cond-mat0502034/def.tex
1: 
2: \documentstyle[aps,epsfig]{revtex}
3: \textwidth160mm
4: \textheight205mm
5: \oddsidemargin0mm
6: \topmargin0mm
7: \def\be{\begin{equation}}
8: \def\ee{\end{equation}}
9: \begin{document}
10:  \title{Entanglement entropy with interface defects}
11:  \author{
Ingo Peschel\\
12: {\small Fachbereich Physik, Freie Universit\"at Berlin,} \\
13: {\small Arnimallee 14, D-14195 Berlin, Germany}}
14:  \maketitle
15:  \begin{abstract}
16:  We consider a section of a half-filled chain of free electrons and its
17:  entanglement with the rest of the system in the presence of one or 
18:  two interface defects. We find a logarithmic behaviour of the 
19:  entanglement entropy with constants depending continuously on 
20:  the defect strength.
21: 
22:  
23:  
24: \end{abstract}
25: 
26: \vspace{1cm}
27: 
28: The entanglement of different parts of a quantum system in the total
29: wave function can be viewed as the result of a coupling across their
30: common interface. Correspondingly, the entanglement entropy turns
31: out to be proportional to the interface area in simple models 
32: \cite{Bom86,Sred93,Plenio04}.
33: Modifications of the interface will therefore modify the entanglement
34: and in one-dimensional systems with short-range interactions, where the 
35: interface reduces to a point, a single defect is expected to have a marked
36: influence. This was recently pointed out by Levine \cite{Levine04} who
37: applied this idea to the case of a Luttinger liquid with one impurity. He
38: obtained within 
39: perturbation theory a reduction of the entanglement entropy $S$ which is 
40: strong for repulsive and weak for attractive interactions, if one considers 
41: a large subsystem. Roughly speaking, this corresponds to the known influence
42: of interactions on the impurity strength in this case \cite{Kane/Fisher92}.
43: Levine's approach using bosonization is interesting, but it also suggests 
44: to look directly at the simplest case and to investigate the problem for 
45: non-interacting electrons with specific defects.\\
46: 
47: 
48: In the present paper we therefore study free electrons hopping on a chain
49: for the case of half filling. In spin language, this corresponds to an XX model.
50: This is a critical system and the entanglement entropy between a subsystem
51: of length $L$ and the rest is given by the conformal result 
52: \cite{Holzhey94,Calabrese04,Vidal03,Latorre03,Jin04}
53: \be 
54:   S= \frac{c} {3} \ln L + k
55:   \label{eqn:entro1}
56: \ee
57: with the value $c=1$ for the central charge. How does this change in the
58: presence of a defect at the chosen interface ? For one case, the answer is
59: known : if the defect is such that it cuts the chain, the subsystem has a free
60: end and thus only one connection to the rest remains. This changes $c \rightarrow
61: c_{eff}=c/2$ and $k \rightarrow k'$, but the logarithm is unaffected \cite{Calabrese04}.
62: In the following we show that the same holds for arbitrary defects, either at
63: one or at both interfaces. These defects can be changed bonds or changed
64: site energies. The calculations are numerical and based on the determination
65: of the reduced density matrix $\rho$ from the one-particle correlation function
66: of the total chain \cite{Peschel03,Vidal03,Cheong/Henley02}. The necessary 
67: formulae are given in Section I. Eigenvalue spectra of $\rho$ for various defects, 
68: which differ in a characteristic way from the spectrum of the homogeneous system, 
69: are presented in Section II. From the eigenvalues, the entanglement entropy 
70: $S= -tr(\rho \ln \rho)$ is obtained and discussed in Section III. We determine the 
71: effective central charge $c_{eff}$ for bond and site defects and discuss its 
72: behaviour in terms of a simple model. In contrast to the constant $k$, it is always 
73: reduced by a defect and therefore the entanglement always becomes smaller if 
74: the size $L$ of the subsystem is large enough. A brief summary is given in Section IV.
75: 
76: 
77: \section{Basic formulae}
78: 
79: 
80: We consider  a system of free fermions hopping between neighbouring 
81: sites of an infinite linear chain. The corresponding Hamiltonian reads \\
82:  \be
83:   \hat{H}=- \sum_{n}  t_n (c_n^{\dagger} c_{n+1} + c_{n+1}^{\dagger} c_{n})
84:   + \sum_{n} \Delta_n c_n^{\dagger}c_{n} 
85:   \label{eqn:hop}
86:  \ee
87: where 
$t_n$ is the hopping matrix element and the 'hat' denotes quantities of the 
88: total system. In the following, we set $t_n =1 $ and $\Delta_n = 0$ except at the boundaries 
89: of the subsystem which consists of the sites $i = 1, 2, ...L$. We will mainly consider the case
90: of one bond defect at the left boundary, $t_0 = t$, but we will also give results for
91: two equal bond defects at the two boundaries, $t_0 = t_L= t$, 
92: and one or two site defects next to the boundary,
93: $\Delta_1 = \Delta$ or $\Delta_1 = \Delta_L = \Delta$.\\
94: 
95: The total system is assumed to be half filled and in its ground state $|0>$. The reduced density
96: matrix then has the form \cite{Peschel03,Peschel04}
97: \be
98:    \rho= {\cal K} \exp{(-\sum_{i,j} H_{ij} c_i^{\dagger} c_j )}
99:    \label{eqn:rho1}
100:  \ee
101:  where $\cal K$ is a normalization constant and
102:  the matrix $H_{ij}$ follows from the one-particle correlation function
103:  of the total system
104:  \be
105:   \hat{C}_{mn} = <0|\; c_m^{\dagger} c_n\; |0>
106:   \label{eqn:cf}
107:  \ee
108:  via the relation
109:  \be
110:     H = \ln{\,[(1-C)/C\,]}
111:   \label{eqn:HC}
112:  \ee
113:  Here $C$ denotes the $L \times L$ submatrix of $\hat{C}$ with 
114:  the sites restricted to the subsystem. For a homogeneous infinite system,
115:  $C$ is given by
116:  \be
117:   C_{ij} = C^{0}(i-j) =  \frac{\sin \bigl[\frac{\pi} {2}\,(i-j)\bigr]} {{\pi}\,(i-j)} 
118:   \label{eqn:cf1}
119:  \ee
120:  With defects, the translational invariance is lost and for single defects  
121:  $C$ has the general form
122:  \be
123:   C_{ij} = C^{0}(i-j) -  C^{1}(i+j)
124:   \label{eqn:cf2}
125:  \ee
126:  Thus $C_{ij}$ is the difference of a Toeplitz matrix depending on $(i-j)$
127:  and a Hankel matrix depending on $(i+j)$.
128:  Physically, the term $C^{1}$ leads to oscillatory behaviour of the correlations
129:  as one moves along the chain. For site defects, these are the Ruderman-
130:  Kittel oscillations if one looks at the density $C_{ii}$. For bond defects, the
131:  density is unaffected but the oscillations appear e.g. in the nearest-neighbour
132:  (bond) correlation function $C_{i,i+1}$. The quantity  $C^{1}$ can be obtained
133:  by using the scattering phase shifts and including the contribution of possible 
134:  localized states caused by the defect. Thus a single weak bond 
135:  $t= e^{\:\nu} \leq 1$ in an infinite total system leads to
136:  \be
137:       C^{1}(l) = - \frac {1} {2} \sinh \nu \; (e^{-\nu} I_l - e^{\nu} I_{l-2})
138:    \label{eqn:cf3}
139:   \ee
140:   where
141:  \be
142:    I_l = \int_{0}^{\pi /2} \frac {dq} {\pi} \frac {\cos(ql)} {\sinh^2 \nu + \sin^2 q}
143:    \label{eqn:int}
144:   \ee
145:  For $t=0$, this simplifies to $C^{1}(l)= C^{0}(l)$ and $C$ assumes the known form 
146:  for a system with an open end. For a strong bond, $t > 1$, one has to add the
147:  contribution
148:  \be 
149:   C^{1}(l) =- \sinh \nu \; \exp{(-\nu (l-1))}
150:   \label{eqn:boundstate}
151:  \ee
152:  from the occupied bound state at energy $-2\cosh \nu$ below the band.
153:  Similarly, a site defect $\Delta = 2 \sinh \nu >  0$ leads to
154:  \be
155:  C^{1}(l) =  \frac {1} {2} \sinh \nu \; ( I_{l-1} - I_{l-3}  +  2  \sinh \nu \; I_{l-2})
156:  \label{eqn:cf4}
157:   \ee
158:  By diagonalizing $C$ one obtains its eigenvalues $\zeta_k$  
159:  ($0 <\zeta_k <1$),  from which the eigenvalues $\varepsilon_k = \ln ((1-\zeta_k)/\zeta_k)$ 
160:  of $H$ ($-\infty < \varepsilon_k < \infty$) follow. The reduced density 
161:  matrix then takes the diagonal form
162:  \be
163:    \rho = {\cal K} \exp{(-\sum_{k} \varepsilon_ k c_k^{\dagger} c_k)}
164:    \label{eqn:rho2}
165:   \ee
166:   The entanglement entropy is defined by $S= -tr(\rho \ln \rho)$ (we are using the
167:   natural logarithm, not the one to basis 2) and reads
168:   in terms of the $\zeta_k$
169:   \be
170:    S = -\sum_{k} \bigl[\,\zeta_k \ln \zeta_k + (1-\zeta_k) \ln(1-\zeta_k) \bigr]
171:    \label{eqn:entro2a}
172:   \ee
173:   and in terms of the $\varepsilon_k$
174:  \be
175:   S= \sum_{k} \left [ \, \ln{[1+exp{(-\varepsilon_k)}]} +  \frac {\varepsilon_k}
176:         {\exp{(\varepsilon_k)}+1} \right ]
177:   \label{eqn:entro2b}
178:  \ee
179:     
180:  Only the $\zeta_k$ which are not too close to 0 or 1 resp. the $\varepsilon_k$
181:  which are not much larger than 1 in magnitude give a sizeable contribution to the
182:  entropy.
183:  
184: \section{Density-matrix spectra}
185: 
186: We have calculated the correlation matrix $C$ for the case of single defects  
187: by evaluating the integrals for $C^{1}$ numerically. From this the low-lying
188: single-particle eigenvalues $\varepsilon_k$ with  $ |\varepsilon_k| \leq 25$
189: were obtained. Higher values cannot be reached with standard double precision
190: routines because the corresponding $\zeta_k$ lie too close to 0 or 1. However, as
191: noted above, they are unimportant for $S$. In the following
192: we always consider even $L$.\\
193: 
194: 
195: Spectra for one bond defect are shown in Fig. 1 for a subsystem of $L=50$ sites.
196: The eigenvalues come in pairs ($\varepsilon,-\varepsilon$) due to the particle-hole
197: symmetry of the problem.
198: For a homogeneous system, $t=1$, one finds the slightly bent curve known
199: from previous investigations \cite{Chung/Peschel01,Cheong/Henley03}.\\
200: 
201: For a weak defect, one can see a shift of the dispersion curve which is upward
202: for positive $\varepsilon_k$ and downward for negative ones. This shift
203: becomes stronger as the bond becomes weaker. In addition, oscillations appear which
204: also increase for weaker bonds. Most importantly, however, more and more of the low
205: $\varepsilon_k$ follow a steeper dispersion curve as $t$ approaches zero. This
206: curve has a slope approximately twice as large as for the homogeneous system and no wiggles 
207: and represents the result one obtains directly for the system with a free end.
208: Thus the spectrum with the defect is a mixture of those for the two cases $t=1$ and $t = 0$.\\
209: 
210: For a strong defect, the situation is basically the same. However, here the two lowest eigenvalues 
211: $\varepsilon_0$ and $\varepsilon_1$ play a special role because they approach zero as 
212: $t\rightarrow \infty$ . This moves the left and right part of the spectra two units apart. 
213: This feature can be understood from the correlations in this limit. The two
214: 
215: \pagebreak
216:  
217: \begin{figure}
218: \centerline{\psfig{file=./defectfig/efig1.eps,width=7cm,angle=-90}
219: \psfig{file=./defectfig/efig4.eps,width=7cm,angle=-90}}
220: \vspace{1mm}
221: \caption{Low-lying single-particle eigenvalues for a subsystem of L=50
222: sites with $\it{one}$ boundary defect.\\
223:  Left : Weak defects, $t=1; 10^{-1}; 10^{-2}; 10^{-3}; 10^{-4}$ 
224:  (from bottom to top in the right part of the figure). \\
225:  Right : Strong defects, $t=10; 10^2; 10^3; 10^4$ (also from bottom to top).
226:  The lines are guides for the eye. \\
227: The numbering is such that positive $\varepsilon_k$ 
228: have positive values of $k$. }
229: \label{fig1.eps}
230: \end{figure}
231: 
232: 
233: localized states 
234: below and above the band then exhaust the local Hilbert space at the two bond sites and 
235: effectively cut the system. Thus $C_{1j} = 0$ for $j > 1$ and the diagonal term $C_{11} = 1/2$ gives 
236: one eigenvalue zero. The other one appears because the remaining part of the subsystem has
237: an odd number ($L-1$) of sites. It is absent, if the full subsystem has $L+1$ sites. Then the spectrum 
238: for $t \gg 1$ is exactly the same as for $L$ sites and $1/t \ll 1$ up to the remaining zero eigenvalue.
239: 
240: 
241: \begin{figure}
242: \centerline{\psfig{file=./defectfig/efig2.eps,width=7cm,angle=-90}
243: \psfig{file=./defectfig/efig3.eps,width=7cm,angle=-90}}
244: \vspace{1mm}
245: \caption{Low-lying single-particle eigenvalues for a subsystem of $L=50$
246: sites with $\it{two}$ boundary defects. \\Left : weak bond defects. Right : strong bond defects.
247: The parameters are the same as in Fig. 1.}
248: \label{fig2.eps}
249: \end{figure}
250: 
251: 
252: 
253: 
254: Spectra for two bond defects are shown in Fig. 2. These were calculated by finding
255: the eigenfunctions of $\hat{H}$ for a total system of size $N=1000$ numerically. This
256: introduces small finite-size effects but these are not visible on the scale of the figure.
257: There 
258: is a shift of the curves as for a single defect, but now it affects $\it{all}$ eigenvalues
259: if the defects are weak. Thus a gap is opened in the spectrum. For $t \rightarrow 0$  
260: all $\varepsilon_k$ diverge. This is reflects the fact that the 
261: subsystem is, in this limit, completely decoupled from the rest and the ground state $|0>$ 
262: becomes a product state. Then $\rho$ has one eigenvalue 1 if all single-particle
263: levels are unoccupied and all other eigenvalues vanish. If the defects are strong,
264: two $\varepsilon_k$ go to zero while the remaining eigenvalues again diverge. The
265: zero eigenvalues are related to the localized states at the two bonds which belong
266: to the subsystem as well as to the environment. The ground state $|0>$ then has four 
267: components which differ by the location of the electrons on the two bonds and the four
268: resulting eigenvalues 1/4 of $\rho$ are produced by the two vanishing $\varepsilon_k$.\\
269: 
270: For comparison, the spectra for two site defects are shown in Fig. 2. These defects 
271: destroy the particle-hole symmetry of the problem and therefore the spectrum of the 
272: $\varepsilon_k$ is in general no longer symmetric with respect to zero. Apart from that,
273: it shows the same features as for two bond defects. In particular, also the site defects
274: cut off the subsystem if the site energy $\Delta$ becomes large and thus all 
275: $\varepsilon_k$ except two diverge for $\Delta\rightarrow \infty$. In this limit, the reflection
276: symmetry of the spectrum is also restored.
277: 
278: \begin{figure}
279: \centerline{\psfig{file=./defectfig/efig6.eps,width=8cm,angle=-90}}
280: \vspace{1mm}
281: \caption{Low-lying single-particle eigenvalues for a subsystem of $L=50$
282: sites with two $\it{site}$ defects at the boundaries. The site energies are 
283: $\Delta= 0; 2; 5; 10; 100$ from bottom to top in the right part of the figure.}
284: \label{fig3.eps}
285: \end{figure}
 
286:  
287:  \section{Entanglement entropy}
288:  
289:  Using the values of the $\varepsilon_k$ or the $\zeta_k$, one can obtain the
290:  the entanglement entropy from (\ref{eqn:entro2a}),(\ref{eqn:entro2b}). Results 
291:  for single bond defects are shown in Fig. 4 for system sizes between 
292:  $L=20$ and $L=100$.
293: 
294: 
295:   The plot on the right hand side shows that $S$ varies logarithmically with
296:   $L$ in all cases. There is no indication of a $\ln^2 L$-term as found in 
297:   \cite{Levine04}. There is only a small variation of the slope with $L$ and
298:  its asymptotic value can be determined very well by an extrapolation in $1/L$.
299:  It is remarkable that one can see this logarithmic behaviour so well, since for 
300:  all sizes considered here only about 20 single-particle eigenvalues contribute 
301:  if the accuracy is set to $10^{-10}$. Thus one is far away from a limit in which 
302:  the eigenvalue spectrum becomes dense and where the logarithmic dependence 
303:  could result from the density of states.
304: 
305:   
306:  \begin{figure}
307: \centerline{\psfig{file=./defectfig/efig7.eps,width=6.5cm,angle=-90}
308:  \psfig{file=./defectfig/efig7a.eps,width=6.5cm,angle=-90}}
309: \vspace{1mm}
310: \caption{Entanglement entropy as a function of $L$ for one bond defect. 
311: From bottom to top: t=0.1; 0.5; 1.0; 2.0}
312: \label{fig4.eps}
313: \end{figure}
314: 
315:  
316:  Thus we find that the entanglement entropy has the same form (\ref{eqn:entro1})
317:  as for the homogeneous system, but with an effective value $c_{eff}$ which 
318:  depends on the strength of the defect. The same holds for two bond defects 
319:  and also for site defects (see below). We first discuss the case of one bond 
320:  defect further.\\
321:  
322:  
323:   The asymptotic value of $c_{eff}$, determined by extrapolating the data from
324:  $L=20-100$, is shown in Fig. 5 as a function of the defect strength. 
325:  
326:  \begin{figure}
327: \centerline{\psfig{file=./defectfig/efig8.eps,width=7.0cm,angle=-90}}
328: \vspace{1mm}
329: \caption{Effective central charge as a function of the bond strength for one bond defect.
330:  Full line : numerical result, dashed line : approximation Eqn. (\ref{eqn:cfit}).}
331: \label{fig5.eps}
332: \end{figure}
333: 
334:  One sees that
335:  it varies smoothly between the limits $c_{eff}=1/2$  for a subsystem with a free end
336:  and  $c_{eff}=1$ for a homogeneous system. The figure shows $c_{eff}$ only for $t < 1$
337:  because one finds that $c_{eff}$ is symmetric under $t \rightarrow 1/t$.
338:  
339:  
340:  One can try to find a simple analytic function which fits the numerical curve in Fig. 5.
341:  Since the defect is characterized by the scattering phases, one could expect that
342:  $c_{eff}$ depends on these. Because one is dealing with an asymptotic property, the
343:  phase shifts at the Fermi surface should enter. These are given by
344:  
345:   \be
346:    \delta_{\pm} = \pm \,\bigl[\,2 \,\hbox{arctg}\,(t) - \frac {\pi} {2}\, \bigr]
347:    \label{eqn:phases}
348:   \ee
349:   
350:  for functions symmetric and antisymmetric about the defect, respectively. One is then
351:  lead to the expression
352:  
353:   \be
354:    c_{eff} = \frac {1} {2} + \frac {1} {2}\,
355:    \bigl[\, \frac {4} {\pi} \,\hbox{arctg}(t)\; \frac {4} {\pi} \,\hbox{arctg}(1/t)\,\bigr]^2
356:    \label{eqn:cfit}
357:   \ee
358:   
359:  The expression in the brackets also appears (as an exponent) if one calculates the time 
360:  autocorrelation function of a spin in an XX spin chain with a bond defect 
361:  \cite{Peschel/Schotte84}. The function 
362:  (\ref{eqn:cfit}) has the correct limits for $t=0$ and $t=1$ and is symmetric under
363:  $t \rightarrow 1/t$. As can be seen from the figure, it approximates the numerical data 
364:  quite well. The deviations are less than $10^{-2}$, but the agreement is not perfect.
365:  Especially for $t \rightarrow 0$, the data do not seem to vary quadratically in $t$ but
366:  with a power 1.8. Replacing the power 2 in (\ref{eqn:cfit}) by this value improves the
367:  fit over the whole intervall substantially. The deviations become less than 
368:  $10^{-3}$ and are almost invisible on the scale of the figure.\\
369:  
370:  The two terms in (\ref{eqn:cfit}) can be viewed as the contributions of the unperturbed
371:  and the perturbed interface, respectively.
372:  One can understand this result from the form of the single-particle spectrum. In the
373:  previous section it was found that it consists of two pieces.  If one assumes that in the 
374:  limit $L \rightarrow \infty$  the levels in these two branches become equidistant and
375:  dense such that (for positive $\varepsilon_k$)
376:  \be
377:    \varepsilon_k= \left \{ \begin {array} {c} 2k\;a/\ln L ,\;\;\; 
378:                              \varepsilon_k < \alpha  \; \;\;  \\[0.3cm]
379:           k\; a/\ln L , \;\;\;  \varepsilon_k > \alpha \;\;\;  \; \;\; \end{array} \right.
380:   \label{eqn:spec1}
381:  \ee
382:  where $\alpha = \alpha(t)$ depends on $t$ such that $\alpha = 0$ for $t=1$ and 
383:  $\alpha = \infty$ for $t=0$ and $a$ is a constant, the entanglement entropy
384:  can be written as
385:  \be
386:   S =  2 \ln L \, \left [ \,\frac {1} {2a} \int_{0}^{\alpha} d\varepsilon \; s(\varepsilon) +
387:                         \frac {1} {a}  \int_{\alpha}^{\infty} d\varepsilon \;s(\varepsilon) \right ]
388:    \label{eqn:entro2}
389:   \ee
390:   
391:   Here $s(\varepsilon)$ is given by the bracket in (\ref{eqn:entro2b}) 
392:   and the factor of 2 in front takes care of the
393:   negative levels. Denoting the second integral by $G(\alpha)$, one obtains
394:   \be
395:     S = 2 \ln L \; \bigl[\, \frac {1} {2} (G(0) - G(\alpha)) + G(\alpha) \,\bigr]
396:         =  \ln L \; \bigl[\,G(0) + G(\alpha)\,\bigr]
397:    \label{eqn:entro3}
398:    \ee
399:   Thus each interface brings its own amplitude for the logarithm : the unperturbed
400:   one $G(0)$ and the perturbed one $G(\alpha)$.
401:   For the homogeneous system the bracket must have the value 1/3, from which
402:   $G(0)=1/6$ follows. Therefore (\ref{eqn:entro3}) gives for $c_{eff}$
403:   \be
404:     c_{eff} = \frac {1} {2} + 3\; G(\alpha)
405:    \label{eqn:ceff1}
406:   \ee
407:   If the exact dependence of $\alpha$ on $t$ were known, one could evaluate the
408:   integral $G(\alpha)$ and determine $c_{eff}$ in this way. The oscillations of the
409:   $\varepsilon_k$ which are neglected in (\ref{eqn:spec1}) could also be included.\\
410:   
411:   The discussion so far has dealt with one defect. However, one can apply it also to
412:   the case of two defects. The results of Section III show that then the lowest
413:   eigenvalues are just missing and one also finds that the gap in the spectrum for
414:   two defects  
415:   corresponds roughly to the bending point in the spectrum for one defect. 
416:   Therefore one has only the second region in (\ref{eqn:spec1}) and finds instead of
417:   (\ref{eqn:entro3})
418:   \be
419:     S =   \ln L \; [2 \;G(\alpha)]
420:    \label{eqn:entro4}
421:    \ee
422:   from which the value of $c_{eff}$ can be read off. This also gives the relation
423:   \be 
424:     c^{(2)}_{eff} = 2c^{(1)}_{eff} - 1
425:     \label{eqn:ceff2}
426:   \ee
427:   between the effective c-values for two defects and for one defect. Due to the
428:   lower accuracy of the calculations for two defects, this relation could not be
429:   checked precisely, but the deviations are less than some percent and vanish near
430:   $t=0$ and $t=1$.\\
431:   
432:  Results for the case of one site defect are shown in Fig. 6. Here the entanglement
433:  entropy is symmetric under $\Delta \rightarrow -\Delta$ and only positive values
434:  of $\Delta$ have to be considered. One sees that $c_{eff}$ decreases again from
435:  the value 1 to 1/2 as the perturbation becomes stronger, because
436:  the defect cuts the chain in the limit $\Delta \rightarrow \infty$ and thus acts
437:  like a bond with $t=0$ or $t=\infty$. 
438:     
439:  \begin{figure}
440: \centerline{\psfig{file=./defectfig/efig10.eps,width=8cm,angle=-90}}
441: \vspace{1mm}
442: \caption{Effective central charge as a function of the site energy $\Delta$ for one site defect.
443:  The values were obtained by extrapolation in $1/L$}
444: \label{fig6.eps}
445: \end{figure}
  
446:     
447:   
448:   Finally, we present results for the constant $k$ in the entropy, see (\ref{eqn:entro1}).
449:   They were obtained by extrapolating $S(L)-(c_{eff}/3) \; ln L$ from the values $L=20-100$.
450:   As Fig. 7 shows, this quantity $k$ is not symmetric under $t \rightarrow 1/t$ for one bond
451:   defect. Rather it increases from about 0.5 for $t=0$ to about 1.2 for large
452:   values of $t$, with a small initial dip (which depends sensitively on $c_{eff}$).
453:    At $t=1$ we find a value $k=0.726$ which is in
454:   agreement with the findings in \cite{Latorre03} but differs slightly from the value
455:   obtained from the asymptotic analysis in \cite{Jin04}. The limiting value for 
456:   $t \rightarrow \infty$ is found to differ from the value at $t=0$ by $\ln 2$. If one considers 
457:   systems with an odd number of sites $L+1$ for $t \gg 1$, this also holds and follows in this
458:   case directly from the identity of the single-particle spectra up to one zero eigenvalue 
459:   (which contributes $\ln 2$) noted in Section III. 
460:   The relatively large value of $k$ can make the entropy with a defect larger
461:   than without one if the subsystem is small enough (compare Fig. 4). However, in the limit 
462:   $L \rightarrow \infty$ the effect of the reduced $c_{eff}$ always dominates and leads
463:   to a reduction of the entanglement.
464:   
465:  \begin{figure}
466: \centerline{\psfig{file=./defectfig/efig9.eps,width=8cm,angle=-90}}
467: \vspace{1mm}
468: \caption{Constant $k$ in the entanglement entropy as a function of the bond strength
469:  $t$ for one bond defect. The values were obtained by extrapolation in $1/L$}
470: \label{fig7.eps}
471: \end{figure}
472:   
473: \section{Conclusion}
474: 
475: We have investigated  a chain of free electrons in its ground state at half filling. 
476: The entanglement properties were calculated for the case of one or
477: two defects located at the contact points between the subsystem and the rest of the
478: chain. The density-matrix spectra were seen to differ in a characteristic way from
479: those for a homogeneous system. Nevertheless, the logarithmic dependence of 
480: the entanglement entropy on the size persists, only the amplitude changes 
481: and becomes dependent on the defect strength.
482: The mechanism could be described by a simple model based on the features of
483: the single-particle spectra. A simple approximate formula for the effective central
484: charge was also given. Although we focussed on bond defects, the same 
485: features are found for site defects. The logarithmic law means that the system remains 
486: critical, as expected for a localized perturbation. In the two-dimensional system
487: associated with the quantum chain, the defect becomes a line as noted and used
488: in \cite{Levine04}.
489: Such a two-dimensional system with a cut (because one considers the reduced density
490: matrix) and one or two straight defect lines perpendicular to this cut can be mapped 
491: conformally to a strip \cite{Peschel04} but the defect lines are bent by 
492: the mapping. For a single defect, the line runs only in the left or right half of the strip,
493: coming from one point on the boundary and going back to another one. 
494: For two defects, the two lines are attached to different edges and touch in 
495: the center of the strip. The more complicated density-matrix spectra found here must 
496: be related to this feature. It would be interesting to derive them analytically. Also
497: one wonders if the asymptotic formulae used in \cite{Jin04,Keating/Mezzadri04} to 
498: calculate $S$ can be generalized to the present case.\\
499: 
500: $\it{Acknowledgement}$ The author thanks K.D. Schotte for discussions and help with
501: the figures.
502:  
503:  
504:  
505: 
506:  
507: \begin{thebibliography}{99}
508: 
509: \bibitem{Bom86} L. Bombelli, R. K. Koul, J. Lee and R. D. Sorkin, 
510:  Phys. Rev. D {\bf 34}, 373 (1986)
511: \bibitem{Sred93} M. Srednicki
512:  Phys. Rev. Lett. {\bf 71}, 666 (1993)
513: \bibitem{Plenio04} M. B. Plenio, J. Eisert, J. Dreissig and M. Cramer,
514: preprint quant-phys/0405142 
515: \bibitem{Levine04} G. C. Levine,
516: Phys. Rev. Lett. {\bf 93} 266402 (2004)
517: \bibitem{Kane/Fisher92} C. L. Kane and M. P. Fisher,
518: Phys. Rev. B {\bf 46}, 15233 (1992)
519: \bibitem{Holzhey94} Ch. Holzhey, F. Larsen and F. Wilczek,
520:  Nucl. Phys. B {\bf 424}, 443 (1994)
521: \bibitem{Calabrese04} P. Calabrese and J. Cardy,
522:  J. of Statistical Mechanics, P06002 (2004)
523: \bibitem{Vidal03} G. Vidal, J. I. Latorre, E. Rico and A. Kitaev,
524:  Phys. Rev. Lett. {\bf 90}, 227902 (2003)
525: \bibitem{Latorre03}  J. I. Latorre, E. Rico and G. Vidal, 
526:  Quant. Inf. and Comp. {\bf 4}, 48 (2004), 
527:  preprint: quant-ph/0304098
528: \bibitem{Jin04} B.-Q. Jin and V. E. Korepin, 
529:  J. Stat. Phys. {\bf 116}, 79 (2004),
530: \bibitem{Peschel03} I. Peschel, J. Phys. A {\bf 36}, L205 (2003)
531: \bibitem{Cheong/Henley02} S.-A. Cheong and C. L. Henley,
532:  Phys. Rev. B {\bf 69}, 075111 (2004),
533:  preprint: cond-mat/0206196 
534: \bibitem{Cheong/Henley03} S.-A. Cheong and C. L. Henley,
535:  Phys. Rev. B {\bf 69}, 075112 (2004)
536:  \bibitem{Peschel04} I. Peschel,
537:  J. of Statistical Mechanics P06004 (2004)
538: \bibitem{Chung/Peschel01} M.-C. Chung and I. Peschel,
539:  Phys. Rev. B {\bf 64}, 064412 (2001)
 
540: \bibitem{Peschel/Schotte84} I. Peschel and K. D. Schotte,
541:  Z. Physik B {\bf 54} 305 (1984)
542: \bibitem{Keating/Mezzadri04} J. P. Keating and F. Mezzadri,
543: Preprint quant-ph/0407047
544: 
545: 
546: \end{thebibliography}
547: 
548: 
549: 
550: 
551: 
552: 
553: 
554: 
555: 
556: 
557: 
558: \end{document}
559: 
560: 
561: 
562: 
563: 
564: 
565: 
566: 
567: 
568: 
569: 
570: 
571: 
572: