. They have been used for
teleportation, dense coding and entanglement swapping [49–51, 140]. Bell states
have also found application in remote state preparation, where a known state is
prepared in a distant laboratory [141]. Hence, it is of general interest to distinguish
Bell states without disturbing them.The complete set of Bell states are,
|Bi〉 =
1√2(|0201〉+ |1211〉), 1√
2(|0201〉 − |1211〉),
1√2(|0211〉+ |1201〉), 1√
2(|0211〉 − |1201〉)
. (C.10)
Bell states form an orthogonal set. Hence one can design a circuit for Bell
state discrimination using only phase estimation. The circuit diagram is same
as that shown in Fig. C.2a with different U1 and U2. For eigen-value arrays
Appendix C. Non-Destructive Discrimination using Phase Estimation 146
e1 = 1,−1, 1,−1, e2 = −1, 1, 1,−1. U1 and U2 are obtained as,
U1 =
0 0 0 1
0 0 1 0
0 1 0 0
1 0 0 0
and U2 =
0 0 0 −1
0 0 1 0
0 1 0 0
−1 0 0 0
. (C.11)
The controlled operators (C−U1 and C−U2) for phase estimation which involves
3-qubit operators can be written as the product of 2-qubit operators as,
C − U1 = C −NOT 31 × C −NOT 3
2 ,
C − U2 = C − π21 × C −NOT 4
1 × C −NOT 42 × C − π2
1.(C.12)
Here qubits 1 and 2 are work qubits in which the Bell states are encoded and 3
and 4 are the ancilla qubits. Here C − NOT ij represents C-NOT operation with
control on ith qubit and target on jth qubit. The splitting of three qubit operator
into two qubit operators is needed for the implemention of C − U1 and C − U2.
There already exists an algorithm for non-destructive discrimination of Bell state
by Gupta et al. [139], which has also been experimentally implemented in NMR by
Jharana et al. [52]. The circuit of Gupta et al. [139] is based on parity and phase
measurement and will fail for a superposed state which has no definite parity.
However for non-destructive discrimination of Bell states using the present phase
estimation algorithm is similiar to Gupta’s circuit where the parity estimation is
replaced by modified phase estimation.
C.2.5 GHZ state discrimination:
GHZ states are maximally entangled multi qubit states [142]. GHZ states have
been used in several quantum algorithms such as quantum secret sharing, con-
trolled dense coding and quantum key distribution [143, 144]. These algorithms
make use of entanglement and hence it is important to discriminate GHZ states
by preserving their entanglement.
Appendix C. Non-Destructive Discrimination using Phase Estimation 147
All n qubit GHZ states form an orthogonal set (without definite parity). Hence a
circuit can be designed for discriminating general n qubit GHZ states using only
phase estimation.
Consider the case of three qubit GHZ states, which are
|Gi〉 =
1√2(|000〉+ |111〉), 1√
2(|000〉 − |111〉),
1√2(|001〉+ |110〉), 1√
2(|001〉 − |110〉),
1√2(|010〉+ |101〉), 1√
2(|010〉 − |101〉),
1√2(|011〉+ |100〉), 1√
2(|011〉 − |100〉).
. (C.13)
Here we need three ancilla qubits and have to implement three controlled operators
for state discrimination. Verification of the NMR experiment to discriminate such
states has been carried out here using Matlab® and the parameters of a four qubit
NMR system, as described in Sec. C.4.
C.3 Experimental Implementation by NMR
C.3.1 Non-Destructive Discrimination of two qubit orthog-
onal states:
Experimental implementation of the quantum state discrimination(QSD) algo-
rithm has been performed here for 2 qubit case for the orthogonal set (Eqn. C.8).
The discrimination circuit diagram shown in Fig. C.2a, needs a 4 qubit system.
As two ancilla qubits are independent of each other, following [52] one can split the
experiment into 2 measurements with a single ancilla qubit (Fig.C.2b and C.2c).
The NMR implementation of the discrimination algorithm starts with (i) prepara-
tion of the pseudo-pure state followed by (ii) creation of input state (iii) Quantum
Phase estimation with operators Uj. Finally the measurement on ancilla qubits
yields the result.
The experiment has been carried out at 300K in 11.7T field in a Bruker AV 500
spectrometer using a triple resonance QXI probe. The system chosen for the
implementation of the discrimination algorithm is Carbon-13 labeled Dibromo-
fluoro methane 13CHFBr2, where 1H, 19F and 13C act as the three qubits[78].
Appendix C. Non-Destructive Discrimination using Phase Estimation 148
Figure C.2: (a) Two qubit State discrimination circuit for Experimental im-plementation in three qubit NMR quantum computer, (b) and (c) are splittingof the circuit-(a) into two circuits with single ancilla measurements.
The 1H, 19F and 13C resonance frequencies at this field are 500, 470 and 125
MHz, respectively. The scalar couplings between the spins are: JHC = 224.5Hz,
JHF = 49.7Hz and JFC = −310.9Hz(Fig. C.3).
The NMR Hamiltonian for a three qubit weakly coupled spin system is [14],
H =3∑i=0
νiIiz +
3∑i<j=1
JijIizIjz , (C.14)
where νi are the Larmor frequencies and the Jij are the scalar couplings. The
starting point of any algorithm in an NMR quantum information processor is the
equilibrium density matrix, which under high temperature and high field approx-
imation is in a highly mixed state represented by[78],
ρeq ∝ γHIHz + γCI
Cz + γF I
Fz = γH(IHz + 0.94IFz + 0.25ICz ). (C.15)
There are several methods for creating pseudo pure states (PPS) in NMR from
equlibrium state [28–30, 145]. We have utilized the spatial averaging technique
[145] for creating pseudo pure states as described in [78]. The spectra for equlib-
rium and |000〉 PPS are shown in Fig. C.3.
For Phase Estimation algorithm, due to its high sensitivity, proton spin has been
utilized as the ancilla qubit; and the two qubit states, to be discriminated, are
Appendix C. Non-Destructive Discrimination using Phase Estimation 149
FLUORINECARBONPROTON
a.
b.
Figure C.3: The three qubit NMR sample used for experimental implemen-tation. The nuclear spins 1H, 19F and 13C are used as the three qubits. (a)Equlibrium Spectra of proton, carbon and fluorine, (b) Spectra corresponds tothe created |000〉 pseudo pure state. These spectra are obtain by using 90o
measuring pulse on each spin.
encoded in carbon and fluorine spins. As the measurements are performed only
on ancilla qubit, we record only proton spectra for non-destructive discrimination
of the state of carbon and fluorine. The state of the ancilla qubit can be identified
by the relative phase of the spectra. We set the phase such that a positive peak
indicates that the proton was initially in state |0〉.
C.3.1.1 Implementation of Controlled-U1 and U2:
For the set of orthogonal states given in Eqn. C.8 the U1 and U2 are given in
Eqn. C.9. Let H1 and H2 be the effective Hamiltonians for Controlled-U1 and
Controlled-U2 propagators such that,
Controlled-U1 = exp(iH1),
Controlled-U2 = exp(iH2).(C.16)
where H1 and H2, in terms of product operators [14] are obtained as,
H1 = (π
4I − π
2I1z −
π
2I3x + πI1
z I3x),
H2 = (π
4I − π
2I1z − πI2
z I3x + 2πI1
z I2z I
3x)
(C.17)
Since the various terms in H1 and H2 commute with each other, one can write,
Appendix C. Non-Destructive Discrimination using Phase Estimation 150
Figure C.4: The pulse sequence for Controlled-U1 and Controlled-U2 oper-ators for two qubit orthogonal states shown in Eqn. C.9 (Here narrow pulsesindicate (π2 ) pulses and broad pulses indicate π pulses with the phase givenabove the pulse).
200 0 0 0 0 200 200 200 200 200 200 200 Hz
Figure C.5: Ancilla (proton spin) spectra of final state for two qubit statediscrimination algorithm for (i) |φ1〉, (ii) |φ2〉, (iii) |φ3〉, (iv) |φ4〉 states. A1
and A2 are results of two measurements on single ancilla (Fig. C.2b and C.2crespectively) qubit-1 and 2 (here it is two experiments with same ancilla qubit).These spectra are obtained with a 90o measuring pulse on the ancilla(proton)qubit at the end of the pulse sequence.
Controlled− U1 = exp(i(π
4I − π
2I1z −
π
2I3x + πI1
z I3x)) =
exp(iπ
4I)× exp(−iπ
2I1z )× exp(−iπ
2I3x)× exp(iπI1
z I3x),
Controlled− U2 = exp(i(π
4I − π
2I1z − πI2
z I3x + 2πI1
z I2z I
3x)) =
exp(iπ
4I)× exp(−iπ
2I1z )× exp(−iπI2
z I3x)× exp(i2πI1
z I2z I
3x).
(C.18)
As the decomposed terms commute with each other, these propagators can be
easily implemented in NMR(Fig. C.4). Single spin operators such as Ix, Iy are
implemented using R.F pulses. The Iz operator is implemented using composite z
rotation pulses in NMR ((π2)−x(
π2)y(
π2)x) [95, 146]. Two spin product terms such
as I izIjx are implemented using scalar coupling Hamiltonian evolution sandwiched
between two (π2)y pulse on j spin [78]. The three spin product operator terms are
implemented using cascades of two spin operator evolutions(Tseng et al. [147]).
Appendix C. Non-Destructive Discrimination using Phase Estimation 151
The experimental results are shown in Fig. C.5. Proton spectra shows the state
of ancilla qubit, which in-turn can be used for discrimination of two qubit state
in carbon and fluorine spins. Positive peaks in Fig. C.5 means ancilla is in qubit
state |0〉 and negative peak indicates ancilla qubit is in state |1〉. Thus spectra in
Fig. C.5 indicates that (i), (ii), (iii), (iv) are respectively |φ1〉, |φ2〉, |φ3〉, |φ4〉 (Tab.
C.1). To compute fidelity of the experiment, complete density matrix tomography
has been carried out (Fig. C.6). The experimental results are in agreement with
the Tab. C.1 with an ‘average absolute deviation’ [32] of 4.0% and ‘maximum
absolute deviation’ [32] of 7.2%, providing the desired discrimination.
C.4 Three qubit GHZ state Discrimination using
Matlab® Simulation:
Non-destructive discrimination of the three qubit maximally entangled (GHZ)
states using only Phase Estimation algorithm as described in Sec. C.2 in NMR
has also been performed using a Matlab® simulation. This simulation verifies
the principle involved but does not include any decoherence or pulse imperfection
effects. The three qubit GHZ states form a set Gi given by Eqn. C.13 can be
re-expressed as,
|Gi〉 = |φ1〉, |φ1〉, |φ3〉, ........|φ8〉 (C.19)
The discrimination of a 3-qubit GHZ state using phase estimation requires 3
work qubits and 3 ancilla. We divide the 6 qubit quantum circuit into three
circuits. Each circuit has three work qubits and a single ancilla. There are several
possibilities for eigen-value sets which will satisfy the sets of conditions discussed
in Sec. C.2. Consider one such set,
e1e2e3
=
e1 = 1,−1, 1,−1, 1,−1, 1,−1,e2 = −1, 1, 1,−1, 1,−1,−1, 1,e3 = −1, 1, 1,−1,−1, 1, 1,−1.
. (C.20)
For this eigen-value set (18), the Controlled− Uj operators can be written as
Appendix C. Non-Destructive Discrimination using Phase Estimation 152
Figure C.6: Density Matrix Tomography of the initial and final states ofquantum state discrimination circuit. First qubit is the ancilla. It is evidentthat the state of 2nd and 3rd qubits are preserved. (here 1→ |000〉, 2→ |001〉,3→ |010〉, 4→ |011〉, 5→ |100〉, 6→ |101〉, 7→ |110〉, 8→ |111〉.)
Appendix C. Non-Destructive Discrimination using Phase Estimation 153
Figure C.7: Pulse sequence for Controlled operators in GHZ state discrim-ination. (Here narrow pulses indicate (π2 ) pulses and broad pulses indicate πpulses with the phase given above the pulse).
Appendix C. Non-Destructive Discrimination using Phase Estimation 154
Figure C.8: The chemical structure, the chemical shifts and spin-spin couplingof a 13C labelled Crotonic Acid. The four 13C spins act as four qubits [2].
state Measurement-1 Measurement-2 Measurement-3|φ1〉 |0〉 |1〉 |1〉|φ2〉 |1〉 |0〉 |0〉|φ3〉 |0〉 |0〉 |0〉|φ4〉 |1〉 |1〉 |1〉|φ5〉 |0〉 |0〉 |1〉|φ6〉 |1〉 |1〉 |0〉|φ7〉 |0〉 |1〉 |0〉|φ8〉 |1〉 |0〉 |1〉
Table C.2: State of ancilla qubits for different input states of Eqn. C.13 andC.19.
Controlled− U1 = C −NOT a1 × C −NOT a2 × C −NOT a3 ,
Controlled− U2 = C − π23 × C −NOT a1 × C −NOT a2×
C −NOT a3 × C − π23,
Controlled− U3 = C − π13 × C −NOT a1 × C −NOT a2×
C −NOT a3 × C − π13.
(C.21)
Here 1,2 and 3 are the work qubits, in which the GHZ state is encoded and ‘a’ is
the ancilla qubit. Pulse sequence for these operators are shown in Fig. C.7. The
results of ancilla qubit measurements are tabulated in Tab. C.2.
NMR simulation has been carried out using the parameters of a well known 4-qubit
system, crotonic acid with all carbons labelled by 13C (Fig. C.8) [2]. The density
matrix tomography of the Matlab® experiment for a few selected(|φ1〉, |φ4〉 and
|φ7〉) GHZ states are shown in Fig. C.9. This confirms that the method of Phase
Estimation discussed in Sec. C.2 can be used for discrimination of GHZ states
without destroying them.
Appendix C. Non-Destructive Discrimination using Phase Estimation 155
First Experiment Second Experiment Third Experiment
(i) |0〉a( 1√2(|000〉+ |111〉)) |1〉a( 1√
2(|000〉+ |111〉)) |1〉a( 1√
2(|000〉+ |111〉))
2 4 6 8 10 12 14 16
24
68
1012
1416
−0.5
−0.25
0
0.25
0.5
2 4 6 8 10 12 14 16
24
68
1012
1416
−0.5
−0.25
0
0.25
0.5
2 4 6 8 10 12 14 16
24
68
1012
1416
−0.5
−0.25
0
0.25
0.5
(ii) |1〉a( 1√2(|001〉 − |110〉)) |1〉a( 1√
2(|001〉 − |110〉)) |1〉a( 1√
2(|001〉 − |110〉))
2 4 6 8 10 12 14 16
24
68
1012
1416
−0.5
−0.25
0
0.25
0.5
2 4 6 8 10 12 14 16
24
68
1012
1416
−0.5
−0.25
0
0.25
0.5
2 4 6 8 10 12 14 16
24
68
1012
1416
−0.5
−0.25
0
0.25
0.5
Figure C.9: Matlab® simulation results for GHZ state discrimination. Thesimulated spectras are shown for three GHZ states |φ1〉 and |φ4〉. It is evidentfrom final density matrix that the GHZ states are preserved. (Here 1→ |0000〉,2 → |0001〉, 3 → |0010〉, 4 → |0011〉, 5 → |0100〉, 6 → |0101〉, 7 → |0110〉,8 → |0111〉, 9 → |1000〉, 10 → |1001〉, 11 → |1010〉, 12 → |1011〉, 13 → |1100〉,14→ |1101〉, 15→ |1110〉, 16→ |1111〉. First qubit is the ancilla)
C.5 Conclusion
A general scalable method for non-destructive quantum state discrimination of
a set of orthogonal states using quantum phase estimation algorithm has been
descibed, and experimently implemented for a two qubit case by NMR. As the
direct measurements are performed only on the ancilla, the discriminated states
are preserved. The generalization of the algorithm is illustrated by discrimination
of GHZ states using a Matlab® simulation.
Bibliography
[1] http://www.mathworks.in/help/matlab/ref/expm.html.
[2] C. Negrevergne, R. Somma, G. Ortiz, E. Knill, and R. Laflamme. Liquid-
state NMR Simulations of Quantum Many-Body Problems. Physical Review
A, 71:032344, 2005.
[3] M. A. Nielson and I. L. Chuang. Quantum Computation and Quantum
Information. Cambridge University Press, 2002.
[4] http://www.nobelprize.org/nobel prizes/physics/laureates/.
[5] Felix Bloch. Nuclear Induction. Physical review, 70(7):460, 1946.
[6] Edward M. Purcell, H. C. Torrey, and Robert V. Pound. Resonance Ab-
sorption by Nuclear Magnetic Moments in a Solid. Physical review, 69:37,
1946.
[7] James Keeler. Understanding NMR Spectroscopy. Wiley com, 2011.
[8] Anil Kumar, R. R. Ernst, and K. Wuthrich. A Two-Dimensional Nu-
clear Overhauser Enhancement (2D NOE) Experiment for the Elucidation
of Complete Proton-Proton Cross-Relaxation Networks in Biological Macro-
molecules. Biochemical and Biophysical Research Communications, 95:1,
1980.
[9] Kurt Wuthrich. NMR Studies of Structure and Function of Biological Macro-
molecules (Nobel Lecture). Angewandte Chemie International Edition, 42
(29):3340, 2003.
[10] John Cavanagh, Wayne. J. Fairbrother, Arthur G. Palmer III, and
Nicholas J. Skelton. Protein NMR spectroscopy: Principles and Practice.
Academic Press, 1995.
157
Bibliography 158
[11] Christopher M. Dobson and Peter J. Hore. Kinetic Studies of Protein Folding
Using NMR Spectroscopy. Nature Structural & Molecular Biology, 5:504,
1998.
[12] Lewis E. Kay, G. Marius Clore, A. Bax, and Angela M. Gronenborn.
Four-Dimensional Heteronuclear Triple-resonance NMR Spectroscopy of
Interleukin-1 beta in Solution. Science, 249(4967):411, 1990.
[13] Paul C. Lauterbur. Image Formation by Induced Local Interactions: Exam-
ples Employing Nuclear Magnetic Resonance. Nature, 242(5394):190, 1973.
[14] R. R. Ernst, G Bodenhausen, and A. Wokaun. Principles of Nuclear Mag-
netic Resonance in One and Two Dimensions. Oxford University Press,
1990.
[15] Scott A. Huettel, Allen W. Song, and Gregory McCarthy. Functional Mag-
netic Resonance Imaging, volume 1. Sinauer Associates Sunderland, 2004.
[16] http://cbaweb2.med.unc.edu/henson mrm/pages/mrmfaq.html#MRMAnchor.
[17] V. V. Krishnan and B. Rupp. Macromolecular Structure Determination:
Comparison of X-ray Crystallography and NMR Spectroscopy. John Wiley
& Sons, Ltd, 2001.
[18] Neil A. Farrow, Ranjith Muhandiram, Alex U. Singer, Steven M. Pascal,
Cyril M. Kay, Gerry Gish, Steven E. Shoelson, Tony Pawson, Julie D.
Forman-Kay, and E. Kay, Lewis. Backbone Dynamics of a Free and a
Phosphopeptide-Complexed Src Homology 2 Domain Studied by 15N NMR
Relaxation. Biochemistry, 33(19):5984, 1994.
[19] Anthony Mittermaier and E. Kay, Lewis. New Tools Provide New Insights
in NMR Studies of Protein Dynamics. Science, 312(5771):224, 2006.
[20] E. Kay, Lewis. Protein Dynamics from NMR. Biochemistry and cell biology,
76:145, 1998.
[21] Alex D. Bain. Chemical Exchange in NMR. Progress in Nuclear Magnetic
Resonance Spectroscopy, 43:63, 2003.
[22] Lloyd Jackman. Dynamic Nuclear Magnetic Resonance Spectroscopy. Else-
vier, 1975.
Bibliography 159
[23] Jiangfeng Du, Nanyang Xu, Xinhua Peng, Pengfei Wang, Sanfeng Wu, and
Dawei Lu. NMR Implementation of a Molecular Hydrogen Quantum Simula-
tion with Adiabatic State Preparation. Physical Review Letters, 104:030502,
2010.
[24] A. Sørensen, L. M Duan, J. I. Cirac, and Peter Zoller. Many-particle En-
tanglement with Bose-Einstein Condensates. Nature, 409(6816):63, 2001.
[25] Jharana Rani Samal, Arun K. Pati, and Anil Kumar. Experimental Test
of the Quantum No-Hiding Theorem. Physical Review Letters, 106:080401,
2011.
[26] Vikram Athalye, Soumya Singha Roy, and T. S. Mahesh. Investigation of
the Leggett-Garg Inequality for Precessing Nuclear Spins. Physical Review
Letters, 107:130402, 2011.
[27] Xinhua Peng, Jingfu Zhang, Jiangfeng Du, and Dieter Suter. Quantum
Simulation of a System with Competing Two- and Three-Body Interactions.
Physical Review Letters, 103:140501, 2009.
[28] David G. Cory, Amr F. Fahmy, and Timothy F. Havel. Ensemble Quantum
Computing by NMR Spectroscopy. Proceedings of the National Academy of
Sciences, 94:1634, 1997.
[29] Neil A. Gershenfeld and Isaac L. Chuang. Bulk Spin-Resonance Quantum
Computation. Science, 275(5298):350, 1997.
[30] T. S. Mahesh and Anil Kumar. Ensemble Quantum-information Process-
ing by NMR: Spatially Averaged Logical Labeling Technique for Creating
Pseudopure States. Physical Review A, 64:012307, 2001.
[31] I. L. Chuang, N. Gershenfeld, M. G. Kubinec, and D. W. Leung. Bulk Quan-
tum Computation with Nuclear Magnetic Resonance: Theory and Experi-
ment. Proceedings of the Royal Society of London. Series A: Mathematical,
Physical and Engineering Sciences, 454(1969):447, 1998.
[32] Avik Mitra, Arindam Ghosh, Ranabir Das, Apoorva Patel, and Anil Kumar.
Experimental Implementation of Local Adiabatic Evolution Algorithms by
an NMR Quantum Information Processor. Journal of Magnetic Resonance,
177:285, 2005.
Bibliography 160
[33] Darrell Whitley. A Genetic Algorithm Tutorial. Statistics and Computing,
4:65, 1994.
[34] J. H. Holland. Adaptation in Natural and Artificial Systems: An Introductory
Analysis with Applications to Biology, Control, and Artificial Intelligence.
The MIT Press, 1992.
[35] Khaled Rasheed and Andrew Gelsey. Fourth International Conference on
Artificial Intelligence in Design, 1996.
[36] D. M. Deaven and K. M. Ho. Molecular Geometry Optimization with a
Genetic Algorithm. Physical Review Letters, 75:288, 1995.
[37] M. Ji, K. Umemoto, C. Z. Wang, K. M. Ho, and R. M. Wentzcovitch.
Ultrahigh-Pressure Phases of H2O Ice Predicted Using an Adaptive Genetic
Algorithm. Physical Review B, 84(22):220105, 2011.
[38] I. Rata, A. A. Shvartsburg, M. Horoi, T. Frauenheim, K. W. M. Siu, and
K. A. Jackson. Single-Parent Evolution Algorithm and the Optimization of
Si Clusters. Physical Review Letters, 85:546, 2000.
[39] P. Pahari and S. Chaturvedi. Determination of Best-fit Potential Parameters
for a Reactive Force Field Using a Genetic Algorithm. Journal of Molecular
Modeling, page 1, 2012.
[40] Ray Freeman and Wu Xili. Design of Magnetic Resonance Experiments by
Genetic Evolution. Journal of Magnetic Resonance, 75:184, 1987.
[41] Yong Pang and Gary X. Shen. Improving Excitation and Inversion Accuracy
by Optimized RF Pulse Using Genetic Algorithm. Journal of Magnetic
Resonance, 186:86, 2007.
[42] Matthias Bechmann, John Clark, and Angelika Sebald. Genetic Algorithms
and Solid State NMR Pulse Sequences. Journal of Magnetic Resonance, 228:
66, 2013.
[43] E. M. Fortunato, M. A. Pravia, N. Boulant, G. Teklemariam, T. F. Havel,
and D. G. Cory. Design of Strongly Modulating Pulses to Implement Precise
Effective Hamiltonians for Quantum Information Processing. The Journal
of chemical physics, 116:7599, 2002.
Bibliography 161
[44] N. Khaneja, T. Reiss, C. Kehlet, T. Schulte-Herbruggen, and S. J. Glaser.
Optimal Control of Coupled Spin Dynamics: Design of NMR Pulse Se-
quences by Gradient Ascent Algorithms. Journal of Magnetic Resonance,
172:296, 2005.
[45] A. Ajoy, R. K. Rao, A. Kumar, and P. Rungta. Algorithmic Approach to
Simulate Hamiltonian Dynamics and An NMR Simulation of Quantum State
Transfer. Physical Review A, 85(3):030303, 2012.
[46] W. S. Price. Pulsed-field Gradient Nuclear Magnetic Resonance as a Tool for
Studying Translational Diffusion: Part II. Experimental Aspects. Concepts
in Magnetic Resonance, 10:197, 1998.
[47] T. S. Mahesh, K. Dorai, Arvind, and Anil Kumar. Implementing Logic Gates
and the Deutsch-Jozsa Quantum Algorithm by Two-Dimensional NMR Us-
ing Spin-and Transition-Selective Pulses. Journal of Magnetic Resonance,
148:95, 2001.
[48] A. Einstein, B. Podolsky, and N. Rosen. Can Quantum-Mechanical Descrip-
tion of Physical Reality be Considered Complete ? Physical review, 47:777,
1935.
[49] C. H. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres, and W. K.
Wootters. Teleporting an Unknown Quantum State via Dual Classical and
Einstein-Podolsky-Rosen Channels. Physical Review Letters, 70:1895, 1993.
[50] C. H. Bennett and S. J. Wiesner. Communication via One-and Two-Particle
Operators on Einstein-Podolsky-Rosen States. Physical Review Letters, 69
(20):2881, 1992.
[51] Jian-Wei Pan, Dik Bouwmeester, Harald Weinfurter, and Anton Zeilinger.
Experimental Entanglement Swapping: Entangling Photons That Never In-
teracted. Physical Review Letters, 80:3891, 1998.
[52] Jharana Rani Samal, M. Gupta, P. K. Panigrahi, and A. Kumar. Non-
destructive Discrimination of Bell States by NMR Using a Single Ancilla
Qubit. Journal of Physics B: Atomic, Molecular and Optical Physics, 43:
095508, 2010.
Bibliography 162
[53] M. Carravetta and M. H. Levitt. Long-Lived Nuclear Spin States in High-
Field Solution NMR. Journal of the American Chemical Society, 126(20):
6228, 2004.
[54] G. Pileio, M. Carravetta, and M. H. Levitt. Storage of Nuclear Magnetization
as Long-Lived Singlet Order in Low Magnetic Field. Proceedings of the
National Academy of Sciences, 107(40):17135, 2010.
[55] Soumya Singha Roy and T. S. Mahesh. Initialization of NMR Quantum
Registers Using Long-Lived Singlet States. Physical Review A, 82:052302,
2010.
[56] A. J Shaka, James Keeler, Tom Frenkiel, and Ray Freeman. An improved
Sequence for Broadband Decoupling: WALTZ-16. Journal of Magnetic Res-
onance, 52:335, 1983.
[57] Soumya Singha Roy and T. S. Mahesh. Density Matrix Tomography of
Singlet States. Journal of Magnetic Resonance, 206:127, 2010.
[58] P. Shor. Polynomial-Time Algorithms for Prime Factorization and Discrete
Logarithms on a Quantum Computer. SIAM Journal on Computing, 26:
1484, 1997.
[59] Lov K. Grover. A Fast Quantum Mechanical Algorithm for Database Search.
In Proceedings of the twenty-eighth annual ACM symposium on Theory of
computing, STOC ’96, page 212. ACM, 1996.
[60] Seth Lloyd. Universal Quantum Simulators. Science, 273(5278):1073, 1996.
[61] Daniel A. Lidar and Haobin Wang. Calculating the Thermal Rate Constant
with Exponential Speedup on a Quantum Computer. Physical Review E, 59:
2429, 1999.
[62] Richard Feynman. Simulating Physics with Computers. International Jour-
nal of Theoretical Physics, 21:467, 1982.
[63] B. P. Lanyon, J. D. Whitfield, G. G. Gillett, M. E. Goggin, M. P. Almeida,
I. Kassal, J. D. Biamonte, M. Mohseni, B. J. Powell, M. Barbieri, A. Aspuru-
Guzik, and A. G. White. Towards Quantum Chemistry on a Quantum
Computer. Nature Chemistry, 2:106, 2010.
Bibliography 163
[64] A. Aspuru-Guzik, Anthony D. Dutoi, Peter J. Love, and Martin Head-
Gordon. Simulated Quantum Computation of Molecular Energies. Science,
309(5741):1704, 2005.
[65] Ivan Kassal, Stephen P. Jordan, Peter J. Love, Masoud Mohseni, and Aln
Aspuru-Guzik. Polynomial-Time Quantum Algorithm for the Simulation of
Chemical Dynamics. Proceedings of the National Academy of Sciences, 105
(48):18681, 2008.
[66] V. S. Manu and Anil Kumar. Fast and Accurate Quantification Using Ge-
netic Algorithm Optimized 1H-13C Refocused Constant-Time INEPT. Jour-
nal of Magnetic Resonance, 234:106, 2013.
[67] Michael J. Bremner, Jennifer L. Dodd, Michael A. Nielsen, and Dave Bacon.
Fungible Dynamics: There are Only Two Types of Entangling Multiple-qubit
Interactions. Physical Review A, 69:012313, 2004.
[68] Charles D. Hill. Robust Controlled-NOT Gates from Almost Any Interac-
tion. Physical Review Letters, 98:180501, 2007.
[69] M. C. Arnesen, S. Bose, and V. Vedral. Natural Thermal and Magnetic
Entanglement in the 1D Heisenberg Model. Physical Review Letters, 87:
017901, 2001.
[70] L. Zhou, H. S. Song, Y. Q. Guo, and C. Li. Enhanced Thermal Entanglement
in An Anisotropic Heisenberg XYZ Chain. Physical Review A, 68:024301,
2003.
[71] Xiaoguang Wang. Boundary and Impurity Effects on the Entanglement of
Heisenberg Chains. Physical Review E, 69:066118, 2004.
[72] Jingfu Zhang, Gui Lu Long, Wei Zhang, Zhiwei Deng, Wenzhang Liu, and
Zhiheng Lu. Simulation of Heisenberg XY Interactions and Realization of a
Perfect State Transfer in Spin Chains Using Liquid Nuclear Magnetic Reso-
nance. Physical Review A, 72:012331, 2005.
[73] I. Dzyaloshinsky. A Thermodynamic Theory of Weak Ferromagnetism of
Antiferromagnetics. Journal of Physics and Chemistry of Solids, 4:241, 1958.
[74] Toru Moriya. New Mechanism of Anisotropic Superexchange Interaction.
Physical Review Letters, 4:228, 1960.
Bibliography 164
[75] D. C. Dender, P. R. Hammar, Daniel H. Reich, C. Broholm, and G. Aeppli.
Direct Observation of Field-Induced Incommensurate Fluctuations in a One-
Dimensional S = 1/2 Antiferromagnet. Physical Review Letters, 79:1750,
1997.
[76] Masahumi Kohgi, Kazuaki Iwasa, Jean-Michel Mignot, Bjorn Fak, Philipp
Gegenwart, Michael Lang, Akira Ochiai, Hidekazu Aoki, and Takashi Suzuki.
Staggered Field Effect on the One-Dimensional S = 12
Antiferromagnet
Yb4As3. Physical Review Letters, 86:2439, 2001.
[77] M. Greven, R. J. Birgeneau, Y. Endoh, M. A. Kastner, M. Matsuda,
and G. Shirane. Neutron Scattering Study of the Two-dimensional Spin
S=1/2 Square-lattice Heisenberg Antiferromagnet Sr2CuO2Cl2. Zeitschrift
fr Physik B Condensed Matter, 96:465, 1995.
[78] Avik Mitra, K. Sivapriya, and Anil Kumar. Experimental Implementation
of a Three Qubit Quantum Game with Corrupt Source Using Nuclear Mag-
netic Resonance Quantum Information Processor. Journal of Magnetic Res-
onance, 187:306, 2007.
[79] Valerie Coffman, Joydip Kundu, and William K. Wootters. Distributed
Entanglement. Physical Review A, 61:052306, 2000.
[80] C. H. Bennett and Stephen J. Wiesner. Communication via One- and Two-
particle Operators on Einstein-Podolsky-Rosen States. Physical Review Let-
ters, 69:2881, 1992.
[81] Charles H. Bennett, Gilles Brassard, Claude Crepeau, Richard Jozsa, Asher
Peres, and William K. Wootters. Teleporting an Unknown Quantum state
via Dual Classical and Einstein-Podolsky-Rosen Channels. Physical Review
Letters, 70:1895, 1993.
[82] Y. C. Hou, G. F. Zhang, Y. Chen, and H. Fan. Preservation of Entanglement
in a Two-qubit-spin Coupled System. Annals of Physics, 327:292, 2012.
[83] Agedi N. Boto, Pieter Kok, Daniel S. Abrams, Samuel L. Braunstein,
Colin P. Williams, and Jonathan P. Dowling. Quantum Interferometric Op-
tical Lithography: Exploiting Entanglement to Beat the Diffraction Limit.
Physical Review Letters, 85:2733, 2000.
Bibliography 165
[84] Hwang Lee, Pieter Kok, and Jonathan P. Dowling. A Quantum Rosetta
Stone for Interferometry. Journal of Modern Optics, 49(14):2325, 2002.
[85] Barry C. Sanders. Quantum Dynamics of the Nonlinear Rotator and the
Effects of Continual Spin Measurement. Physical Review A, 40:2417, 1989.
[86] Tomohisa Nagata, Ryo Okamoto, Jeremy L. O’Brien, Keiji Sasaki, and
Shigeki Takeuchi. Beating the Standard Quantum Limit with Four-
Entangled Photons. Science, 316(5825):726, 2007.
[87] Jonathan A. Jones, Steven D. Karlen, Joseph Fitzsimons, Arzhang Ardavan,
Simon C. Benjamin, G. Andrew D. Briggs, and John J. L. Morton. Magnetic
Field Sensing Beyond the Standard Quantum Limit Using 10-Spin NOON
States. Science, 324(5931):1166, 2009.
[88] Adriano Barenco, David Deutsch, Artur Ekert, and Richard Jozsa. Condi-
tional Quantum Dynamics and Logic Gates. Physical Review Letters, 74:
4083, 1995.
[89] K. R. K. Rao and Anil Kumar. Entanglement in a 3-spin Heisenberg XY
Chain with Nearest Neighbor Interactions, Simulated in an NMR Quan-
tum Simulator. International Journal of Quantum Information, 10:1250039,
2012.
[90] Debbie. W. Leung, Isaac. L. Chuang, Fumiko Yamaguchi, and Yoshihisa
Yamamoto. Efficient Implementation of Coupled Logic Gates for Quantum
Computation. Physical Review A, 61:042310, 2000.
[91] Sougato Bose. Quantum Communication Through an Unmodulated Spin
Chain. Physical Review Letters, 91:207901, 2003.
[92] Matthias Christandl, Nilanjana Datta, Artur Ekert, and Andrew. J. Lan-
dahl. Perfect State Transfer in Quantum Spin Networks. Physical Review
Letters, 92:187902, 2004.
[93] Stephanie Simmons, Jonathan A. Jones, Steven D. Karlen, Arzhang Arda-
van, and John J. L. Morton. Magnetic Field Sensors Using 13-spin Cat
States. Physical Review A, 82:022330, 2010.
[94] Malcolm H. Levitt. Composite Pulses. Progress in Nuclear Magnetic Reso-
nance Spectroscopy, 18:61, 1986.
Bibliography 166
[95] Malcolm H. Levitt and Ray Freeman. NMR Population Inversion Using a
Composite Pulse. Journal of Magnetic Resonance, 33:473, 1979.
[96] Jonathan. A. Jones. Robust Ising Gates for Practical Quantum Computa-
tion. Physical Review A, 67:012317, 2003.
[97] C. Di Franco, M. Paternostro, and M. S. Kim. Perfect State Transfer on
a Spin Chain without State Initialization. Physical Review Letters, 101:
230502, 2008.
[98] E. L. Hahn. Spin Echoes. Physical Review, 80:580, 1950.
[99] Malcolm H. Levitt and Ray Freeman. Composite Pulse Decoupling. Journal
of Magnetic Resonance, 43:502, 1981.
[100] Malcolm H. Levitt, Ray Freeman, and Thomas Frenkiel. Broadband Het-
eronuclear Decoupling. Journal of Magnetic Resonance, 47:328, 1982.
[101] B. M. Fling. Improvement on Broadband Decoupling in Liquids. Journal of
Magnetic Resonance, 60:424, 1984.
[102] Malcolm H. Levitt. Symmetrical Composite Pulse Sequences for NMR Pop-
ulation Inversion. I. Compensation of Radio frequency Field Inhomogeneity.
Journal of Magnetic Resonance, 48:234, 1982.
[103] A. J. Shaka and Ray Freeman. Composite Pulses with Dual Compensation.
Journal of Magnetic Resonance, 55:487, 1983.
[104] Malcolm H. Levitt and R. R. Ernst. Improvement of Pulse Performance in
NMR Coherence Transfer Experiments. Molecular Physics, 50:1109, 1983.
[105] A. J. Shaka, Christopher Bauer, and Ray Freeman. Selective Population
Transfer Effects in Nuclear Overhauser Experiments. Journal of Magnetic
Resonance, 60:479, 1984.
[106] http://www.wolfram.com/mathematica/.
[107] S. Odedra, M. J. Thrippleton, and S. Wimperis. Dual-Compensated An-
tisymmetric Composite Refocusing Pulses for NMR. Journal of Magnetic
Resonance, 225:81, 2012.
Bibliography 167
[108] S. Odedra and S. Wimperis. Improved Background Suppression in 1H MAS
NMR Using Composite Pulses. Journal of Magnetic Resonance, 221:41–50,
2012.
[109] C. K. Anand, A. D. Bain, A. T. Curtis, and Z. Nie. Designing Optimal
Universal Pulses Using Second-Order, Large-Scale, Non-linear Optimization.
Journal of Magnetic Resonance, 219:61, 2012.
[110] S. Odedra and S. Wimperis. Use of Composite Refocusing Pulses to Form
Spin Echoes. Journal of Magnetic Resonance, 214:68, 2012.
[111] G. T. Genov, B. T. Torosov, and N. V. Vitanov. Optimized Control of Mul-
tistate Quantum Systems by Composite Pulse Sequences. Physical Review
A, 84(6), 2011.
[112] N. V. Vitanov. Arbitrarily Accurate Narrowband Composite Pulse Se-
quences. Physical Review A, 84(6), 2011.
[113] U. Holzgrabe. Quantitative NMR Spectroscopy in Pharmaceutical Applica-
tions. Progress in Nuclear Magnetic Resonance Spectroscopy, 57:229, 2010.
[114] S. Heikkinen, M. M. Toikka, P. T. Karhunen, and I. A. Kilpelainen. Quanti-
tative 2D HSQC (Q-HSQC) via Suppression of J-dependence of Polarization
Transfer in NMR Spectroscopy: Application to Wood Lignin. Journal of the
American Chemical Society, 125(14):4362, 2003.
[115] D. D. Werstler. Quantitative 13C NMR Characterization of Aqueous
Formaldehyde Resins: 1. Phenol-formaldehyde Resins. Polymer, 27:750,
1986.
[116] S. L. Patt and J. N. Shoolery. Attached Proton Test for Carbon-13 NMR.
Journal of Magnetic Resonance, 46:535, 1982.
[117] W. A. Anderson and R. Freeman. Influence of a Second Radio frequency
Field on High-resolution Nuclear Magnetic Resonance Spectra. The Journal
of Chemical Physics, 37:85, 1962.
[118] Gareth A. Morris and Ray Freeman. Enhancement of Nuclear Magnetic Res-
onance Signals by Polarization Transfer. Journal of the American Chemical
Society, 101:760, 1979.
Bibliography 168
[119] Terry J. Henderson. Sensitivity-Enhanced Quantitative 13C NMR Spec-
troscopy via Cancellation of 1JCH Dependence in DEPT Polarization Trans-
fers. Journal of the American Chemical Society, 126(12):3682, 2004.
[120] Bin Jiang, Nan Xiao, Huili Liu, Zhiming Zhou, Xi-an Mao, and Maili Liu.
Optimized Quantitative DEPT and Quantitative POMMIE Experiments for13C NMR. Analytical Chemistry, 80(21):8293, 2008.
[121] A. V. Makela, I. Kilpelainen, and S. Heikkinen. Quantitative 13C NMR Spec-
troscopy Using Refocused Constant-Time INEPT, Q-INEPT-CT. Journal
of Magnetic Resonance, 204:124, 2010.
[122] Stephen Wimperis and Geoffrey Bodenhausen. Heteronuclear Coherence
Transfer Over a Range of Coupling Constants. A Broadband-INEPT Exper-
iment. Journal of Magnetic Resonance, 69:264, 1986.
[123] A. M. Torres, W. A. Bubb, D. J. Philp, and P. W. Kuchel. Improved
J-compensated Sequences Based on Short Composite Pulses. Journal of
Magnetic Resonance, 194:81, 2008.
[124] S. Ehni and B. Luy. A Systematic Approach for Optimizing the Robustness
of Pulse Sequence Elements with Respect to Couplings, Offsets, and B1-field
Inhomogeneities (COB). Magnetic Resonance in Chemistry, 50:S63, 2012.
[125] N. T. Nyberg and O. W. Sørensen. Multiplicity-Edited Broadband HMBC
NMR Spectra. Magnetic Resonance in Chemistry, 44:451, 2006.
[126] D. P. Burum and R. R. Ernst. Net Polarization Transfer via a J-ordered
State for Signal Enhancement of Low-Sensitivity Nuclei. Journal of Magnetic
Resonance, 39:163, 1980.
[127] A. Chipperfield, P. J. Fleming, H. Pohlheim, and C. M. Fonseca. Genetic
Algorithm Toolbox for Use with MATLAB. 1993.
[128] A. Nielsen, Michael and L. Chuang, Isaac. Quantum computation and Quan-
tum Information. Cambridge university press, 2010.
[129] Edward Fredkin and Tommaso Toffoli. Conservative Logic. International
Journal of Theoretical Physics, 21:219, 1982.
[130] M. Fleischhauer and M. D. Lukin. Quantum Memory for Photons: Dark-
state Polaritons. Physical Review A, 65:022314, 2002.
Bibliography 169
[131] Kyriakos N. Sgarbas. A Quantum Probability Splitter and Its Application
to Qubit Preparation. Quantum Information Processing, 12:601, 2013.
[132] http://www.mathworks.in/help/optim/ug/optimtool.html.
[133] http://www.mathworks.in/help/gads/ga.html.
[134] Jonathan Walgate, Anthony J. Short, Lucien Hardy, and Vlatko Vedral. Lo-
cal Distinguishability of Multipartite Orthogonal Quantum States. Physical
Review Letters, 85:4972, 2000.
[135] S. Virmani, M.F. Sacchi, M.B. Plenio, and D. Markham. Optimal local
discrimination of two multipartite pure states. Physics Letters A, 288:62,
2001.
[136] Sibasish Ghosh, Guruprasad Kar, Anirban Roy, Aditi Sen(De), and Ujjwal
Sen. Distinguishability of Bell States. Physical Review Letters, 87:277902,
2001.
[137] P. W. Shor. Algorithms for Quantum Computation: Discrete Logarithms
and Factoring. In Foundations of Computer Science, 1994 Proceedings., 35th
Annual Symposium on, page 124. IEEE, 1994.
[138] G. Brassard, P. Høyer, and A. Tapp. Quantum Counting. ICALP, 1998.
[139] Manu Gupta and Prasanta K. Panigrahi. Deterministic Bell State Discrim-
ination. arXiv preprint quant-ph/0504183, 2005.
[140] M. Zukowski, A. Zeilinger, M. A. Horne, and A. K. Ekert. “Event-Ready-
Detectors” Bell Experiment via Entanglement Swapping. Physical Review
Letters, 71:4287, 1993.
[141] Arun K. Pati. Minimum Classical Bit for Remote Preparation and Measure-
ment of a Qubit. Physical Review A, 63:014302, 2000.
[142] M. Greenberger, Daniel, A. Horne, Michael, and Anton Zeilinger. Going
beyond Bells theorem. In Bells theorem, Quantum Theory and Conceptions
of the Universe, page 69. Springer, 1989.
[143] Jiu-Cang Hao, Chuan-Feng Li, and Guang-Can Guo. Controlled Dense Cod-
ing Using the Greenberger-Horne-Zeilinger State. Physical Review A, 63:
054301, 2001.
Bibliography 170
[144] Li Xiao, Gui Lu Long, Fu-Guo Deng, and Jian-Wei Pan. Efficient Multiparty
Quantum-Secret-Sharing Schemes. Physical Review A, 69:052307, 2004.
[145] David G. Cory, Mark D. Price, and Timothy F. Havel. Nuclear Magnetic
Resonance Spectroscopy: An Experimentally Accessible Paradigm for Quan-
tum Computing. Physica D: Nonlinear Phenomena, 120:82, 1998.
[146] O. W. Sørensen, G. W. Eich, M. H. Levitt, G. Bodenhausen, and R. R.
Ernst. Product Operator Formalism for the Description of NMR Pulse Ex-
periments. Progress in Nuclear Magnetic Resonance Spectroscopy, 16:163,
1984.
[147] C. H. Tseng, S. Somaroo, Y. Sharf, E. Knill, R. Laflamme, T. F. Havel, and
D. G. Cory. Quantum Simulation of a Three-Body-Interaction Hamiltonian
on an NMR Quantum Computer. Physical Review A, 61:012302, 1999.
Index
adaptive feasible, 20
adiabatic evolution, 41
adiabatic quantum computation, 9
AED, 52, 53
angular momentum, 2
APT, 89
average experimental deviation, 50
average Hamiltonian, 50, 113
Avogadro’s number, xxvii
BB-INEPT, 90
Bell state, 32, 140, 143, 145
Bloch, 1
Bloch sphere, 7, 8
Bodenhausen, 90
Bohr magneton, xxvii
Boltzmann constant, xxvii
Boltzmann distribution, 11
Boltzmann statistics, 6
chemical shift, 4, 27, 91
coherence transfer, 69
composite pulse, 69, 72, 92
concurrence, 53
controlled controlled NOT, 104, 105
controlled Hadamard, 103, 104
controlled NOT, 9, 28, 30, 31, 69, 72, 86
controlled operators, 9, 141, 153
controlled phase, 41, 49, 51
controlled SWAP, 105, 106, 113, 138
cost function, 14
crossover, 16, 19, 92, 116
crossover fraction, 92
crossover heuristic, 19
crossover intermediate, 19
crossover scattered, 19
D WAVE, 10
Darwinian evolution, 14
David. J. Wineland, 7
decision variable, 14
decoherence, 10, 53
delay zeroing, 116
dense coding, 145, 146
density matrix, 11, 12
density matrix tomography, 151, 152
DEPT, 96, 99
Deutsch-Jozsa algorithm, 10, 38
diamagnetic shielding, 3, 4
dipolar coupling, 5
Dzyaloshinskii-Moriya, 43
Einstein-Podolsky-Rosen, 145
electron mass, xxvii
elementary charge, xxvii
elite count, 92, 115
elite individuals, 16
entanglement dynamics, 53
entanglement preservation, 54, 55
entanglement swapping, 145
equilibrium state, 11
Ernst, 1
excitation, 72
expm, 71
171
Index 172
factorization, 38
ferromagnetism, 43
Feynman, 37
fidelity profile optimization, 37–39, 42,
44, 51, 52
fidelity space, 38
fitness function, 16, 25, 26, 39, 72, 73,
92, 93
GAQIC, 20, 90, 91, 93, 98
genetic algorithm, 16
GHZ state, 140, 146, 151
gradient ascent pulse engineering, 23
Grover’s search algorithm, 11
Hadamard gate, 32
Heisenberg-XY, 39, 46, 49, 52
Hilbert space, 140, 143
homonuclear NMR, 27
INEPT, 89, 91
inversion, 72
Jharana, 146
John Holland, 14
Leggett-Garg inequality, 11, 38
linear spin chain, 52
LOCC, 139
logical labelling, 12
long lived singlet state, 24
longitudinal relaxation time, 49
magnetic dipoles, 5
Mathematica, 70
Matlab, 74, 83
methine, 91
methyl, 91
methylene, 91
min-max difference, 93, 96
MRI, 1
mutation, 16, 92, 115
mutation gaussian, 20
mutation operators, 19
no-hiding theorem, 11
NOE, 89
non-destructive discrimination, 146
NOON, 57
Nyberg, 90
objective function, 14, 16
original modulation, 96
parity estimation, 146
Pauli, 1
Pauli matrices, 8
permeability, xxvii
phase estimation algorithm, 139, 140,
142, 145–148
Planck constant, xxvii
polarization transfer, 72, 89–91
population inversion, 69
proton mass, xxvii
pseudo pure state, 11, 30, 32, 50, 147–
149
pulse zeroing, 116
pulsed field gradients, 24
Purcell, 1
Q-INEPT-CT, 90, 100
QIP, 37
quadrupolar interaction, 5
quantification, 89, 91
quantification efficiency, 93, 95, 96
quantitative INEPT, 16, 20
quantum counter, 106–108
Index 173
quantum gates, 8
quantum key distribution, 146
quantum probability splitter, 110, 111
quantum secret sharing, 146
quantum simulation, 11
quantum state discrimination, 140, 141
quantum state tomography, 12, 34, 50
Rabi, 1
recombination operators, 14
refocused INEPT, 91, 95–97
remote state preparation, 145
representation scheme, 24, 72, 92
reproduction operators, 18
rf pulse, 5
Roulette, 17, 116
SALLT, 12
scalar coupling, 4
Serge Haroche, 7
Shor’s algorithm, 11
single point crossover, 19
single qubit rotation, 8, 27
singlet state, 30, 32, 34
spatial averaging, 12, 30, 50, 148
spectral editing, 97
spin quantum numbers, 2
spin selective pulses, 26
spin-lock, 34
spin-orbit coupling, 43
SSQB, 143
state to state optimization, 24, 30
Stern and Gerlach, 1
stochastic uniform, 17
stopping criteria, 16
strongly modulated pulses, 23
superconductor, 10
teleportation, 145
temporal averaging, 12
thermal equilibrium, 6, 30, 34
time restricted modulation, 96
Torres, 90
transfer functions, 91
transverse relaxation time, 49
trapped ions, 10
two point crossover, 19
UOD, 38, 103
W state, 109, 110
WALTZ-16, 34
Zeeman, 2
Zeeman interaction, 4