1: \documentclass[aps,prl,twocolumn,showpacs,balancelastpage]{revtex4}
2: %\documentclass[aps,prl,preprint,showpacs]{revtex4}
3:
4: \usepackage{amsmath,amssymb,amsfonts}
5: \usepackage{bm}
6: \usepackage{epsfig}
7:
8: %% macros
9: \newcommand{\nop}[1]{#1}
10: \newcommand{\bra}[1]{\langle{#1}|}
11: \newcommand{\ket}[1]{|{#1}\rangle}
12: \newcommand{\braket}[2]{\langle{#1}|{#2}\rangle}
13: \newcommand{\expval}[1]{\langle{#1}\rangle}
14: \newcommand{\gwfexp}[1]{\langle{#1}\rangle}%_{\text{G}}^{\phantom{+}}}
15: \newcommand{\ham}{\hat{H}}%{\hat{H}_{\text{G}}}
16:
17: \begin{document}
18:
19: \title{Gossamer metals}
20:
21: \author{Marcus Kollar}
22:
23: % \email[Email address: ]{kollar@itp.uni-frankfurt.de}
24:
25: \affiliation{In\-sti\-tut f\"{u}r Theore\-ti\-sche Phy\-sik,
26: Jo\-hann-Wolf\-gang-Goethe-Uni\-ver\-si\-t\"{a}t Frank\-furt,
27: Ro\-bert-Mayer-Stra\ss{}e~8, D-60054~Frankfurt am~Main, Germany.}
28:
29: \date{August 26, 2003}
30:
31: \begin{abstract}
32: Laughlin's construction of exact gossamer ground states is applied
33: to normal metals. We show that for each variational parameter
34: $0\leq g\leq1$, the paramagnetic or ferromagnetic Gutzwiller wave
35: function is the exact ground state of an extended Hubbard model
36: with correlated hopping, with arbitrary particle density,
37: non-interacting dispersion, and lattice dimensionality. The
38: susceptibility and magnetization curves are obtained, showing that
39: the Pauli susceptibility is enhanced by correlations. The
40: elementary quasiparticle excitations are gapless, except for a
41: half-filled band at $g=0$, where a Mott transition from metal to
42: insulator occurs.
43: \end{abstract}
44:
45: \pacs{71.27.+a, 71.10.Fd, 71.30.+h}
46:
47: \maketitle
48:
49:
50: Progress in the understanding of many-body effects in strongly
51: correlated electron systems, such as quantum magnets, narrow-band
52: transition metal compounds, fractional quantum Hall systems, or
53: high-temperature superconductors, has depended on a variety of
54: theoretical tools. Important information about the electronic
55: structure can often be obtained from \emph{ab initio} calculations,
56: which are however less reliable if interactions between electrons
57: are dominant over their kinetic energy. On the other hand, the
58: study of idealized model systems, containing only the presumably
59: relevant degrees of freedom, can provide insight into microscopic
60: physical mechanisms. However, since such models are rarely exactly
61: solvable, analytical and numerical calculations usually involve
62: approximations or extrapolations. In view of these limitations,
63: support for proposed physical notions has occasionally come from an
64: inverse strategy: starting from a correlated many-body wavefunction
65: one constructs a hopefully ``reasonable'' model Hamiltonian for
66: which it is the exact ground state. Correlated quantum phases may
67: then be classified according to their elementary excitations or
68: correlation functions. This approach has been useful in particular
69: for the understanding of the fractional quantum Hall effect,
70: spin-Peierls or Haldane-gap antiferromagnets, and quantum
71: rotors~\cite{Arovas92}.
72:
73: Recently, Laughlin~\cite{Laughlin02a} developed a new approach to
74: high-temperature superconductivity, viewing the insulating state as
75: a superconductor with very low superfluid density. Pursuing the
76: above strategy, he proposed that the ground-state wavefunction of
77: such a ``gossamer superconductor'' is obtained from the BCS
78: mean-field product state by applying the Gutzwiller correlation
79: operator ($0\leq g\leq1$),
80: \begin{align}
81: \hat{K}(g)
82: =
83: g^{\sum_{i}\!\hat{D}_{i}}
84: =
85: \prod_{\scriptstyle i}
86: \left[1-(1-g)\hat{D}_{i}\right]
87: ,\label{eq:K}
88: \end{align}
89: where $\hat{D}_{i}=\hat{n}_{i\uparrow}\hat{n}_{i\downarrow}$ is the
90: operator for double occupation at lattice site $i$, and
91: constructed a corresponding model Hamiltonian. Elementary
92: excitations~\cite{Laughlin02a}, the transition from superconductor
93: to Mott insulator~\cite{Zhang02a}, magnetic
94: instabilities~\cite{Bernevig03a}, and related mean-field
95: Hamiltonians~\cite{Bernevig03a,Yu02ff} were also studied in this
96: context.
97:
98: The purpose of this letter is the application of Laughlin's gossamer
99: paradigm to normal metals, i.e., itinerant electrons on a lattice
100: without broken (discrete) translational symmetries. (In particular,
101: antiferromagnetic or superconducting phases are excluded.) It is
102: well known that a metallic system can be driven into an insulating
103: state by strong electronic correlations. This type of transition
104: from metal to insulator, the Mott transition, occurs for example in
105: transition metal oxides, and has been analyzed by a variety of
106: theoretical methods \cite{Gebhard97a}. These include the
107: variational Gutzwiller wavefunction (GWF) \cite{Gutzwiller63ff}
108: obtained by acting with (\ref{eq:K}) on an uncorrelated Fermi sea.
109: In general the GWF describes a correlated metal, except for the
110: insulating state with one immobile particle at each lattice site
111: that results at $g=0$ for a half-filled band. When used as a
112: variational wavefunction for the Hubbard model and evaluated within
113: the Gutzwiller approximation \cite{Gutzwiller63ff}, this
114: Brinkman-Rice (BR) transition \cite{Brinkman70a} occurs at finite
115: critical Hubbard interaction $U_c^{\text{BR}}$. While the
116: Gutzwiller approximation becomes exact in the limit of infinite
117: dimensions \cite{Metzner89a}, the BR transition is shifted to
118: $U_c^{\text{BR}}=\infty$ in finite dimensions
119: \cite{vanDongen89a,Gulacsi91ff}. However, the reliability of these
120: variational results is limited, as the true ground state of the
121: Hubbard model in infinite dimensions may behave rather differently
122: \cite{Georges96a}; for example the number of doubly occupied sites
123: in general does not vanish at the transition as in the BR scenario.
124: Furthermore, the analysis of elementary excitations is hampered by
125: the fact that the true ground state is lower in energy, and on these
126: grounds the GWF has been criticized as inadequate for describing the
127: Mott transition \cite{Millis91a}. Some of these difficulties are
128: resolved for models with exact GWF ground states, which we now
129: proceed to construct.
130:
131:
132: \emph{Metallic gossamer ground state.} %
133: In general a gossamer ground state is built as follows
134: \cite{Laughlin02a}. Starting from an uncorrelated product wave
135: function $\ket{\phi}$ and operators
136: $\hat{b}_{\bm{k}\sigma}^{\phantom{+}}$ such that
137: $\hat{b}_{\bm{k}\sigma}^{\phantom{+}}\ket{\phi}=0$ for all $\bm{k}$
138: and $\sigma$, one applies an invertible many-body correlator
139: $\hat{K}$ to obtain a correlated wavefunction $\ket{\psi}$ $=$
140: $\hat{K}\ket{\phi}$, and defines
141: $\hat{\tilde{b}}_{\bm{k}\sigma}^{\phantom{+}}$ $=$
142: $\hat{K}\hat{b}_{\bm{k}\sigma}^{\phantom{+}}\hat{K}^{-1}$. Then
143: $\ket{\psi}$ is an exact ground state of the hermitian Hamiltonian
144: $\ham$ $=$ $\sum_{\bm{k}\sigma}\tilde{E}_{\bm{k}\sigma}
145: \hat{\tilde{b}}_{\bm{k}\sigma}^{+}
146: \hat{\tilde{b}}_{\bm{k}\sigma}^{\phantom{+}}$
147: for arbitrary $\tilde{E}_{\bm{k}\sigma}\geq0$, since $\ham\geq0$
148: and $\ham\ket{\psi}=0$.
149:
150: In the present context we use the Gutzwiller correlator (\ref{eq:K})
151: as in Refs.~\onlinecite{Laughlin02a,Zhang02a,Bernevig03a,Yu02ff},
152: which is invertible for $g\neq0$, $\hat{K}(g)^{-1}=\hat{K}(g^{-1})$,
153: but start from a product state containing spin-up and spin-down
154: fermions, characterized by the occupation numbers
155: $n_{\bm{k}\sigma}^{0}$ (with $n_{\bm{k}\sigma}^{0}$ $=$ $0$ or $1$),
156: \begin{align}
157: \ket{\phi}
158: &=
159: \!\!\!\!
160: \prod_{~~~~~\bm{k}\sigma~(n_{\bm{k}\sigma}^{0}=1)~}
161: \hat{c}_{\bm{k}\sigma}^{+}
162: \;\ket{0}
163: \,.\label{eq:phi}
164: \end{align}
165: This state is annihilated by the operators
166: $\hat{b}_{\bm{k}\sigma}^{\phantom{+}}$ $=$
167: $(1-n_{\bm{k}\sigma}^{0})\hat{c}_{\bm{k}\sigma}^{\phantom{+}}$ $+$
168: $n_{\bm{k}\sigma}^{0}\hat{c}_{\bm{k}\sigma}^{+}$. After some
169: algebra, we can rewrite the Hamiltonian $\ham$ as
170: \begin{align}
171: \ham
172: &=
173: \hat{H}_t
174: +
175: \hat{H}_h
176: +
177: \hat{H}_U
178: +
179: \hat{H}_X
180: +
181: \hat{H}_Y
182: +
183: \text{const}
184: \,,\label{eq:H_tXYU}
185: \\
186: \hat{H}_t
187: &=
188: \sum_{i\neq j,\sigma}
189: \nop{T}_{ij\sigma}
190: \hat{c}_{i\sigma}^{+}
191: \hat{c}_{j\sigma}^{\phantom{+}}
192: \,,%\label{eq:H_t}
193: ~~%\\
194: \hat{H}_h
195: =%&=
196: -h
197: \sum_{i}
198: (
199: \hat{n}_{i\uparrow}
200: -
201: \hat{n}_{i\downarrow}
202: )
203: \,,%\label{eq:H_h}
204: \label{eq:H_t+h}
205: \\
206: \hat{H}_X
207: &=
208: \sum_{i\neq j,\sigma}
209: X_{ij\sigma}
210: (\hat{n}_{i\bar{\sigma}}
211: +\hat{n}_{j\bar{\sigma}})
212: \hat{c}_{i\sigma}^{+}
213: \hat{c}_{j\sigma}^{\phantom{+}}
214: \,,\label{eq:H_X}
215: \\
216: \hat{H}_Y
217: &=
218: \sum_{i\neq j,\sigma}
219: Y_{ij\sigma}
220: \hat{n}_{i\bar{\sigma}}
221: \hat{n}_{j\bar{\sigma}}
222: \hat{c}_{i\sigma}^{+}
223: \hat{c}_{j\sigma}^{\phantom{+}}
224: \,,%\label{eq:H_Y}
225: ~~%\\
226: \hat{H}_U
227: =%&=
228: U\sum_i
229: \hat{n}_{i\uparrow}
230: \hat{n}_{i\downarrow}
231: \,,%\label{eq:H_U}
232: \label{eq:H_Y+U}
233: \end{align}
234: with the constant term depending only on $g$ and the total particle
235: density $n$ $=$ $\hat{n}_{\uparrow}+\hat{n}_{\downarrow}$, which is
236: fixed; we mostly consider densities $n$ $\leq$ $1$ since
237: $\ket{\psi(0)}$ $=$ $0$ otherwise. Here $\nop{T}_{ij\sigma}$ is the
238: Fourier transform of $\nop{E}_{\bm{k}\sigma}$ $=$
239: $(1-2n_{\bm{k}\sigma}^0)\tilde{E}_{\bm{k}\sigma}$, and the other
240: parameters are given by
241: \begin{align}
242: h
243: &=
244: -\frac{1}{2L}
245: \sum_{\bm{k}\sigma}
246: \sigma\,
247: (1-(1+g^2)n_{\bm{k}\sigma}^0))
248: \tilde{E}_{\bm{k}\sigma}
249: \,,\label{eq:h}
250: \\
251: U
252: &=
253: \frac{1-g^2}{g^2L}
254: \sum_{\bm{k}\sigma}
255: (1-(1-g^2)n_{\bm{k}\sigma}^0))
256: \tilde{E}_{\bm{k}\sigma}
257: \,,\label{eq:U}
258: \\
259: X_{ij\sigma}
260: &=
261: \frac{1-g}{2g}
262: \left[
263: (1-g)\nop{T}_{ij\sigma}
264: +
265: (1+g)\tilde{T}_{ij\sigma}
266: \right]
267: ,\label{eq:X}
268: \\
269: Y_{ij\sigma}
270: &=
271: \frac{(1-g)^2}{2g^2}
272: \left[
273: (1+g^2)\nop{T}_{ij\sigma}
274: +
275: (1-g^2)\tilde{T}_{ij\sigma}
276: \right]
277: ,\label{eq:Y}
278: \end{align}
279: where $\tilde{T}_{ij\sigma}$ is the Fourier transform of
280: $\tilde{E}_{\bm{k}\sigma}$, and $L$ is the number of lattice sites.
281:
282: A model with arbitrary non-interacting dispersion
283: $\epsilon_{\bm{k}}$ can now be obtained as follows. For given band
284: dispersion $\nop{E}_{\bm{k}\sigma}$ we construct the Fermi sea via
285: $n_{\bm{k}\sigma}^0$ $=$ $\Theta(-\nop{E}_{\bm{k}\sigma})$ and let
286: $\tilde{E}_{\bm{k}\sigma}$ = $|\nop{E}_{\bm{k}\sigma}|$ $\geq$ $0$.
287: Then we put $\nop{E}_{\bm{k}\sigma}$ $=$ $\epsilon_{\bm{k}}$ $-$
288: $\epsilon_{\text{F}\!\sigma}$ and adjust the Fermi energies
289: $\epsilon_{\text{F}\!\sigma}$ so that the starting wavefunction
290: $\ket{\phi}$ is a Fermi sea with desired densities $n_\sigma$ $=$
291: $\frac{1}{L}\sum_{\bm{k}}n_{\bm{k}\sigma}^0$.
292: In the following we assume $\sum_{\bm{k}}\epsilon_{\bm{k}}$ $=$
293: $0$ for convenience,
294: hence $\epsilon_{0\sigma}$ $\equiv$
295: $\frac{1}{L}\sum_{\bm{k}}\epsilon_{\bm{k}}n_{\bm{k}\sigma}^0$ $\leq$
296: $0$.
297:
298: For each $0\leq g\leq1$ the Gutzwiller wavefunction $\ket{\psi(g)}$
299: $=$ $\hat{K}(g)\ket{\phi}$ is the exact ground state of the extended
300: Hubbard Hamiltonian (\ref{eq:H_tXYU}). It contains the kinetic
301: energy $\hat{H}_t$ of a single band $\epsilon_{\bm{k}}$, which is
302: independent of $g$, and an optional Zeeman term $\hat{H}_h$, absent
303: for $\hat{n}_{\uparrow}$ $=$ $\hat{n}_{\downarrow}$. For $g<1$,
304: $\ham$ contains interactions that involve at most two sites: a
305: repulsive on-site interaction $\hat{H}_U$ and correlated hopping
306: terms $\hat{H}_X$ and $\hat{H}_Y$, whose amplitudes are related by
307: $gY_{ij\sigma}$ $=$ $(1-g)^2(\nop{T}_{ij\sigma}+X_{ij\sigma})$.
308: Note that $X_{ij\sigma}$, $Y_{ij\sigma}$, and $U$ all diverge in the
309: limit $g\to0$. Similar interaction terms appear in models with
310: superconducting gossamer ground states
311: \cite{Laughlin02a,Zhang02a,Bernevig03a,Yu02ff}, but here those
312: states cannot be lower in energy than $\ket{\psi(g)}$. Apart from
313: $g$, the magnetic field and the strength and range of the
314: interactions depend on the chosen band dispersion
315: $\epsilon_{\bm{k}}$ and the densities $n_\sigma$. To illustrate the
316: behavior of the amplitude $\tilde{T}_{ij\sigma}$ appearing in
317: (\ref{eq:X})-(\ref{eq:Y}) we now discuss several examples.
318:
319:
320: \emph{One-dimensional systems.} %
321: The dispersion for a one-dimensional ring with nearest-neighbor
322: hopping $-t<0$ is $\epsilon_k=-2t\cos(k)$.
323: For the Fourier transform of
324: $\tilde{E}_{k\sigma}=|\epsilon_k-\epsilon_{\text{F}\!\sigma}|$ we
325: find
326: \begin{align}
327: \tilde{T}_{j\pm1,j\sigma}
328: &=
329: t\,
330: \big[
331: 2n_\sigma-1
332: +
333: \tfrac{1}{\pi}
334: \sin(\pi n_\sigma)
335: \big]
336: \,,\label{eq:tildeT_1}
337: \\
338: \tilde{T}_{j+r,j\sigma}
339: &=
340: \frac{4t}{\pi(r^2-1)}
341: \big[
342: r\sin(\pi n_\sigma r)\cos(\pi n_\sigma)
343: \\
344: &~~~
345: +
346: \cos(\pi n_\sigma r)\sin(\pi n_\sigma)
347: \big]
348: \,,~~|r|\geq2
349: ,\label{eq:tildeT_2}
350: \end{align}
351: which falls off algebraically at large distances. At half-filling
352: ($n_\sigma$ $=$ $1/2$) it is on the order of $1/r^2$ and alternates
353: in sign for even $r$, while vanishing for odd $r$. This long-range
354: behavior of $\tilde{T}_{ij\sigma}$ is rather generic. As another
355: example we consider ``$1/r$'' hopping, $\nop{T}_{j+r,j\sigma}$ $=$
356: $it(-1)^{r}/r$ with dispersion $\epsilon_k$ $=$ $tk$, for which the
357: corresponding Hubbard model was solved by Gebhard \emph{et al.} (see
358: Ref.~\onlinecite{Gebhard97a} for a review). We obtain
359: \begin{align}
360: \tilde{T}_{j+r,j\sigma}
361: &=
362: \frac{(-1)^rt}{\pi r^2}
363: \big[
364: 1-i\pi(2n_\sigma-1)r-e^{-2\pi in_\sigma r}
365: \big]
366: \,,\label{eq:tildeT_lr}
367: \end{align}
368: again with contributions proportional to $1/r$ (absent for
369: half-filling) and $1/r^2$.
370: Similar power-law behavior is typically found in dimensions $D=2,3$.
371:
372:
373: \emph{Infinite-dimensional systems.} %
374: Nearest-neigh\-bor hopping $t=1/\sqrt{2D}$ on a hypercubic lattice
375: with dispersion $\epsilon_{\bm{k}}$ $=$ $-2t\sum_{\alpha=1}^D\cos
376: k_\alpha$ yields the density of states $\rho_{\text{hc}}(\epsilon)$
377: $=$ $\exp(-\epsilon^2/2)/\sqrt{2\pi}$ in the limit $D\to\infty$
378: \cite{Metzner89a,Georges96a}.
379: In order to construct the corresponding amplitude
380: $\tilde{T}_{ij\sigma}$ further assumptions about its symmetry are
381: necessary. Following Ref.~\onlinecite{Bluemer03a} we assume that it
382: depends only on the ``taxi-cab'' distance $||\bm{R}||$ $=$
383: $\sum_{\alpha=1}^D|R_\alpha|$ and use the appropriate scaling
384: $\tilde{T}_{ij\sigma}$ $=$
385: $\tilde{T}_{r,\sigma}^*/\smash[b]{\sqrt{2^r\scriptstyle\binom{D}{r}}}$
386: where $r$ $=$ $||\bm{R}_i-\bm{R}_j||$ $\geq$ $0$. We then obtain
387: \begin{align}
388: \tilde{T}_{r\sigma}^*
389: &=
390: \int%\limits_{-\infty}^{\infty}
391: |\epsilon-\epsilon_{\text{F}\!\sigma}|
392: \,
393: \rho_{\text{hc}}(\epsilon)
394: \,
395: \frac{\text{He}_r(\epsilon)}{\sqrt{r!}}
396: \,d\epsilon
397: \,,
398: \\
399: \tilde{T}_{2r+1,\sigma}^*
400: &=
401: (1-2n_\sigma)
402: \,
403: \delta_{r,0}
404: \,,
405: \\
406: \tilde{T}_{2r+2,\sigma}^*
407: &=
408: 2\,
409: \rho_{\text{hc}}(\epsilon_{\text{F}\!\sigma})
410: \frac{\text{He}_{2r}(\epsilon_{\text{F}\!\sigma})}{\sqrt{(2r+2)!}}
411: \,,
412: \end{align}
413: where $\text{He}_{n}(x)$ are Hermite polynomials. At half-filling we
414: find $\tilde{T}_{2r\sigma}^*\sim r^{-5/4}$, corresponding to an
415: effective correlated hopping range $\sum_r\tilde{T}_{r\sigma}^*{}^2$
416: of order unity. For other densities of states $\rho(\epsilon)$, in
417: particular those with finite bandwidth, it is also possible
418: construct a corresponding dispersion $\epsilon_{\bm{k}}$
419: \cite{Bluemer03a}, and then derive $\nop{T}_{ij\sigma}$ and
420: $\tilde{T}_{ij\sigma}$ in a similar fashion.
421:
422:
423: \emph{Response to external magnetic field.} %
424: Returning to the case of arbitrary dispersion and densities, we note
425: that according to the equation of state (\ref{eq:h}) the
426: ground-state magnetization $m$ $=$
427: $\hat{n}_{\uparrow}-\hat{n}_{\downarrow}$ is nonzero if an external
428: magnetic field $h$ is present. For the homogeneous susceptibility
429: $\chi$ we obtain
430: \begin{align}
431: \chi(h)^{-1}
432: &=
433: \frac{\partial h}{\partial m}
434: =
435: \frac{1}{4}
436: \sum_\sigma
437: \frac{
438: 1-(1-g^2)n_{\sigma}
439: }{
440: \rho(\epsilon_{\text{F}\!\sigma})
441: }
442: \,.\label{eq:chi_h}
443: \end{align}
444: \begin{figure}[t]
445: \centerline{\epsfig{file=figure1.eps,width=\hsize}}
446: \caption{Magnetization curves $m(h)$
447: for a one-dimensional ring with nearest-neighbor hopping at
448: half-filling. Inset: same for $1/r$ hopping.
449: \label{fig:m}}
450: \end{figure}
451: In the limit of zero field this reduces to $\chi(0)$ $=$
452: $\chi_0/[1-(1-g^2)n/2]$. As expected the system behaves like a
453: correlated paramagnet, i.e., the interactions enhance the Pauli
454: susceptibility $\chi_0$ $=$ $2\rho(\epsilon_{\text{F}})$ of the
455: uncorrelated system. However, it should be kept in mind that the
456: interaction parameters (\ref{eq:U})-(\ref{eq:Y}) do not remain
457: constant when the parameters $h$ or $m$ are varied.
458: Fig.~\ref{fig:m} shows the magnetization as a function of magnetic
459: field for a one-dimensional ring at half-filling. Interestingly, for
460: nearest-neighbor hopping the upward curvature of these magnetization
461: curves is very similar to Bethe-ansatz results for the pure Hubbard
462: model \cite{Lieb68a,Takahashi69a}, where a metal-insulator
463: transition occurs at $U_c=0^+$. By contrast, for $1/r$ hopping the
464: magnetization curves are strictly linear, $m$ $=$ $\chi(0)h$, due to
465: the constant density of states.
466:
467:
468: \emph{Metal-insulator transition.} %
469: For an unpolarized half-filled band ($n$ $=$ $1$,
470: $\epsilon_{\text{F}\!\sigma}$ $=$ $0$), the ground-state wave
471: function $\ket{\psi(g)}$ describes a metal for $g>0$ and an
472: insulator for $g=0$. In the insulating state there are no doubly
473: occupied sites, the discontinuity of $n_{\bm{k}\sigma}$ at the Fermi
474: surface vanishes, and the kinetic energy $\gwfexp{\hat{H}_t}$ is
475: zero. This Mott metal-insulator transition in the ground state of
476: $\ham$ occurs at infinite interactions (\ref{eq:U})-(\ref{eq:Y}), in
477: contrast to the variational BR transition, or numerical results for
478: the Hubbard model in infinite dimensions \cite{Georges96a}.
479:
480: Nevertheless we may, somewhat artificially, shift the transition to
481: finite interactions as follows. Clearly $\ket{\psi(g)}$ remains the
482: ground state when we multiply $\ham$ by a positive $g$-dependent
483: factor, although qualitatively different Hamiltonians may then
484: result in the limit $g\to0$. For example, for the Hamiltonian
485: $\ham^{(1)} = g \ham$ the $X$ term has a finite limit, while
486: $\ham^{(2)} = g^2 \ham$ yields a vanishing $X$ term and finite $Y$
487: and $U$ terms; in both cases the quadratic kinetic energy vanishes
488: at $g=0$. In particular we may conclude that for any dispersion
489: $\epsilon_{\bm{k}}$ the Hamiltonian
490: \begin{align}
491: \ham'
492: &=
493: \sum_{i\neq j,\sigma}
494: Y_{ij}'
495: \hat{n}_{i\bar{\sigma}}
496: \hat{n}_{j\bar{\sigma}}
497: \hat{c}_{i\sigma}^{+}
498: \hat{c}_{j\sigma}^{\phantom{+}}
499: +
500: U'
501: \sum_i
502: \hat{n}_{i\uparrow}
503: \hat{n}_{i\downarrow}
504: \,,\label{eq:H_prime}
505: \end{align}
506: where $Y_{ij}'$ is the Fourier transform of $\epsilon_{\bm{k}}
507: (1-n_{\bm{k}\sigma}^0)$, has the exact ground state
508: $\ket{\psi(g=0)}$ at half-filling if $U'$ $\geq$ $U_c'$, with
509: critical interaction $U_c'$ $=$ $-\sum_\sigma \epsilon_{0\sigma}$
510: $\equiv$ $|\epsilon_0|$. Interestingly, the uncorrelated kinetic
511: energy also sets the energy scale of the BR transition in the
512: Gutzwiller approximation \cite{Brinkman70a}, where $U_c^{\text{BR}}$
513: $=$ $8|\epsilon_0|$. Although $\ham'$ is not a standard Hubbard
514: Hamiltonian, it nonetheless appears to be the simplest model with a
515: BR-type transition to an exact insulating ground state at finite
516: Hubbard interaction.
517:
518: \begin{figure}[t]
519: \centerline{\epsfig{file=figure2.eps,width=\hsize}}
520: \caption{Expectation values of parts of $\ham$,
521: (\ref{eq:H_t+h})-(\ref{eq:H_Y+U}), for nearest-neighbor hopping
522: in $D$ $=$ $1$ (arrows) and $D$ $=$ $\infty$ at half-filling.
523: Inset: double occupancy for $1/r$ hopping in $D$ $=$ $1$, as
524: compared to the pure Hubbard model.\label{fig:H_tXYU}}
525: \end{figure}
526:
527: \emph{Ground-state expectation values.} %
528: The separate expectation values of the kinetic energy $\hat{H}_t$
529: and the interaction terms $\hat{H}_X$, $\hat{H}_Y$, and $\hat{H}_U$
530: can be calculated from the quantities $n_{\bm{k}\sigma}$ $=$
531: $\gwfexp{\hat{c}_{\bm{k}\sigma}^{+}
532: \hat{c}_{\bm{k}\sigma}^{\phantom{+}}}$, $d$ $=$ $\frac{1}{L}
533: \sum_{i} \gwfexp{\hat{n}_{i\uparrow} \hat{n}_{i\downarrow}}$,
534: $x_{ij\sigma}$ $=$ $\gwfexp{\hat{n}_{i\bar{\sigma}}
535: \hat{c}_{i\sigma}^{+} \hat{c}_{j\sigma}^{\phantom{+}}}$, and
536: $y_{ij\sigma}$ $=$ $\gwfexp{\hat{n}_{i\bar{\sigma}}
537: \hat{n}_{j\bar{\sigma}} \hat{c}_{i\sigma}^{+}
538: \hat{c}_{j\sigma}^{\phantom{+}}}$. Using the methods of
539: \cite{Metzner87ff,Kollar01ff} it can be shown that for the GWF the
540: Fourier transforms of the latter are given by
541: \begin{align}
542: x_{\bm{k}\sigma}
543: &=
544: n_{\bm{k}\sigma}^{0}
545: \bigg[
546: n_{\bar{\sigma}}
547: -\frac{1-n_{\bm{k}\sigma}}{1-g}
548: \bigg]
549: -
550: \frac{(1-n_{\bm{k}\sigma}^{0})g\,n_{\bm{k}\sigma}}{1-g}
551: -
552: d
553: \,,\label{eq:xks-alternate}
554: \\
555: y_{\bm{k}\sigma}
556: &=
557: n_{\bm{k}\sigma}^{0}
558: \bigg[
559: n_{\bar{\sigma}}
560: -\frac{1-n_{\bm{k}\sigma}}{(1-g)^2}
561: \bigg]
562: +
563: \frac{(1-n_{\bm{k}\sigma}^{0})
564: g^2n_{\bm{k}\sigma}}{(1-g)^2}
565: -
566: d
567: \,,\!\!
568: \label{eq:yks-alternate}
569: \end{align}
570: i.e., only the momentum-space distribution $n_{\bm{k}\sigma}$ and
571: the double occupancy $d$ are needed. They may be evaluated
572: numerically by Monte-Carlo methods, but are also available in closed
573: form under certain circumstances. For one-dimensional systems with
574: symmetric Fermi sea ($n_{k\sigma}^{0}$ $=$ $n_{-k\sigma}^{0}$) both
575: quantities have been calculated analytically
576: \cite{Metzner87ff,Kollar01ff}. In dimension $D=2,3$ high-order
577: perturbative methods can be used \cite{Gulacsi91ff}. The Gutzwiller
578: approximation \cite{Gutzwiller63ff}, with piecewise constant
579: momentum distribution $n_{\bm{k}\sigma}$, is recovered in infinite
580: dimensions \cite{Metzner89a}.
581:
582: The expectation values of the various parts of $\ham$ are shown in
583: Fig.~\ref{fig:H_tXYU} for nearest-neighbor hopping in $D$ $=$ $1$
584: and $D$ $=$ $\infty$ at half-filling. We note that
585: $\gwfexp{\hat{H_X}}$ approaches a constant for $g\to0$, while
586: $\gwfexp{\hat{H_Y}}$ and $\gwfexp{\hat{H_U}}$ diverge. This behavior
587: occurs for all $D$, since $d\sim g^2\ln(1/g)$ in one dimension
588: \cite{Metzner87ff}, $d=o(g)$ in all finite dimensions
589: \cite{vanDongen89a,Gulacsi91ff}, and $d$ $\sim$ $g$ in infinite
590: dimensions. We may thus conclude that the penalty that $\hat{H}_U$
591: imposes on double occupancies is compensated by assisted hopping due
592: to the nonstandard three-body interaction $\hat{H}_Y$.
593:
594: The effect of correlated hopping is also apparent when comparing to
595: the pure Hubbard ring with $1/r$ hopping, which features a
596: metal-insulator transition at $U_c$ $=$ $2\pi t$ with continuous
597: nonzero double occupancy $d$ \cite{Gebhard97a}. For comparison with
598: previous studies of variational wavefunctions in the vicinity of
599: this transition \cite{Gebhard94a,Dzierzawa95a}, $d$ vs.\ $U$ is
600: shown in the inset of Fig.~\ref{fig:H_tXYU}. The results for both
601: models with $1/r$ hopping agree for weak interactions, but the
602: energy gain from correlated hopping leads to a larger number of
603: doubly occupied sites for strong coupling in the model
604: (\ref{eq:H_tXYU}), as expected.
605:
606:
607: \emph{Quasiparticle excitations.} %
608: The known ground state of $\ham$ suggests that it might also be
609: possible to calculate dynamical properties of the model, such as the
610: spectral function. Unfortunately the construction of exact excited
611: states is not straightforward, be it with one added or removed
612: particle, or with charge or spin excitations. We therefore proceed
613: by considering the variational states
614: \cite{Laughlin02a,Buenemann03a}
615: \begin{align}
616: \ket{\bm{k}\sigma}
617: =
618: \hat{K}\hat{b}_{\bm{k}\sigma}^{+}\ket{\phi}
619: =
620: \left\{
621: \begin{array}{ll}
622: \hat{K}\hat{c}_{\bm{k}\sigma}^{+}\ket{\phi}
623: &
624: \text{~if~}n_{\bm{k}\sigma}^{0}=0
625: \\
626: \hat{K}\hat{c}_{\bm{k}\sigma}^{\phantom{+}}\ket{\phi}
627: &
628: \text{~if~}n_{\bm{k}\sigma}^{0}=1
629: \end{array}
630: \right.
631: \,,
632: \end{align}
633: whose mean energy is
634: \begin{align}
635: \nop{E}_{\bm{k}\sigma}^{\pm}
636: =
637: \frac{
638: \bra{\bm{k}\sigma}
639: \ham
640: \ket{\bm{k}\sigma}}{\braket{\bm{k}\sigma}{\bm{k}\sigma}}
641: =
642: \frac{
643: \braket{\psi}{\psi}}{
644: \braket{\bm{k}\sigma}{\bm{k}\sigma}}
645: \tilde{E}_{\bm{k}\sigma}\;
646: \,,
647: \end{align}
648: where the commutator relations
649: ${[\hat{b}_{\bm{k}\sigma}^{\phantom{+}},}$
650: ${\hat{b}_{\bm{k'}\sigma'}^{+}]}$ $=$
651: $\delta_{\bm{k}\bm{k}'}\delta_{\sigma\sigma'}$ and
652: ${[\hat{b}_{\bm{k}\sigma}^{\phantom{+}},}$
653: ${\hat{b}_{\bm{k'}\sigma'}^{\phantom{+}}]}$ $=$ $0$ were used. The
654: states $\ket{\bm{k}\sigma}$ are mutually orthogonal and their energy
655: is thus an upper bound to the quasiparticle energy for momentum
656: $\bm{k}$ and spin $\sigma$. The variational energy to add a
657: particle (i.e., $n_{\bm{k}\sigma}^{0}=0$) is
658: \begin{align}
659: \nop{E}_{\bm{k}\sigma}^{+}
660: &=
661: \frac{
662: |\epsilon_{\bm{k}}-\epsilon_{\text{F}\!\sigma}|
663: }{
664: 1-(1+g)[(1-g)n_{\bar{\sigma}}+(1+g)n_{\bm{k}\sigma}]
665: }
666: \,,
667: \end{align}
668: while for the removal of a particle (with $n_{\bm{k}\sigma}^{0}=1$)
669: \begin{align}
670: \nop{E}_{\bm{k}\sigma}^{-}
671: &=
672: \frac{
673: g^2
674: |\epsilon_{\bm{k}}-\epsilon_{\text{F}\!\sigma}|
675: }{
676: (1+g)[(1-g)n_{\bar{\sigma}}+(1+g)n_{\bm{k}\sigma}]-1-2g
677: }
678: \,.
679: \end{align}
680: \begin{figure}[t]
681: \centerline{\epsfig{file=figure3.eps,width=\hsize}}
682: \caption{Quasiparticle excitations in a one-dimensional
683: ring with nearest-neighbor hopping $t>0$.\label{fig:E}}
684: \end{figure}
685: Clearly the quasiparticle excitations are gapless, since
686: $\nop{E}_{\bm{k}\sigma}^{\pm}$ $\to$ $0$ close to the Fermi surface.
687: Fig.~\ref{fig:E} shows these energies for one-dimensional
688: nearest-neighbor hopping at half-filling.
689:
690:
691: \emph{Conclusion.} %
692: We have constructed and characterized a new class of itinerant
693: electron models for which the metallic Gutzwiller wavefunction is an
694: exact ground state, due to the interplay of Hubbard interaction and
695: correlated hopping. For a half-filled band a Mott metal-insulator
696: transition similar to the Brinkman-Rice scenario occurs,
697: illustrating Mott's original idea of a quantum phase transition
698: entirely due to charge correlations without magnetic ordering.
699: Further study of the elementary excitations in these models should
700: be fruitful.
701:
702: This work was supported in part by the DFG via Forschergruppe
703: FOR~412.
704:
705: \begin{thebibliography}{99}
706:
707: \bibitem{Arovas92}%
708: For a review, see D.~P.\ Arovas and S.~M.\ Girvin, in: {\em Recent
709: Progress in Many-Body Theories}, Vol.\ 3, edited by T.~L.\
710: Ainsworth, C.~E.\ Campbell, B.~E.\ Clements, and E.\ Krotschek
711: (Plenum Press, New York, 1992), p.\ 315.
712:
713: \bibitem{Laughlin02a}%
714: R.~B.\ Laughlin, \eprint{cond-mat/0209269}.
715:
716: \bibitem{Zhang02a}%
717: F.~C.\ Zhang, \eprint{cond-mat/0209272}.
718:
719: \bibitem{Bernevig03a}%
720: B.~A.\ Bernevig, R.~B.\ Laughlin, and D.~I.\ Santiago,
721: \eprint{cond-mat/0303045}.
722:
723: \bibitem{Yu02ff}%
724: Y.\ Yu, \eprint{cond-mat/0211131}; \eprint{cond-mat/0303501}.
725:
726: \bibitem{Gebhard97a}%
727: F.\ Gebhard, \textit{The Mott metal-insulator transition: models
728: and methods} (Springer, Berlin 1997).
729:
730: \bibitem{Gutzwiller63ff}%
731: M.~C.\ Gutzwiller,
732: Phys.\ Rev.\ Lett.\ {\bf 10}, 159 (1963);
733: Phys.\ Rev.\ {\bf 134}, A 923 (1964);
734: Phys.\ Rev.\ {\bf 137}, A 1726 (1965).
735:
736: \bibitem{Brinkman70a}%
737: W.~F.\ Brinkman and T.~M.\ Rice,
738: Phys.\ Rev.\ B {\bf 2}, 4302 (1970).
739:
740: \bibitem{Metzner89a}%
741: W.\ Metzner and D.\ Vollhardt,
742: Phys.\ Rev.\ Lett.\ {\bf 62}, 324 (1989).
743:
744: \bibitem{vanDongen89a}%
745: P.~G.~J.\ van Dongen, F.\ Gebhard, and D.\ Vollhardt, Z.\ Phys.\ B
746: {\bf 76}, 199 (1989).
747:
748: \bibitem{Gulacsi91ff}%
749: Z.\ Gul\'{a}csi and M.\ Gul\'{a}csi,
750: Phys.\ Rev.\ B {\bf 44}, 1475 (1991).
751: Z.\ Gul\'{a}csi,
752: M.\ Gul\'{a}csi, and
753: B.\ Jank\'{o},
754: Phys.\ Rev.\ B {\bf 47}, 4168 (1993).
755:
756: \bibitem{Georges96a}%
757: For a review, see A.\ Georges, G.\ Kotliar, W.\ Krauth, and M.~J.\
758: Rozenberg, Rev.\ Mod.\ Phys.\ {\bf 68}, 13 (1996).
759:
760: \bibitem{Millis91a}%
761: A.~J.\ Millis and S.~N.\ Coppersmith,
762: Phys.\ Rev.\ B {\bf 43}, 13770 (1991).
763:
764: \bibitem{Metzner87ff}%
765: W.\ Metzner and D.\ Vollhardt,
766: Phys.\ Rev.\ Lett.\ {\bf 59}, 121 (1987);
767: Phys.\ Rev.\ B {\bf 37}, 7382 (1988);
768: \emph{ibid.}~{\bf 39}, 12339 (1989).
769:
770: \bibitem{Kollar01ff}%
771: M.\ Kollar and D.\ Vollhardt, Phys.\ Rev.\ B {\bf 63}, 045107
772: (2001); \emph{ibid.}~{\bf 65}, 155121 (2002).
773:
774: \bibitem{Gebhard94a}%
775: F.\ Gebhard and A.\ Girndt, Z.\ Phys.\ B {\bf 93}, 445 (1994).
776:
777: \bibitem{Dzierzawa95a}%
778: M.\ Dzierzawa, D.\ Baeriswyl, M.\ Di Stasio, Phys.\ Rev.\ B {\bf
779: 51}, 1993 (1995).
780:
781: \bibitem{Bluemer03a}%
782: N.\ Bl\"{u}mer and P.~G.~J.\ van Dongen,
783: \eprint{cond-mat/0303204}. N.\ Bl\"{u}mer, Ph.~D.\ thesis
784: (Universit\"{a}t Augsburg, 2002).
785:
786: \bibitem{Lieb68a}%
787: E.~H.\ Lieb and F.~Y.\ Wu,
788: Phys.\ Rev.\ Lett.\ {\bf 20}, 1445 (1968).
789:
790: \bibitem{Takahashi69a}%
791: M.~Takahashi, Progr.\ Theor. Phys. {\bf 42}, 1098 (1969).
792:
793: \bibitem{Buenemann03a}%
794: J.\ Buenemann, F.\ Gebhard, and R.\ Thul, Phys.\ Rev.\ B {\bf 67},
795: 075103 (2003).
796:
797: \end{thebibliography}
798:
799: \end{document}
800: