1: %format latexe
2: \input epsf
3: %\input psfig.tex
4: \documentstyle[12pt,cites,doublespace]{article}
5: %\input graphics
6: %\pagestyle{empty}
7: \topmargin -1.2cm %-0.7cm
8: \oddsidemargin -0.30 cm
9: \textheight 25.0 cm
10: \textwidth 16.7 cm
11: %\openup 12 pt
12: %\newcommand\ds{\displaystyle}
13: %\renewcommand{\baselinestretch}{1.5}
14: \setstretch{1.5}
15: \setlength{\parindent}{3.0ex}
16: \newfont{\kleiner}{eurm10}
17: %\parindent 2pt
18: \begin{document}
19: %\pagestyle{empty}
20: \vspace*{0.8cm}
21: %\begin{center}
22: {\noindent
23: ELECTRON CORRELATIONS IN THE HIGH $\rm T_c$-COMPOUNDS}\\
24:
25: {\noindent Gernot STOLLHOFF}\\
26: Max-Planck-Institut f\"ur
27: Festk\"orperforschung, D-70569 Stuttgart,
28: Germany \\
29:
30: %\end{center}
31:
32:
33: \begin{singlespace}
34: \noindent
35: Ab-initio correlation results for an idealized high $T_c$-compound
36: are compared to density functional (DF) calculations for the same system.
37: It is shown that and why the DF-charge distribution is wrong. The
38: largest deficiency arises for the $Cu$-$d_{x^2-y^2}$-occupation,
39: arising from strong atomic correlations but mostly from anomalous neighbor
40: $Cu$-spin correlations. Both features are beyound the range of the
41: homogeneous electron gas approximation underlying the DF-schemes.
42: The ab-initio results also exclude a description of the real
43: system in a Mott-Hubbard scenario, that is mostly chosen in theory.
44: Conditions for models are derived that are able to describe the
45: high-$T_c$-compounds.\\
46: \end{singlespace}
47:
48: {\noindent 1. INTRODUCTION}
49:
50: In the past, two approaches have dominated the attempts to
51: understand the electronic structure of the high-$T_C$ compounds.
52: The first is by ab-initio DF-calculations within the local density
53: approximation (LDA), and the second is by very particular models.
54:
55: The LDA calculations suffer from the underlying homogeneous electron gas
56: approximation for exchange and correlations. They are able to represent
57: properties like equilibrium lattice constants, phonons, or the
58: Fermi surface, but are unable to reproduce the magnetic phase at half
59: filling, and have deficiencies for the band masses. The obtained
60: electron lattice coupling is too small for the explanation
61: of superconductivity.
62:
63: The particular models are mostly restricted to
64: the $Cu$-$O$-planes. The description of the electrons in
65: these planes is usually further restricted to a single band Hubbard model
66: in the strong correlation limit or to a t-J model. Such a model can explain
67: the magnetism of the so called half-filled planes, and deliberate
68: Fermi surfaces can be generated by particular parametrizations. A
69: connection to the microscopic reality however, or even the unequivocal
70: construction of a plausible new mechanism of superconductivity, could
71: not yet be obtained.
72:
73: We had been able to pursue a different approach that
74: neither suffers from the uncontrollable shortcomings of
75: the LDA, nor from an
76: equally uncontrollable ad hoc choice of a particular
77: model. With the help of the Local Ansatz (LA), we were able
78: to perform an ab-initio correlation
79: calculation for an idealized high-$T_c$ compound$^1$.
80:
81: It is generally agreed that a basic understanding for
82: the high $T_c$-compounds can be gained from
83: the treatment of a simplified idealized compound, namely a
84: single charged $Cu$-$O$-plane. Actually most models make
85: use of this approximation.
86: Our calculation was restricted to such a system, or rather to its
87: so called infinite layer generalization into three dimensions.
88:
89: In the following, we will compare the LA-results to similar
90: LDA-calculations, and will unequivocally determine
91: the deficiencies of the LDA, and their causes. We will also
92: depict a few important correlation results and give a critical
93: valuation of the most often used models.\\
94:
95: \noindent 2. COMPUTATION METHOD
96:
97: The LA allows the
98: ab-initio computation of the correlated
99: ground state of a solid. Starting point for the correlation treatment
100: is a self consistent field (SCF) or Hartree-Fock calculation for
101: this solid, resulting in the uncorrelated single-particle
102: ground state $\Psi_{SCF}$. This SCF-calculation was performed
103: by the program Crystal$^2$. The latter program uses atom centered
104: Gauss type orbitals (GTO) as a basis, as does the LA.
105:
106: Within the LA,
107: the following variational ansatz is made for the correlated ground
108: state:
109: \begin{eqnarray}
110: | \Psi_{\rm corr} \rangle & = & e^{-S} | \Psi_{\rm SCF} \rangle
111: \label{eq:lanst0} \\
112: S & = & \sum_{\nu} \eta_{\nu} O_{\nu}
113: \label{eq:lanst01} \\
114: O_{\nu} & = & \left\{ \matrix{
115: n_{i \uparrow} n_{i \downarrow} \cr
116: n_i n_j \cr
117: \vec{s}_i \cdot \vec{s}_j \cr
118: \{n_{i \uparrow} (a_{i \downarrow}^{\dagger}a_{j\downarrow}-
119: a_{j \downarrow}^{\dagger}a_{i\downarrow})\} +
120: \{ \uparrow \leftrightarrow \downarrow \} \cr
121: n_i } \label{eq:lanst}
122: \right. \ \ .
123: \end{eqnarray}
124: The $\eta$'s serve as variational
125: parameters. The $n_{i \sigma}$ and $\vec{s}_i$ are density and spin operators for an
126: electron in the local state $a_{i \uparrow}^{\dagger}$, represented by
127: the orbital
128: \begin{equation}
129: g_i(\vec{r}) = \sum_j \gamma_{i j} f_j(\vec{r}) \label{eq:local}
130: \end{equation}
131: where the $f_i(\vec{r})$ are the (GTO like) basis orbitals.
132: The operators have an
133: obvious meaning. The first operator $n_{i \uparrow} n_{i \downarrow}$,
134: for example, when applied to $ | \Psi_{\rm SCF} \rangle $, projects out all
135: configurations with two electrons in orbital $ g_i(\vec{r})$. In connection
136: with the variational parameter $ \eta_{\nu}$, it
137: partially suppresses those configurations. Similarly, the operators $n_i n_j$
138: describe density correlations between electrons in local orbitals
139: $g_i(\vec{r})$ and $ g_j(\vec{r})$.
140: The operators $ \vec{s}_i \cdot \vec{s}_j $ generate
141: spin correlations. The fourth kind of operators is of the form of
142: $[O_{\nu},H_0]_-$, where $H_0$ represents the single-particle Hamiltonian.
143: These operators refine the
144: ansatz with respect to the band energy of the electrons involved.
145: The original operators of eq.
146: \ref{eq:lanst} are next modified by subtracting the
147: contracted contributions in each of them.
148: The corrected operators when applied to $ | \Psi_{\rm SCF} \rangle$
149: contain only two-particle excitations, and the corrected
150: last kind of operators in eq. \ref{eq:lanst} covers local single
151: particle excitations, i.e. it allows for changes in occupations.
152:
153: The variational parameters $\eta_{\nu}$ are chosen to optimize
154: the energy
155: \begin{eqnarray}
156: E_G & = & \frac{\langle\Psi_{\rm corr} | H | \Psi_{\rm corr} \rangle}
157: {\langle\Psi_{\rm corr} | \Psi_{\rm corr} \rangle} \\
158: & = & \langle\Psi_{\rm corr} | H | \Psi_{\rm corr} \rangle_{c} \ \ .
159: \end{eqnarray}
160: In the last equation,
161: the subscript ${}_{c}$ indicates that only connected
162: diagram contributions are summed up.
163: This expression cannot be evaluated exactly.
164: The standard approximation is an expansion in powers of $\eta$, up to second
165: order,
166: \begin{eqnarray}
167: E_G & = &
168: E_{\rm SCF} + E_{\rm corr} \\
169: E_{\rm corr} & = & - \sum_{\nu} \eta_{\nu} \langle
170: O^{\dagger}_{\nu} H\rangle \\
171: 0 & = & - \sum_{\nu} \eta_{\nu} \langle O^{\dagger}_{\nu} H\rangle
172: + \sum_{\nu,\mu} \eta_{\nu} \eta_{\mu} \langle O^{\dagger}_{\nu}
173: HO_{\mu}\rangle_c
174: \ \ . \label{eq:expan}
175: \end{eqnarray}
176: Here, $\langle A \rangle$ means the expectation value of the operator
177: $A$ within
178: $| \Psi_{\rm SCF}\rangle$.
179: This is a weak correlation approximation that blows up when
180: correlations turn too strong. For the high-$T_c$'s no such problems
181: arose, indicating that the Mott-Hubbard limit does not apply.
182:
183: The local orbitals in eq.~\ref{eq:local}
184: are connected to a single atom only and are built from
185: its basis orbitals. Standard Quantum Chemistry
186: (QC) methods also start from a SCF-calculation and add correlations
187: with the help of one-and two-particle operators. However, the QC-operators
188: are constructed from orthogonal sets of occupied and/or unoccupied
189: orbitals, a construction that fails for metals. The restriction to
190: local orbitals and the particular construction of correlation operators
191: from these local orbitals is the essential approximation of the LA
192: in comparison tho QC. It allowes even to treat metals with QC accuracy.
193:
194: For the high-$T_c$ application, only operators built from
195: atomic orbitals were used. The atomic orbitals are
196: unequivocally determined from the SCF-ground state
197: by the condition that
198: they are built from basis orbitals on the respective atoms only and
199: that they cover a maximal fraction of the full occupied space.
200: The resulting orbitals are next L\"owdin-orthogonalized to each other.
201: More localized subatomic orbitals were not used in this applicaltion.
202:
203: The calculated system is formally be described as $SrCuO_2$.
204: It is half-filled and has an antiferromagnetic
205: ground state. However, in all calculations, not this
206: antiferromagnetic ground state but a metastable non magnetic state
207: was treated since we are not interested in the magnetic
208: order but in the electronic properties of the doped high-$T_c$ compounds.
209: This approach is necessary because the Crystal program requires
210: unit cells with integer filling, but we wanted to avoid large unit
211: cells. We took care that
212: none of the computed results was influenced by the proximity to the
213: magnetic instability.
214:
215: Being interested on in-plane features, we choose a good basis for
216: the in plane $Cu$- and $O$- atoms, but added no valence basis orbitals
217: for the $Sr$-atoms. This leads to a charging of the planes
218: with 2 electrons per $Cu$-atom, and renders the interplane coupling
219: negligible.
220: For more details of the basis choice, of the structure of the idealized
221: system, and of the correlation calculations, we refer to ref. 1).\\
222:
223: \noindent 3. ATOMIC CHARGE DISTRIBUTION
224:
225: Being unequivocally defined, the atomic orbitals, generated
226: for correlation purposes,
227: are also a perfect basis for a detailed charge
228: analysis. This representation avoids all non orthogonality
229: problems of a standard
230: Mulliken population analysis.
231:
232: The partial charge distributions
233: $n_i(\Psi)=\langle\Psi| \sum_{\sigma}n_{i\sigma}|\Psi\rangle$
234: are presented in table \ref{tab32} for different states $\Psi$.
235: The first row contains the values for $\Psi=\Psi_{SCF}$.
236: The sum over the partial occupations approaches the
237: number of valence electrons within 0.02.
238: This proximity rectifies the condensation into
239: atomic orbitals and the specific method of their computation.
240:
241: With the addition of correlations, a relatively large charge transfer
242: occurs. Ultimately, it is a charge transfer mostly from the
243: $Cu 3d_{x^2-y^2}$-orbitals into the $O 2p$-orbitals. However, for
244: its understanding it is necessary to progress stepwise.
245: A first step is the addition of atomic correlations which
246: lead to a large correlation energy gain.
247: The dominant charge transfer due to the atomic correlations
248: is from the $Cu 3d_{x^2-y^2}$-orbitals
249: to the $Cu 4s,4p$-orbitals, followed by a secondary
250: redistribution
251: from the $Cu 4s,4p$-orbitals to the $O 2p$-orbitals.
252: Over all, 0.18 electrons are removed from the
253: $Cu 3d_{x^2-y^2}$-orbitals, and put into the $Cu 4s,p$-shell (0.13) and
254: the $O 2s,p$-shells (2x0.03).
255: More than 80 percent of this charge transfer arise from the inclusion
256: of the operators
257: $n_{i\uparrow}n_{i\downarrow}$ for the $Cu 3d_{x^2-y^2}$-orbitals, the
258: remaining part stems from the same operators for the $4s,p$-orbitals.
259: For an explanation, we refer to ref 1).
260:
261: When neighbor correlations are included, then an additional charge
262: transfer of the same magnitude as the one due to on-site correlations
263: occurs.
264: It is dominantly from the
265: $Cu 3d_{x^2-y^2}$-orbitals to the $O 2p_b$-orbitals, and is due to
266: a particular spin correlation between neighbor $Cu$-sites that will
267: be discussed later. The longer range contributions that were covered
268: by the present computations lead to a further but small transfer
269: of the same kind. For an error analysis, we refer to ref. 1) again.
270:
271: \begin{table}
272: \begin{center}
273: %\tabcolsep 0.8 pt
274: { \begin{tabular}{|l|r|rrr|lc|}
275: \hline
276: Orbital& HF&on site corr&nn corr&full corr&LDA&\quad\\
277: \hline
278: $Cu 3d_{x^2-y^2}$ & 1.51& 1.33& 1.17& 1.15&1.53&\\
279: $Cu 3d_{z^2}$ & 1.95& 1.94& 1.94& 1.94&1.90&\\
280: $Cu 3d_{xy},3d_{xz},3d_{yz}$ & 2.00& 2.00& 2.00& 2.00&1.99&\\
281: \hline
282: $Cu 4s$ & 0.50& 0.55& 0.57& 0.58& 0.64 &\\
283: $Cu 4p_{pl}$ & 0.30& 0.33& 0.34& 0.34&0.31&\\
284: $Cu 4p_{\perp}$ & 0.09& 0.11& 0.11& 0.11
285: &0.18&\\
286: \hline
287: $O 2s$ & 1.82& 1.82& 1.81& 1.81&1.80&\\
288: $O 2p_b$ & 1.42& 1.48& 1.57& 1.58&1.39&\\
289: $O 2p_{orth}$ & 1.97& 1.96& 1.96& 1.96&1.94&\\
290: $O 2p_{\perp}$ & 1.95& 1.93& 1.92& 1.91&1.92&\\
291: \hline
292: \end{tabular}}
293: \end{center}\protect
294: \caption{
295: Charge distributions for the SCF ground state and with correlations
296: added, in comparison to LDA results. The subindices of the $p$-orbitals
297: have the following meaning: $\perp$ perpendicular to plane,$pl$ in plane,
298: $b$ in bond direction, $orth$ in plane perp. to bond}
299: \label{tab32}
300: \end{table}
301:
302:
303: The latest version of the program Crystal$^2$ also allowes to perform
304: LDA-calculations within the same basis set as used for the
305: SCF-calculation. The resulting LDA charge distribution is
306: analysed in the same way as done before for the SCF-case.
307: Most alternative
308: LDA-schemes have no atom centered basis. Thus,
309: their ground state results can not be fully decomposed
310: into atomic occupations. Only
311: partial results were so far available, mostly from
312: tight binding fits to the energy bands. Here,
313: the first full LDA charge analysis is given.
314:
315: The LDA-result is included in table \ref{tab32}.
316: The occupations of the $3d$-orbitals
317: are in very good agreement with
318: earlier LDA results (for citations, see Ref 1)).
319: There is a close agreement with the SCF-result. Only
320: orbitals almost filled in LDA are even
321: more filled in SCF-approximation, resulting from the well known
322: band spreading of the SCF. Correlations partially undo this.
323: The proximity to SCF indicates that none of the
324: correlation corrections are contained in the LDA result.
325: It is plausible that the anomalous
326: neighbor correlation effects can not be covered by the homogeneous electron
327: gas approximation, but it was somewhat astonishing that the
328: atomic correlation effects are also completely disregarded in LDA.
329:
330: These occupation results
331: demonstrate why LDA must be very deficient for the
332: high $T_c$-compounds.
333: A consequence for the band structure can be easily derived.
334: When keeping the LDA hopping terms in the single-particle Hamiltonian
335: but shifting the crystal field terms so that the resulting single-particle
336: ground state has the correct charge distribution, then
337: the Fermi surface stays the same, but the conduction
338: band shrinks by half, bringing it into much closer agreement to
339: experiment. More details will be given elsewhere$^3$.\\
340:
341:
342:
343: \noindent 4. SPIN CORRELATIONS AWAY FROM HALF FILLING
344:
345: All LA results discussed so far were
346: connected to LDA results. But in contrast to the LDA, the LA
347: obtains also full information
348: for the correlated ground state, and in particular
349: correlation functions. In the following, we will deal
350: with the surprising
351: anomalous neighbor $Cu$-spin
352: correlations. These are connected to
353: a charge transfer from the two $Cu$-atoms into the inbetween $O$-atom
354: because the latter enhances the antiferromagnetic coupling and further
355: reduces the charge fluctuations on all three
356: involved atoms. This is a correlation feature of the delocalized
357: electron system. It is not at all connected to a Mott-Hubbard
358: transition which can be ruled out by our results.
359: These spin correlations
360: mutually enhance each other. Even without
361: any long range magnetic order in the metastable metallic state, they are so
362: big that the repulsive effect of the on-site $Cu$-correlation hole
363: is overcompensated, and that electrons with differing spins have a higher
364: probability to be close to each other (i.e. up to neighbor sites)
365: than without correlations.
366:
367: This finding has a similarity
368: to the so called renomalized valence bond (RVB)
369: picture that had been proposed as a mechanism for
370: superconductivity$^4$, and that also
371: leads to a strong neighbor spin correlation .
372: While the RVB approach is only valid close to the Mott-Hubbard
373: transition, our results
374: arise already for weak correlations and are also present at large dopings.
375:
376: These short range magnetic correlation
377: features are next compared to experiment.
378: This can only be done outside the magnetically
379: ordered phase. While the ab-initio calculations themselves can not be extended
380: to fractional charging due to the Crystal program limitations,
381: a valuable alternative is to condense the half-filled case informations
382: to a model,
383: and to extend this model to differing
384: fillings. Such a model needs to describe the conduction band well,
385: and also needs to explicitly include the electronic interactions.
386: We had chosen the smallest possible model that contains the
387: $Cu$-$3d_{x^2-y^2}$ orbitals, the $O$-$2p_b$ orbitals, and also
388: the $Cu$-$4s$ orbitals. The latter orbitals were needed
389: to obtain a correct occupation of the other two orbitals, necessary
390: for the short range spin correlations.
391: A three band model (without the $4s$-orbitals) turned
392: out to be very deficient. Even our model choice caused shortcomings
393: because we had to move charge from the $4s$-orbitals into the
394: $2p$-orbitals in order to preserve the correct Fermi surface. This
395: might be overcome in the future by extending the model by
396: $4p$-orbitals.
397:
398: The hopping terms for this model were taken
399: from an LDA-band structure fit found in the literature$^5$, and the
400: crystal field terms were obtained from the required occupations.
401: As interaction, Hubbard interaction terms $U_i$ were included for
402: the individual orbitals. Their value was set so that the
403: atomic correlation functions obtained by the LA for the model
404: were the sames as the ab-initio results$^1$.
405: This model describes the atomic correlations correctly but gives,
406: probably for the deviations in the occupations, too small neighbor
407: spin correlations. We therefore performed a parallel calculation
408: with a $U_{3d}$ that was enhanced by 20 percent.
409:
410: With this model we calculated
411: the spin correlation function
412: \begin{equation}
413: S(\vec{Q})
414: =\sum_{i,j,\vec{G} }
415: \langle\Psi_{corr}|\vec{s}_i(0)\vec{s}_j(\vec{G})|\Psi_{corr}\rangle
416: e^{i\vec{Q}(\vec{r}_i-\vec{r}_j-\vec{G})} \ \ \ .
417: \label{eqs3}
418: \end{equation}
419: of the model with 15 percent doping. Here, $i,j$ runs over the atomic
420: orbitals
421: on sites $r_i,r_j$
422: in the unit cell, and $G$ describes the lattice summation.
423: This function can be compared to the corresponding, experimentally
424: measured spin
425: correlation function$^7$ $S(\vec{Q})$ for the
426: metallic compound $La_{0.85}Sr_{0.15}Cu_2O_4$.
427: The latter is no equal time correlation function but only energy integrated
428: up to 0.15eV.
429:
430: \begin{figure}[hbtp]
431: %\unitlength1.0cm%
432: %\begin{picture}(12,8)%
433: %\centerline{\epsfxsize=12mm\epsfbox{fig1.eps}}
434: %\end{picture}%
435: %\centerline{\psfig{figure=fig1.eps,height=8.0cm,width=8.0cm}}
436: \epsfysize=65mm
437: \epsfxsize=75mm
438: \centerline{\epsffile{fig1.eps}}
439: %\vspace{9.0cm}
440: \caption{\protect Equal time spin correlation function
441: S(Q) for $\vec{Q}=(h,h,0)$
442: in comparison to experiment$^6$ (empty circles).
443: Given are the results of the HF-ground state (broken-dotted curve),
444: the 5 atom cluster result (dotted line) and the 9 atom cluster result
445: (continuous line) for $U_{3d}$=6.3eV, and the 9 atom
446: cluster result for
447: $U_{3d}$=7.8eV.}
448: \label{figs1}
449: \end{figure}
450: \vspace{0.5cm}
451: Fig. \ref{figs1} contains the experimental and theoretical
452: results for
453: the diagonal (1,1) axis.
454: The zone boundary is at h=1, the intensity is
455: given per formular unit which here is equivalent to a unit cell or to a
456: single $Cu$ atom.
457:
458: The lowest curve represents the result for the single-particle ground
459: state. It represents the exchange holes. As the finite value at $h=0$
460: indicates, the summation in eq. \ref{eqs3} was not brought to convergency.
461: The maximal deviation occurs for $h=0$ were the contributions from all
462: missing terms add up. Due to dephasing, the correction is very much
463: smaller for finite $h$.
464:
465: Next, short range correlations as they are deduced from a single coherent
466: 5 $Cu$ cluster calculation are included (dotted curve). Here, the nearest
467: neighbor $Cu-Cu$ correlations come into play and cause a peak at
468: the zone boundary ($h=1$).
469: When extending the correlation treatment to a 9 Cu
470: cluster, the peak narrows somewhat and enhances a little (continuous
471: curve). Finally, also the corresponding values with enlarged $U$ (7.8eV
472: instead of 6.3eV) are given (broken curve).
473:
474: As expected, the
475: theoretical equal time correlation function is always larger
476: than the experimental correlation function whose energy integration extends
477: only to 0.15 eV. One would expect that a sizable part of the correlation
478: function, namely the one already obtained by $\Psi_{SCF}$ can
479: definitely not be
480: seen by the slow neutrons. Consequently, all remaining
481: contributions need to be seen in the experiment. This might
482: occur because the electrons partially localize due to a charge density
483: wave formation in the case of the treated compound.
484:
485: It should be noted that neither Hubbard model nor t-J-model
486: results are able to explain such a strong inelastic magnetic
487: scattering at 15 percent doping. Also a 3 band model would
488: definitively fail if reasonable Hubbard interaction terms were chosen.
489: Even the 4 band model used here displays deficiencies.
490:
491: Our results indicate that the origin of the inelastic magnetic
492: scattering might well be completely disconnected
493: from the magnetic order at half
494: filling, and might be explained by the anomalous short range correlations
495: found over a large range of doping. At least the theoretical Q-dependence
496: matches the experiment very well. \\
497:
498: \noindent REFERENCES
499:
500: \begin{singlespace} \noindent
501: 1) G. STOLLHOFF, Phys. Rev. {\bf B 58} (1998) 9826.\\
502: 2) V.R. SAUNDERS et al., Crystal98 User's Manual, University of
503: Torino (1998).\\
504: 3) G. STOLLHOFF, unpublished. \\
505: 4) P.W. ANDERSON, Science {\bf 235} (1987) 1196.\\
506: 5) O. K. ANDERSEN et al., J. Phys, Chem. Solids {\bf 56} (1995) 1573.\\
507: 6) M. HAYDEN et al., Phys. Rev. Lett. {\bf 76} (1996) 1344.\\
508: \end{singlespace}
509: \end{document}
510: