1: \tolerance=10000
2: %\documentstyle[preprint,aps]{revtex}
3: \documentstyle[prl,aps,epsf]{revtex}
4: \baselineskip 0.9truecm
5: \large
6: \begin{document}
7: \twocolumn
8:
9: \title{Single-Pulse Preparation of the Uniform Superpositional
10: State\\ used in Quantum Algorithms}
11:
12: \author{{G.P. Berman$^1$, F. Borgonovi$^{2,3}$, F.M. Izrailev$^4$,
13: and V.I. Tsifrinovich$^5$}}
14:
15: \address{$^1$T-13 and CNLS,
16: Los Alamos National Laboratory, Los Alamos, New Mexico 87545}
17: %
18: \address{$^{2}$Dipartimento di Matematica e Fisica, Universit\`a Cattolica,
19: via Musei 41, 25121 Brescia, Italy}
20: %
21: \address{$^{3}$I.N.F.N., Sezione di Pavia and I.N.F.M.,
22: Gruppo Collegato di Brescia}
23: %
24: \address{$^{4}$Instituto de Fisica, Universidad Autonoma de Puebla, Apdo.
25: Postal J-48, Puebla 72570, Mexico}
26: %
27: \address{$^5$IDS Department, Polytechnic University,
28: Six Metrotech Center, Brooklyn NY 11201}
29:
30: \maketitle
31:
32: \begin{abstract}
33: We examine a single-pulse preparation of the uniform
34: superpositional wave function, which includes all basis states,
35: in a spin quantum computer.
36: The effective energy spectrum and the errors generated by this pulse
37: are studied in detail.
38: We show that, in spite of the finite width of the energy
39: spectrum bands, amplitude and phase errors can be made reasonably small.
40: %
41: \renewcommand{\baselinestretch}{1.656}
42: \pacs{PACS numbers: 03.67.Lx,~03.67.-a,~76.60.-k}
43: %
44: \end{abstract}
45: %
46:
47: 1.~Both the Shor and the Grover quantum algorithms begin with the preparation
48: of a uniform superposition of the basis states.
49: In Shor's algorithm, it is a superposition in the $x$-register
50: for the modular exponentiation: $a^x (\bmod N)$.
51: For Grover's algorithm, it is a superposition of all
52: possible entries of the unsorted data.
53: In the language of computer science,
54: the transformation of the ground state of the $L$-qubit register,
55: %
56: $$
57: |0_{L-1}0_{L-2}...0_10_0\rangle,\eqno(1)
58: $$
59: %
60: into the uniform superposition of all possible basic states,
61: %
62: $$
63: \Psi_{unif}={{1}\over{2^{L/2}}}\sum|n_{L-1}n_{L-2}...n_1n_0\rangle,
64: ~(n_k=0,1),\eqno(2)
65: $$
66: %
67: is provided by the Hadamard transformation \cite{1}.
68: In physical systems, this transformation can be implemented,
69: for example, using two different methods:
70: 1) by a selected $\pi/2$-pulse excitation of each qubit, and
71: 2) by non-selective excitation of all qubits using a single $\pi/2$-pulse.
72: The first method could be used, for example,
73: for a chain of spins connected by the Ising interaction.
74: Unfortunately, in this case non-resonant effects disturb
75: the uniform superposition.
76: The reason is that a $\pi/2$-pulse acts not only on the chosen resonant
77: spin but also on all other spins \cite{2}.
78: Besides, this method requires the application of $L$ pulses.
79: The second method requires only a single $\pi/2$-pulse.
80: This method also cannot provide a perfect uniform superposition (2).
81: The second method is used currently in developed statistical
82: ensemble quantum computation \cite{3,4,5}.
83:
84: In this paper, we analyze the second method. First we discuss the Hamiltonian
85: of the system, then the effective energy spectrum in the rotating
86: reference frame and, finally, the error generated by a $\pi/2$-pulse.
87: We present the results of numerical simulation of a chain which includes
88: 10 spins.
89: We show that, in spite of the finite width of the
90: energy spectrum bands, the amplitude and phase errors can be made
91: acceptably small.
92:
93: 2.~Consider a chain of spin 1/2 nuclei described by the operators $I_k$.
94: Assume that these spins have slightly different Larmor
95: frequencies, $\omega_k$ and are connected by Ising interactions.
96: In a liquid NMR quantum computation, one utilizes a statistical
97: ensemble of such chains \cite{3,4,5}.
98: To prepare a uniform superposition (2),
99: a $\pi/2$-pulse must be polarized along the $(-y)$-axis of the rotating
100: reference frame (if initially the nuclear spins point in the
101: positive $z$-direction).
102: The Hamiltonian of the system in the rotating frame is \cite{2},
103: %
104: $$
105: {\cal H}=\sum_{k=0}^{L-1}[-(\omega_k-\omega)I^z_k+\Omega I^y_k]-2J
106: \sum_{k=0}^{L-2}I^z_kI^z_{k+1},\eqno(3)
107: $$
108: %
109: where $\hbar=1$, $\Omega$ is the amplitude of the
110: pulse in the frequency units (the Rabi frequency),
111: $\omega$ is the frequency of the pulse,
112: and $J$ is the constant of the Ising interaction.
113: During the action of a $\pi$/2-pulse, the second term,
114: $\Omega I^y_k$, in the Hamiltonian (3) is the main one,
115: the two other terms are supposed to provide small corrections to the spin
116: dynamics. To choose the values of parameters,
117: we assume that the following inequalities are
118: satisfied for our spin quantum computer,
119: %
120: $$
121: J\ll|\omega_{k+1}-\omega_k|,~\Omega\ll\omega_k.\eqno(4)
122: $$
123: %
124: Assuming $J/2\pi\sim 0.1$kHz,
125: $(\omega_{k+1}-\omega_k)/2\pi\sim 1$kHz,
126: $\omega_k/2\pi\sim 100$MHz,
127: we shall consider values of $\Omega/2\pi$ up to $10$MHz.
128:
129: 3.~Next, we shall discuss the effective energy spectrum
130: described by the Hamiltonian (3).
131: To understand the behavior of the energy spectrum, we first analyze
132: a system containing two spins only.
133: When the inhomogeneity and the Ising interaction
134: are absent, the energy spectrum consists of three lines:
135: $E_0=-\Omega$, $E_1=0$, and $E_2=\Omega$.
136: The first level corresponds to the state, $|00\rangle_{-y}$,
137: where the index ``-y'' indicates that both spins point in the
138: (-y)-direction ($(-y)$-representation).
139: The twice degenerate level, $E_1$, corresponds
140: to the states, $|01\rangle_{-y}$ and $|10\rangle_{-y}$,
141: and the energy level $E_2$, refers to the state $|11\rangle_{-y}$.
142: First, we consider the effect of inhomogeneity
143: when the Ising interaction is absent.
144: Assume, for example, that $\omega=\omega_0$
145: and $\omega_1=\omega_0+\Delta\omega$.
146: Then, the effective field for spin ``1'' in the frequency units is,
147: $
148: \Omega_{eff}=\sqrt{\Omega^2+\Delta\omega^2}.
149: $
150: %
151: The energy levels are given by the expressions,
152: %
153: $$
154: E_0=-(\Omega+\Omega_{eff})/2,~E_1=(\Omega-\Omega_{eff})/2,\eqno(6)
155: $$
156: %
157: %
158: $$
159: E^\prime_1=(\Omega_{eff}-\Omega)/2,~E_2=(\Omega+\Omega_{eff})/2.
160: $$
161: %
162: The main effect is the
163: splitting of the central line into two lines with the energy separation,
164: %
165: $$
166: \Delta E=E^\prime-E\approx(\Delta\omega)^2/2\Omega.
167: $$
168: %
169: Important fact is that this splitting
170: decreases as $1/\Omega$ as $\Omega$ increases.
171:
172: Let us consider the error generated by a non-resonant spin.
173: If the initial state of this spin is $|0\rangle_z$, then with
174: linear accuracy, $\Delta \omega/\Omega$,
175: the wave function of the spin, $\Psi(t)$, can be written as,
176: %
177: $$
178: \Psi(t)=[\cos(\Omega t/2)+
179: i(\Delta\omega/\Omega)\sin(\Omega t/2)]|0\rangle_z+\sin(\Omega t/2)|1\rangle_z.
180: $$
181: %
182: After a $\pi/2$-pulse ($\Omega t=\pi/2$), we have:
183: %
184: $$
185: \Psi(\pi/2\Omega)\approx
186: {{1}\over{\sqrt{2}}}\Bigg(e^{i\Delta\omega/\Omega}|0\rangle_z+|1\rangle_z\Bigg).
187: $$
188: %
189: Thus, in first order in $\Delta\omega/\Omega$,
190: the inhomogeneity generates only a phase error which decreases as $1/\Omega$.
191:
192: Now, we consider the effect of the Ising interaction
193: when the inhomogeneity is absent.
194: At first sight, the Ising interaction produces an
195: additional ``effective field'' in the $z$-direction,
196: and its influence on the energy spectrum
197: must also decrease as $\Omega$ increases.
198: %
199: \begin{figure}
200: \epsfxsize 6cm
201: \epsfbox{fig1.eps}
202: \narrowtext
203: \caption{
204: The energy spectrum as a function of $
205: \Omega$.
206: }
207: \label{fig1}
208: \end{figure}
209:
210: However, the ``effective field'' approach is not correct in this case.
211: It is easy the check that the ground state is $|00\rangle_{-y}$
212: with a small admixture of the state $|11\rangle_{-y}$.
213: It has the energy, $E_0=-\sqrt{\Omega^2+(J/2)^2}$.
214: The central energy splits into two levels $\pm J/2$, which correspond to
215: symmetric and antisymmetric superpositions of the states $|01\rangle_{-y}$
216: and $|10\rangle_{-y}$, and $E_2=-E_0$.
217: When the Rabi frequency, $\Omega$, increases,
218: the influence of the Ising interaction on the energy levels,
219: $E_0$ and $E_2$ decreases.
220: But the splitting between the central
221: energy levels does not change as $\Omega$ increases:
222: $\Delta E_1=J$. One might expect that this splitting will
223: generate an error which does not decrease as $\Omega$ increases.
224: Fortunately, it does not happen.
225: If both spins point initially in the positive $z$-direction, the wave function, $\Psi(t)$, can be written as,
226: %
227: $$
228: \Psi(t)\approx (1/2)[e^{i\Omega t}|00\rangle_{-y}-
229: e^{-i\Omega t}|11\rangle_{-y}
230: $$
231: %
232: %
233: $$
234: +ie^{iJ t/2}(|01\rangle_{-y}+|10\rangle_{-y})],
235: $$
236: %
237: where we neglected the
238: terms $\sim(J/\Omega)^2$. At the end of the $\pi/2$-pulse, we have,
239: %
240: $$
241: \Psi(\pi/2\Omega)
242: \approx{{i}\over{\sqrt{2}}}
243: \Bigg[|00\rangle_{-y}+|11\rangle_{-y}+
244: e^{i\pi J/4\Omega}(|01\rangle_{-y}+|10\rangle_{-y})\Bigg].
245: $$
246: \begin{figure}
247: \epsfxsize 6cm
248: \epsfbox{fig1c.eps}
249: \narrowtext
250: \caption{The structure of a narrow strip of the central band shown in
251: Fig.\ref{fig1}.}
252: \label{fig1c}
253: \end{figure}
254:
255: For $J=0$, we have a uniform superposition of the basis states
256: (in both $(-y)$- and $z$-representations).
257: For $J\not=0$, to first order in $J/\Omega$,
258: the Ising interaction (similar to the inhomogeneity)
259: generates only a phase error which decreases as $1/\Omega$.
260: The non-vanishing bandwidth, $\Delta E_1=J$, does not cause
261: a non-vanishing error.
262: This happens because the phase error is proportional to the duration
263: of a $\pi/2$-pulse which is proportional to $1/\Omega$.
264:
265: 3.~Next, we present the results of numerical simulations
266: with $L=10$ qubits.
267: These simulations require operations on a digital computer
268: in the Hilbert space with dimension $D=2^{10}=1024$.
269: \begin{figure}
270: \epsfxsize 6cm
271: \epsfbox{bandnew.eps}
272: \narrowtext
273: \caption{Widths $\Delta E_6$,(left) $ \Delta E_4$ (right)
274: of the $6$th (central) and the 4th band vs $\Omega$.
275: Open circles are for $J=0$, crosses for $J=0.1$.
276: Dashed lines are the best fit to $A/\Omega$ (circles) where
277: $A=28.466$ (left) and $A=23.673$(right).
278: }
279: \label{bandnew}
280: \end{figure}
281:
282: If the Ising interaction and the inhomogeneity of the frequencies,
283: $\omega_k$, are both absent, the energy spectrum of the Hamiltonian (3)
284: consists of 11 equidistant levels separated by gaps with value $\Omega$.
285: Both the interaction and the inhomogeneity cause splitting of all inner levels.
286: This leads to formation of energy bands.
287:
288: Fig. \ref{fig1} shows the energy spectrum
289: as a function of $\Omega$, for the following values of parameters,
290: %
291: $$
292: \omega_{k+1}-\omega_k=1,~(k=0,...,9),~J=0.1,~\omega_k-\omega=k-4.5.
293: $$
294:
295: \begin{figure}
296: \epsfxsize 6cm
297: \epsfbox{ampli.eps}
298: \narrowtext
299: \caption{Dependence on $\Omega$ of the maximal
300: (open circles) and average errors (full circles
301: ) for the amplitude modulus, $|A_n|$, on $\Omega$.
302: The full line is the best fit, $0.2787/\Omega^2$, while the dashed is
303: $0.0953/\Omega^2$.
304: Other parameters are the same as in
305: Fig.\ref{fig1}.}
306: \label{ampli}
307: \end{figure}
308:
309:
310: The last equation means that the frequency $\omega$ of a $\pi/2$-pulse is equal
311: to the average Larmor frequency,
312: $\langle\omega_k\rangle=(1/10)\sum_{k=0}^{9}\omega_k$.
313: For each $\Omega$, there are 1024 energy levels.
314: At the scale shown in Fig. \ref{fig1}, each band is represented by a point.
315:
316: \begin{figure}
317: \epsfxsize 6cm
318: \epsfbox{pha.eps}
319: \narrowtext
320: \caption{Dependence of the maximal (open circles) and average errors (full circles
321: ) for the phase errors $|\phi_n|$, on $\Omega$.
322: Full line is the best fit $13.0216/\Omega$, while dashed is
323: $3.6606/\Omega$.
324: Other data are the same as in
325: Fig.\ref{fig1}.}
326: \label{pha}
327: \end{figure}
328:
329:
330: In fact, each band has a complicated structure. As an example, Fig. \ref{fig1c}
331: shows the structure of a narrow strip of the central band
332: as a function of $\Omega$.
333: %
334: Fig. \ref{bandnew} (left) shows the dependence of the width of the
335: central band, $\Delta E_6$, on $\Omega$ (in the logarithmic scale).
336: %
337: When the interaction between spins is absent (circles in
338: Fig. \ref{bandnew} ), $\Delta E_6$ decreases approximately
339: as $1/\Omega$. A finite interaction (crosses in Fig. \ref{bandnew})
340: changes this picture.
341: After the width of the band caused by the inhomogeneity
342: decreases to the value of approximately $6J$
343: (at $\Omega\approx 50$), its value does not decrease.
344: This dependence of $\Delta E_6$ on $\Omega$
345: is qualitatively similar to the results discussed above for two spins.
346: Fig. \ref{bandnew} (right) shows similar dependence for the the width
347: of the 4th band, $\Delta E_4$.
348:
349: \begin{figure}
350: \epsfxsize 6cm
351: \epsfbox{camp.eps}
352: \narrowtext
353: \caption{
354: The distribution of the complex amplitudes,
355: $A_n$, in the complex plane for $\Omega=10^3$
356: (left) and $\Omega=10^4$ (right).
357: Other parameters are the same as in
358: Fig.\ref{fig1}.}
359: \label{camp}
360: \end{figure}
361:
362: Now we consider the errors generated in the process of preparation of a
363: uniform superposition of the basis states, $\Psi_{unif}$.
364:
365: In the rotating frame, the spin dynamics
366: can be described as a superposition of stationary
367: solutions with constant coefficients, which can be found
368: from the initial conditions.
369: In the absence of both the interaction between spins and the
370: inhomogeneity, a $\pi/2$-pulse applied to the ground state,
371: $|0_{L-1}...0_0\rangle_z$, generates a uniform superposition of
372: all $2^L$ basic states with the amplitudes $1/\sqrt{2^L}$.
373:
374: \begin{figure}
375: \epsfxsize 6cm
376: \epsfbox{pofs.eps}
377: \narrowtext
378: \caption{Level spacing distribution for a system of $L=12$ spins;
379: $\Omega=100$; $\omega=\omega_0$, $\omega_k=\omega_0+k$,
380: $J=20$ (upper); $J=100$ (lower).
381: The continuous line shows the
382: Poisson distribution, while the dashed curve shows the GOE distribution.
383: }
384: \label{pofs}
385: \end{figure}
386:
387:
388: To describe the error in the complex amplitude,
389: $A_n=|A_n|\exp(i\Phi_n)$,
390: of the state $|n\rangle_z$
391: $(0\le n\le 2^{L-1})$, we use two quantities:
392: 1) $\eta=|2^{-L/2}-|A_n||$ which describes the error of the amplitude modulus,
393: $|A_n|$,
394: and 2) the phase modulus,
395: $|\Phi_n|$, which describes the phase error (as $\Phi_n=0$ in the ideal case).
396: Fig. \ref{ampli} shows the dependences of the
397: maximal error, $\eta_{max}$ and the average error, $\eta_{ave}$,
398: on the Rabi frequency, $\Omega$.
399: One can see that both quantities decrease approximately as $1/\Omega^2$.
400: %
401: Fig. \ref{pha} shows similar dependences of the phase error on $\Omega$.
402: Both, $|\Phi|_{max}$ and $|\Phi|_{ave}$ decrease approximately as $1/\Omega$.
403: This corresponds to the results derived above for two spins.
404: Note that for parameters we chose (which correspond to $J\ll\Delta \omega$),
405: the main contribution to the error depends on the inhomogeneity. The
406: plots presented in Figs \ref{ampli} and \ref{pha} do not change
407: significantly when $J=0$.
408: Fig. \ref{camp} shows the distribution of the complex
409: amplitudes, $A_n$, in the complex plane for $\Omega=10^3$ and
410: $\Omega=10^4$.
411: One can see that the distribution of $A_n$ has a
412: form of the arc whose length and width decrease as $\Omega$ increases.
413:
414: 4.~In this section, we note that the quantum
415: Hamiltonian (3) belongs to the so-called quantum non-integrable systems.
416: This means that, treated classically, these systems exhibit chaotic
417: behavior for some range of parameters and initial conditions.
418: (See, for example, \cite{6} and references therein.)
419: The quantum properties of classically chaotic systems can be investigated
420: by a detailed analysis of their eigenvalues and eigenfunctions\cite{frep}.
421: This analysis will be the subject of future investigation.
422: Here, we concentrate on the statistics of neighboring energy levels spacing,
423: $P(s)$. Indeed it has been conjectured that $P(s)$ for a chaotic system
424: depends only on some general symmetries\cite{bohi}.
425: In our case (for Hermitian matrices),
426: we can assume the quantum chaos regime as
427: described by a Wigner-Dyson distribution.
428: Fig. \ref{pofs} shows the transition of
429: $P(s)$, from the Poisson distribution
430: to the Wigner-Dyson distribution of the Gaussian orthogonal ensemble (GOE)
431: of Hermitian matrices.
432:
433: As one can see from Fig. \ref{pofs}, the transition to chaos
434: appears for relatively large values of $J$ which are far beyond
435: the range of quantum computation.
436: Also, the appearance of the Poisson distribution, which should be
437: a fingerprint of integrability, appears at relatively high values of
438: $J$, out of quantum computation regime.
439: Level statistics in the quantum computation regime will be the
440: subject of future investigations.
441:
442: 5.~In conclusion, we investigated the errors generated by a
443: single-pulse implementation of the Hadamard transformation for
444: a chain of spins connected by the Ising interaction.
445: In the rotating reference frame, the interaction between spins
446: and the inhomogeneity of the Larmor frequencies split the
447: energy levels into the bands.
448: The characteristic width of the band caused by the inhomogeneity
449: decreases as $1/\Omega$ as the Rabi frequency, $\Omega$, increases.
450: The contribution to the band widths due to the Ising
451: interaction between spins remains constant.
452: In spite of this fact, errors generated in the process of
453: preparation of the uniform wave function decrease monotonically
454: as $\Omega$ increases.
455: When the Rabi frequency, $\Omega$, increases, the errors of the
456: amplitude modulus, $|A_n|$, decrease as $1/\Omega^2$.
457: The phase errors decrease only as $1/\Omega$.
458: For reasonable values of $\Omega$, the errors of the amplitude's
459: modulus, $|A_n|$, become negligible.
460: The phase error can be reduced to the order of $10^{-5}$rad.
461: Thus, both the errors caused by the inhomogeneity and the interaction
462: between spins can be made reasonably small, for a single-pulse
463: generation of the superpositional wave function used
464: in the main quantum algorithms.
465: %
466: \section*{Acknowledgments}
467: %
468: The work of GPB and VIT was supported by the Department of
469: Energy (DOE) under contract
470: W-7405-ENG-36, by the National Security Agency (NSA) and Advanced Research
471: and Development Activity (ARDA). FB acknowledges financial support from INFN and INFM.
472: %
473: \begin{thebibliography}{99}
474: %
475: \bibitem{1}
476: A.O. Pittenger, {\it An Introduction to Quantum Computing
477: Algorithms (Progress in Computer Science
478: and Applied Logic, V. 19) }, Birkh\"auser, Boston, 1999.
479: %
480: \bibitem{2}
481: G.P. Berman, G.D. Doolen, R. Mainieri, V.I. Tsifrinovich,
482: {\it Introduction to Quantum Computers},
483: World Scientific Publishing Company, 1998.
484: %
485: \bibitem{3}
486: D.G. Cory, A.F. Fahmy, T.F. Havel,
487: {Proc. Natl. Acad. Sci. USA}. 94, 1634 (1997).
488: %
489: \bibitem{4}
490: I.L. Chuang, N.A. Gershefeld,
491: M. Kubinec, {\it Phys. Rev. Lett}, {\bf 80}, 3408 (1998).
492: %
493: \bibitem{5}
494: N.A. Gershenfeld, I.L. Chuang, {\it Science}, {\bf 275}, 350 (1997).
495: %
496: \bibitem{6}
497: L.E. Reichl, {\it The Transition to Chaos}, Springer-Verlag, 1992.
498: %
499: \bibitem{frep}
500: F.M.Izrailev,
501: Phys. Rep. {\bf 196}, 299 (1990)
502: %
503: \bibitem{bohi}
504: O.Bohigas, in Proceedings of the 1898 Les Houches Summer School
505: {\it Chaos and Quantum Physics}, eds M.J.Giannoni, A.Voros, J.Zinn-Justin
506: (Elsevier Science Publisher B.V. North-Holland, Amsterdam, 1991), p.89
507: %
508: \end{thebibliography}
509: %
510: \end{document}
511: