1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %%%%%% M\"obius Problem, Jan. 24 2003
3: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
4:
5: \documentclass[aps,prb,twocolumn,floats,showpacs]{revtex4}
6: %\documentclass[prb,preprint,endfloats,showpacs,titlepage,groupedaddress]{revtex4}
7: %\usepackage{epsfig}
8: \usepackage{graphicx}
9: \usepackage{amsmath}
10: \setlength{\textheight}{24.5cm}
11: \newcommand{\half}{\mbox{\small $\frac{1}{2}$}}
12: \newcommand{\hide}[1]{}
13:
14: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
15:
16: \begin{document}
17:
18: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
19:
20: \title{Persistent currents in M\"obius strips}
21:
22: \author{\large Kousuke Yakubo$^1$,
23: Yshai Avishai$^{1,2,3}$ and Doron Cohen$^{2}$
24: }
25:
26: \affiliation{$^1$Department of Applied Physics,
27: Hokkaido University, Sapporo 060-8628, Japan \\
28: $^2$Department of Physics and $^3$Ilse Katz Center for Nanotechnology
29: Ben Gurion University, Beer Sheva, Israel}
30:
31:
32: \begin{abstract}
33: Relation between the geometry of a two-dimensional sample and its
34: equilibrium physical properties is exemplified here for a system
35: of non-interacting electrons on a M\"obius strip. Dispersion
36: relation for a clean sample is derived and its persistent current
37: under moderate disorder is elucidated, using statistical analysis
38: pertinent to a single sample experiment. The flux periodicity is
39: found to be distinct from that in a cylindrical sample, and the
40: essential role of disorder in the ability to experimentally
41: identify a M\"obius strip is pointed out.
42: \end{abstract}
43:
44: \pacs{73.23.Ra, 72.15.Rn, 72.10.-d}
45:
46: \maketitle
47:
48: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
49: \section{Introduction}
50: \label{sec1}
51:
52: An important theme in quantum mechanics is to find a relation
53: between a global geometry of a sample ({\it e.g.}, boundary
54: conditions) and its physical properties. We address this issue by
55: comparing flux periodicity of persistent currents in a cylinder
56: and in a M\"obius strip. The aim is to determine whether the
57: geometrical (in some sense topological) difference is tangible and
58: experimentally observable. At zero temperature, the persistent
59: current $I(\phi)$ in a ring can be expressed as\cite{Imry}
60: %
61: \begin{equation}
62: I(\phi)=-\frac{\partial E(\phi)} {\partial \phi}=
63: \sum_{n=1}^{\infty} I_{n} \sin (2 \pi n \phi),
64: \label{Iphi}
65: \end{equation}
66: %
67: where $\phi$ is the magnetic flux threading the ring in units of
68: $\Phi_0=hc/e$, $E(\phi)$ is the ground state energy, and $I_n$ are
69: the current harmonics.
70:
71: The current $I(\phi)$ is an anti-symmetric and periodic function
72: of $\phi$ with period~1. Possible occurrence of smaller flux
73: periodicity in mesoscopic physics is one of the cornerstones of
74: weak localization. For the cylinder geometry, conductance
75: measurements\cite{AASS} and magnetization of $10^7$ copper
76: rings\cite{Levy} indicate the emergence of periodicity $1/2$. It
77: is shown to be intimately related to the procedure of {\em
78: averaging} over disorder realizations and numbers of electrons in
79: the rings\cite{Gefen,Gilles,AGI,AASS}. Very recently, a
80: microscopic NbSe$_{3}$ M\"obius strip has been
81: fabricated\cite{Tanda}. Obviously, in this case, attention should
82: be focused on a {\it single sample} measurement\cite{Webb} for
83: which there is no self-averaging.
84:
85: Let us first mention several intuitive points relevant to the flux
86: periodicity in the M\"obius strip, based on semi-classical
87: arguments and geometry\cite{Hayashi}. First, recall that the
88: periodicity is related to interference between trajectories (such
89: as Aharonov-Bohm interference between different trajectories or
90: weak-localization interference between time-reversed paths). In
91: the cylinder (M\"obius) geometry, an electron moving in the {\em
92: longitudinal direction} along the ring encircles the system once
93: ({\em twice}) before returning to its initial position. Therefore,
94: we might expect different flux periodicities of the persistent
95: current between the two cases. Second, unlike a cylinder which can
96: be \lq\lq pressed\rq\rq\ into a one-dimensional ring, the
97: M\"obius strip cannot be pressed into a one-dimensional structure.
98: This brings in another important factor, namely, the motion of
99: electrons in the {\em transverse} direction. In a tight-binding
100: model this motion is controlled by the transverse hopping. If it
101: is very weak, the twice-encircling property of the M\"obius strip
102: implies the dominance of even harmonics $I_{2n}$. Contrary, for a
103: strong transverse hopping, the current in the M\"obius strip is
104: expected to be effectively similar to that in the cylindrical
105: strip\cite{Mila}. In the following we are mainly interested in a
106: regime where the transverse hopping is slightly less than with the
107: longitudinal one. Third, the role of disorder should be carefully
108: examined. Weak disorder is not expected to significantly alter
109: interference between semi-classical trajectories discussed above,
110: while strong disorder should result in a reduced sensitivity to
111: the pertinent geometry, due to localization effects. The most
112: intriguing disorder effect might then be expected in a moderate
113: strength of disorder which will be used below. The upshot of the
114: present study is that the periodicity pattern in a M\"obius strip
115: is remarkably distinct from that of a cylinder, and that disorder
116: plays a crucial role in making the statistical effect detectable.
117:
118: \section{Model}
119: \label{sec2}
120:
121: A M\"obius strip is modelled by considering a non-interacting
122: particle in a rectangle of length $L_x$ and width $L_y$, requiring
123: its wave-function $\psi(x,y)$ to satisfy Dirichlet boundary
124: conditions in the $y$ direction, and M\"obius boundary conditions
125: \cite{rmrk} in the $x$ direction:
126: %
127: \begin{align}
128: &\psi(x,-L_{y}/2) = \psi(x,L_{y}/2) = 0 &&\text{(Dirichlet B.C.)}\ ,\\
129: &\psi(x{+}L_x,y) = \psi(x,-y) &&\text{(M\"obius B.C.)}\ .
130: \label{BC}
131: \end{align}
132: %
133: The quantized wave-numbers are $k_y = (\pi/L_y)n_y$ and $k_x =
134: (2\pi/L_x)([\half]_{n_{y}}+n_x)$, where $n_y=1,2,\cdots$ and
135: $n_x=0,\pm1,\pm2,\cdots$. The notation $[\alpha]_{n}$ represents
136: $\alpha$ for $n=\text{even}$ and $0$ for $n=\text{odd}$. In the
137: cylinder geometry, Eq.~(3) should be replaced by
138: $\psi(x+L_{x},y)=\psi(x,y)$, and gives $k_{x}=(2\pi/L_{x})n_{x}$.
139: Thus, only the $n_y=\text{even}$ eigenstates are affected by the
140: switch from the conventional cylinder (periodic) boundary
141: conditions to the M\"obius ones.
142:
143: In the absence of disorder, the energies of the eigenstates both
144: in the M\"obius and cylinder strips are given by the formula
145: %
146: \begin{equation}
147: E_{n_x n_y} =
148: \epsilon_x\left( k_x-\frac {2\pi\phi} {L_x}\right) + \epsilon_y(k_y),
149: \label{Enxny}
150: \end{equation}
151: %
152: where $\epsilon_x$ and $\epsilon_y$ provide the dispersion
153: relation. Equation (\ref{Enxny}) is rather general for clean
154: systems. To be more specific, let us model the M\"obius strip by a
155: tight-binding Hamiltonian. The M\"obius strip is constructed from
156: a rectangular lattice including $N\times 2M$ sites. The rectangle
157: is twisted by $180^{\circ}$, and its two sides are connected, such
158: that longitudinal wire $1$ is attached to wire $2M$, wire $2$ is
159: attached to wire $2M-1$ and so on. The M\"obius strip so
160: constructed includes $M$ longitudinal wires with $2N$ sites on
161: each one. The Hamiltonian is then
162: %
163: \begin{eqnarray}
164: H_{\text{M\"obius}}=
165: \sum_{n=1}^{2N} \sum_{m=1}^{M}\Big[
166: \varepsilon_{nm}c^{\dagger}_{nm} c_{nm}
167: -t_{1}e^{-2 \pi i \phi/N}c^{\dagger}_{nm} c_{n+1m} \Big]
168: \nonumber \\
169: \hspace*{0.5cm}
170: - t_{2} \sum_{n=1}^{2N} \sum_{m=1}^{M-1}c^{\dagger}_{nm+1} c_{nm}
171: - \frac{t_{2}}{2} \sum_{n=1}^{2N}c^{\dagger}_{nM} c_{n+NM}
172: +\text{h.c.}
173: \hspace*{0.5cm}
174: \label{HM}
175: \end{eqnarray}
176: %
177: where $c_{nm}$ is the fermion operator at the site $(n,m)$
178: ($n=1,2,\dots, 2N$, $m=1,2,\dots,M$) and $t_{1}$ and $t_{2}$ are
179: longitudinal and transverse hopping amplitudes respectively. The
180: quantity $\varepsilon_{nm}$ is the site energy. Connecting the two
181: sides of the rectangle without twisting, we obtain a cylindrical
182: strip which includes $2M$ longitudinal wires composed of $N$
183: sites. The Hamiltonian of the cylinder is
184: %
185: \begin{eqnarray}
186: H_{\text{cylinder}}=
187: \sum_{n=1}^{N} \sum_{m=1}^{2M}\Big[
188: \varepsilon_{nm}c^{\dagger}_{nm} c_{nm}
189: -t_{1}e^{-2 \pi i \phi/N}c^{\dagger}_{nm} c_{n+1m} \Big]
190: \nonumber \\
191: \hspace*{0.5cm}
192: - t_{2} \sum_{n=1}^{N} \sum_{m=1}^{2M-1}c^{\dagger}_{nm+1} c_{nm}
193: +\text{h.c.}
194: \hspace*{3.0cm}
195: \label{HC}
196: \end{eqnarray}
197: %
198: Locally the two Hamiltonians (\ref{HM}) and (\ref{HC}) look the
199: same. But there is a couple of essential differences between them:
200: a) The M\"obius Hamiltonian (\ref{HM}) includes an extra term
201: which describes long range hopping between distant parts of the
202: $M$th wire \cite{rmrk}. b) While the magnetic phase accumulated
203: along the longitudinal direction on each link is the same (that
204: is, $2 \pi \phi /N$), the corresponding number of links is
205: different ($2N$ for the M\"obius strip and $N$ for the cylinder).
206: %
207:
208: \section{The Spectrum}
209: \label{sec3}
210:
211: We first consider a system without disorder, namely,
212: $\varepsilon_{nm}=0$. The dispersion relation for an electron in
213: the M\"obius strip reads,
214: %
215: \begin{eqnarray}
216: E_{n_x n_y} =
217: &-& 2t_1 \cos \Bigg[\frac{2\pi}{N}\left( \Big[ \frac{1}{2}\Big]_{n_{y}}
218: +n_x-\phi\right)\Bigg] \nonumber \\
219: &-& 2t_2 \cos \left(\frac{\pi}{2M{+}1} n_y \right)\ ,
220: \label{Enxny1}
221: \end{eqnarray}
222: %
223: where $n_x=1,\cdots,N$ and $n_y=1,\cdots,2M$. Defining new indexes
224: $k=[1]_{n_{y}}+2n_{x}$ and $q=[\half]_{k}+n_{y}/2$, one obtains a
225: more suggestive form,
226: %
227: \begin{eqnarray}
228: E_{kq} =
229: &-&2t_1 \cos \Big[\frac{\pi}{N}(k-2\phi)\Big] \nonumber \\
230: &-&2t_2 \cos \Big[\frac{\pi}{2M+1} \left(2q-[1]_{k}\right) \Big],
231: \label{EkqM}
232: \end{eqnarray}
233: %
234: where $k=1,\cdots,2N$ and $q=1,\cdots,M$. It is instructive to
235: compare it with the energy in the cylinder geometry, $E_{kq}=
236: -2t_1 \cos \big[\frac{2 \pi}{N}(k-\phi)\big] -2t_2 \cos
237: \left(\frac{\pi}{2M+1} q \right)$, where $k=1,\cdots,N$ and
238: $q=1,\cdots,2M$. Despite the apparent similarity between these two
239: spectra, there are at least two important differences. First, the
240: combination of flux and longitudinal momentum is distinct, namely,
241: it is $k-\phi$ for the cylinder and $k-2\phi$ for the M\"obius
242: strip. For a small ratio $t_2/t_1$ this might affect the
243: periodicity of the current\cite{Mila}. Second, the mini-band
244: structure is different.
245: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
246: \begin{figure}[tttt]
247: \includegraphics[width=\hsize]{fig1.eps}
248: \caption{Single-particle energy spectrum as a function of flux
249: threading the M\"obius ring. The $195$th-$200$th energy levels are
250: shown. The parameters are $N=20$, $M=10$, $t_{2}=0.5$, and
251: $W=0.5$. Energies are measured in units of $t_{1}$.}
252: \label{fig1}
253: \end{figure}
254: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
255:
256: We now turn to elucidate the current in disordered M\"obius
257: strips. The random numbers $\varepsilon_{nm}$ are assumed to be
258: uniformly distributed over the range $-W/2\le \varepsilon_{nm}\le
259: W/2$, where $W$ represents the strength of disorder. The
260: Hamiltonian Eq.~(\ref{HM}) [or Eq.~(\ref{HC})] is treated
261: numerically. As an example, the evolution of single-particle
262: energies with flux in a disordered M\"obius strip with $N=20$ and
263: $M=10$ is shown in Fig.~1. The parameters are $t_{2}/t_{1}=0.5$
264: and $W/t_{1}=0.5$. The pattern of avoided crossing turns out to be
265: remarkably different from that for a cylinder (see ref.~5 figure 1
266: therein). It must then be reflected in the behavior of persistent
267: currents.
268:
269: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
270: \begin{figure}
271: \includegraphics[width=8cm]{fig2.eps}
272: \caption{Fourier components of the persistent current for the
273: clean M\"obius strip as a function of the transverse-hopping
274: energy (a) at the half-filling and (b) below the half-filling. The
275: size of the M\"obius strip is given by $N=20$ and $M=10$.}
276: \label{fig2}
277: \end{figure}
278: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
279: The first stage of the analysis is an inspection of the typical
280: values of $I_n$, aiming in determination of their dependence on
281: the ratio $t_2/t_1$. As expected, in the absence of averaging we
282: find typical $I_1$ dominance in case of the cylinder geometry
283: irrespective of the $t_2/t_1$ ratio. For the M\"obius geometry the
284: emerging picture is quite different. Figure 2 shows the Fourier
285: components of the persistent current for a clean M\"obius strip as
286: a function of the ratio $t_{2}/t_{1}$ at and below the
287: half-filling ($N_{e}=200$). For small ratios ($t_2/t_1 < 0.1$) we
288: find, as can be naively expected, $I_2$ dominance.
289: %
290: %
291: %
292: %
293: The expected effect of averaging in the cylinder case is to
294: emphasize the $I_2$ contribution, while in the M\"obius case
295: the expected effect is to emphasize the $I_4$ contribution.
296: %
297: For clean M\"obius strip the $I_{n}$ with odd $n$,
298: as a function of the number of electrons $N_{e}$,
299: is anti-symmetric around half-filling.
300: Therefore $I_{1}$ and $I_{3}$ completely vanish [Fig.~2(a)].
301: See further discussionin Sect.~V.
302: To avoid this particularity at the half-filling,
303: we display in Fig.~2(b) also the case where
304: the number of electrons ($N_{e}=190$) is below
305: half-filling.
306: %
307: For large ratios ($t_2/t_1 > 0.8$) we observe in Fig.~2(b)
308: a cylinder-like regime where there is typically $I_1$ dominance.
309: This is because the strong transverse hopping changes the periodicity
310: of the M\"obius strip to that of the conventional cylinder.
311: %
312: The somewhat unexpected observation is that there is a distinct wide
313: intermediate regime ($0.1 < t_2/t_1 < 0.8$) where $I_1$, $I_2$,
314: $I_3$ and $I_4$ are all comparable. This is the regime which is of
315: experimental relevance.
316: %
317: The expected effect of averaging in this regime is to emphasize
318: both the $I_2$ and the $I_4$ contributions.
319:
320:
321:
322:
323:
324:
325: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
326: \section{Statistical Analysis}
327: \label{sec4}
328:
329: The problem arising in the analysis of persistent currents in
330: disordered M\"obius strips is how to characterize the statistics
331: of the calculated data. It was already pointed out that essential
332: properties of observables result from the averaging procedure and
333: the nature of the underlying statistical ensemble
334: \cite{Gefen,Gilles,AGI}. On the other hand, fabrication of a
335: M\"obius strip requires an outstanding effort\cite{Tanda}, and
336: hence, anticipated measurements of the persistent current would
337: probably be performed on a single sample. Thus, somewhat
338: unfortunately, the important results reported therein and the
339: powerful calculation methods based on super-symmetry might be less
340: useful for {\em single-sample} experiments since there is no
341: averaging.
342:
343: What is then the most efficient way to present our calculated
344: results? The answer is provided by elementary statistics. An
345: experimental result consists of a set of $K$ measurements
346: $I(\phi_{i})$, $i=1,2,\dots,K$ performed on a given sample. This
347: sample is taken out of an ensemble of M\"obius strips with
348: different disorder realizations, electron numbers $N_{e}$, aspect
349: ratios, {\em etc}. The set $\{ I(\phi_{i}) | i=1,\dots,K \}$ can be
350: regarded as an instance of a random vector in a $K$ dimensional
351: space. Alternatively, this instance can be represented by the
352: current harmonics $(I_{1},I_{2},\dots)$ defined via
353: Eq.~(\ref{Iphi}). For our purpose it seems adequate to keep only
354: the first 4~harmonics. The relevant statistical ensemble is then a
355: set of \lq\lq points\rq\rq\ $(I_{1},I_{2},I_{3},I_{4})$ in
356: four-dimensional probability space, each point corresponds to a
357: possible experimental measurement of the current on the {\it
358: entire} $\phi$ interval. Let us denote the number of points within
359: an infinitesimal four-dimensional volume element by
360: $P(I_{1},I_{2},I_{3},I_{4})dI_{1}dI_{2}dI_{3}dI_{4}$. The
361: distribution function $P$ is normalized to ${\cal N}$, the total
362: number of members in the ensemble. The most probable (typical)
363: experimental result is then determined by the quadruple
364: $I_{1},I_{2},I_{3},I_{4}$ at which $P$ is maximal. Another
365: quantity, which seems more informative and easy to analyze, is the
366: distribution
367: %
368: \begin{equation}
369: p_{n}(I_{n})=\int_{0}^{\frac {1} {2} |I_{n}|}
370: P(I_{1},I_{2},I_{3},I_{4}) \prod_{m \ne n} d|I_{m}|.
371: \label{pn}
372: \end{equation}
373: %
374: This corresponds to the possibility of finding a sample whose
375: current $I(\phi)$ is approximately described by $I(\phi) \approx
376: I_n \sin(2\pi n\phi)$. (For a sample counted by $p_{n}(I_{n})$,
377: all the harmonics other than $I_{n}$ are at most half of $I_{n}$
378: in magnitude). The number of members in the ensemble that exhibit
379: $I_n$ dominance is therefore ${\cal N}_{n}=\int_{0}^{\infty}
380: p_{n}(I_{n})d|I_{n}|$. If ${\cal N}_{n} > {\cal N}_{m}$ for any $m
381: \ne n$, the typical periodicity of $I(\phi)$ is dominantly $1/n$.
382: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
383: \begin{figure}[tttt]
384: \includegraphics[width=\hsize]{fig3.eps}
385: \caption{The distributions $p_{n}(I_{n})$ ($n=1,2,3,4$) defined by
386: Eq.~(\ref{pn}) for the cylinder and M\"obius ensembles. The
387: numbers of members with $I_n$ dominance are ${\cal N}_{1}=15,829$,
388: ${\cal N}_{2}=382$, ${\cal N}_{3}=4$, and ${\cal N}_{4}=439$ for
389: the cylinder ensemble, and ${\cal N}_{1}=1,562$, ${\cal
390: N}_{2}=336$, ${\cal N}_{3}=384$, and ${\cal N}_{4}=1,992$ for the
391: M\"obius ensemble.}
392: \label{fig3}
393: \end{figure}
394: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
395: In actual calculations, we assume that the lattice structure, the
396: aspect ratio, and the strength of disorder are fixed, and that the
397: temperature is very low. Then, two quantities are still
398: fluctuating, namely, the filling factor (or the electron number
399: $N_{e}$) and the specific realization of disorder. We generate an
400: ensemble of ${\cal N}={\cal N}^{a}{\cal N}^{b}$ members
401: corresponding to ${\cal N}^{a}$ consecutive values of $N_{e}$,
402: usually around half filling, and ${\cal N}^{b}$ realizations of
403: disorder for each one of them. Actually, for our systems of size
404: $N=20$, $M=10$ with $t_{1}=1$, $t_{2}=0.5$, and $W=0.5$, we take
405: $150 \le N_{e} \le 250$, hence ${\cal N}^{a}=101$ and ${\cal
406: N}^{b}=250$, so that ${\cal N}=25250$. The distributions
407: $p_{n}(I_{n})$ for the cylinder and M\"obius ensembles are shown
408: in Fig.~3.
409:
410: \section{Main Observations}
411: \label{sec5}
412:
413: The most striking result that can be deduced from Fig.~3 is the
414: essential reduction of ${\cal N}_{1}$ for the M\"obius ensemble
415: compared with the cylinder one. For the present ratio $N/2M=1$,
416: there is also a strong tendency towards $\Phi_{0}/4$ periodicity,
417: since ${\cal N}_{4} > {\cal N}_{m \ne 4}$ for the M\"obius
418: ensemble. This result is intriguing, because here we have no
419: averaging procedure which is crucial to get the $1/2$ periodicity
420: in cylindrical strips. However, this $1/4$ periodicity emerges
421: only for the specific ratio $N/2M=1$. We have calculated the
422: distributions $p_n(I_n)$ for M\"obius strips with several aspect
423: ratios. The value of ${\cal N}_{n}$ depends on the aspect ratio.
424: No specific $n$ gives prominent ${\cal N}_{n}$ independently of
425: the aspect ratio. On the other hand, the collapse of $I_{1}$
426: dominance in the M\"obius ensemble is robust and persists in
427: systems with different ratios $N/2M$ as well. We can safely say
428: that ${\cal N}_{1}$, ${\cal N}_{2}$, ${\cal N}_{3}$, and ${\cal
429: N}_{4}$ become all comparable in the M\"obius ensemble.
430:
431: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
432: \begin{figure}
433: \includegraphics[width=8.5cm]{fig4.eps}
434: \caption{$I_{1}$ and $I_{2}$ as a function of $N_{e}$ for the
435: ordered (solid line) and disordered (dots) systems. Parameters
436: describing the systems are the same with those for Fig.~3.}
437: \label{fig4}
438: \end{figure}
439: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
440: The natural question that comes to mind is whether this result is
441: a consequence of the M\"obius geometry or, rather, it is due to
442: the presence of disorder. In order to answer this question, we
443: have performed the calculation of $P^0(I_1,I_2,I_3,I_4)$ for a
444: \lq\lq clean\rq\rq\ M\"obius ensemble (without disorder, only
445: $N_e$ is being changed). We found out that the probability to find
446: any $I_n$ dominance is extremely small. The immediate conclusion
447: is that disorder is essential for the identification of M\"obius
448: strips via $I_{n>1}$ dominance. Does this mean that interference
449: or weak-localization effects due to the presence of disorder is
450: important? To clarify this point, we should understand how
451: $P^0(I_1,I_2,I_3,I_4)$ is modified by disorder. The distribution
452: $P^0(I_1,I_2,I_3,I_4)$ is, in fact, a function defined on a one
453: dimensional curve $[I_1(N_e),I_2(N_e),I_3(N_e),I_4(N_e)]$ in
454: $(I_1,I_2,I_3,I_4)$ space. For this reason, it is unlikely to find
455: a sample where one of the $I_{n}$ is dominant. The effect of
456: disorder is to give some \lq\lq thickness\rq\rq\ to this curve
457: (see Fig.~4). Taking into account that the amplitudes of
458: $I_n(N_e)$ for M\"obius strips are all comparable, the thickness
459: gives a finite probability to find samples where one of the $I_n$
460: is dominant. On the contrary, in the case of cylindrical strips,
461: the amplitude of $I_{1}(N_{e})$ is overwhelmingly larger than
462: those of $I_{n\ne 1}(N_{e})$, which makes it unlikely to find
463: $I_{n\ne 1}(N_{e})$ dominated samples even if we take the
464: statistical effect of disorder into account. We should note here
465: that the function $I_{n}$ with odd $n$ for the clean M\"obius
466: strip is an even function around the half-filling ($N_{e}=200$)
467: and an odd function for odd $n$, while the function $I_{n}$ for
468: arbitrary $n$ is an even function in the cylinder case.
469:
470:
471:
472: %
473: %
474: %
475: %
476: Our findings regarding ${\cal N}_{n}$ for the M\"obius ensemble
477: are based on the fact that the amplitudes of $I_{n}(N_{e})$
478: are all comparable for M\"obius strips. As we have observed
479: in Fig.~2, this is a robust statistical property in the
480: intermediate regime $0.1<t_{2}/t_{1}<0.8$.
481: The choice $t_{2}/t_{1}=0.5$ above, provides typical
482: results for $p_{n}(I_{n})$ and ${\cal N}_{n}$ in case
483: that $t_{2}/t_{1}$ is within this distinct regime.
484:
485:
486:
487:
488:
489:
490:
491:
492: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
493: \section{Conclusions}
494: \label{sec6}
495:
496: We have studied the persistent currents of non-interacting
497: electrons in M\"obius strips. The spectral properties for a clean
498: system were found analytically, and the effect of disorder on the
499: currents was analyzed numerically. We have found that disorder is
500: quite essential for the identification of M\"obius strips. The
501: issue of disorder averaging is not relevant for single sample
502: experiments, and hence, special care is required for statistical
503: analysis of the current harmonics. The fingerprint of the M\"obius
504: geometry is an enhanced probability to find samples in which
505: $I_n$, with $n>1$ dominates. This should be contrasted with the
506: case of cylinder geometry, where there is a clear $I_{1}$
507: dominance. The above assertion regarding the fingerprint of the
508: M\"obius geometry is correct provided the effect of disorder is
509: properly taken into account.
510:
511: We would like to thank T. Nakayama for very helpful discussions.
512: One of the authors (Y.A.) was supported by the Invitation
513: Fellowship for Research in Japan (Short Term) of the Japan Society
514: for the Promotion of Science. Numerical calculations in this work
515: have been mainly performed on the facilities of the Supercomputer
516: Center, Institute for Solid State Physics, University of Tokyo.
517:
518:
519: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
520:
521: \begin{thebibliography}{99}
522:
523: \bibitem{Imry}
524: M. B\"uttiker, Y. Imry and R. Landauer,
525: Phys. Lett. {\bf 96A}, 365 (1983).
526:
527: \bibitem{AASS}
528: B. L. Altshuler, A. G. Aronov, B. Z. Spivak,
529: D. Yu. Sharvin and Yu. V. Sharvin,
530: Pis'ma Zh. Eksp. Teor. Fiz. {\bf 35}, 484 (1982)
531: [JETP Lett. {\bf 35}, 589 (1982)].
532:
533: \bibitem{Levy}
534: L. L\'evy, G. Dolan, J. Dunsmir and H. Bouchiat,
535: Phys. Rev. Lett. {\bf 64}, 2074 (1990).
536:
537: \bibitem{Gefen}
538: H. F. Cheung, E. K. Riedel and Y. Gefen,
539: Phys. Rev. Lett. {\bf 62}, 587 (1989).
540:
541: \bibitem{Gilles}
542: H\'el\`ene Bouchiat and Gilles Montambaux,
543: J. Phys. France {\bf 50}, 2695 (1989).
544:
545: \bibitem{AGI}
546: B. L. Altshuler, Y. Gefen and Y. Imry,
547: Phys. Rev. Lett. {\bf 66}, 88 (1991).
548:
549: \bibitem{Tanda}
550: S.~Tanda, T.~Tsuneta, Y.~Okajima, K.~Inagaki, K.~Yamaya, and N.~Hatakenaka,
551: Nature {\bf 417}, 397 (2002).
552:
553: \bibitem{Webb}
554: %S. Washburn and R. A. Webb, Adv. Phys. {\bf 35}, 395 (1986).
555: V. Chandrasekhar, R. A. Webb, M. J. Brady, M. B. Ketchen, W. J. Galager,
556: and A. Kleinsasser, Phys. Rev. Lett. {\bf 67}, 3578 (1991).
557:
558: \bibitem{Hayashi}
559: It has been theoretically predicted that a novel state appears
560: in a superconducting M\"obius strip under a magnetic field:
561: M.~Hayashi and H.~Ebisawa, J. Phys. Soc. Jpn. {\bf 70}, 3495 (2001).
562:
563: \bibitem{Mila}
564: Fredric Mila, C. A. Stafford and Sylvian Capponi,
565: Phys. Rev. {\bf B57}, 1457 (1998).
566:
567: \bibitem{rmrk}
568: If, instead of Eq.~(2), we assume the periodic B.C.~in
569: the $y$ direction, the M\"obius B.C.~of Eq.~(3)
570: are modified as $\psi(x+L_x,y)=\psi(x,y+L_y)$. This leads to a
571: long-range hopping term in Eq.~(5), that does not
572: \lq\lq discriminate\rq\rq\ the $M$th wire. Yet another possibility
573: is to {\em keep} Eq.~(3). This possibility is of {\em academic}
574: interest, since it corresponds to an geometry that cannot be embedded
575: in a 3D Euclidean space. In the latter case, there is a symmetry
576: breaking that prevents having a stationary current in the $y$
577: direction.
578: \end{thebibliography}
579: \end{document}
580: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
581: