cond-mat0307292/m07.tex
1: %
2: %  Manuscript for the proceedings 
3: %  of The Summer School "New Magnetics" (Bedlewo, 2003)
4: %  To be published in Molecular Physics Reports
5: %
6: %  Corrected version as of September 24, 2003:
7: %  minor corrections in the text and some references updated
8: %
9: %\documentclass[11pt]{elsart}
10:  \documentclass[11pt,a4paper]{article}
11: % Fine-tune margins:
12:  \setlength{\textwidth}{15.3cm}
13:  \setlength{\oddsidemargin}{0.3cm}
14:  \setlength{\evensidemargin}{0.3cm}
15:  \setlength{\textheight}{21.5cm}
16:  \setlength{\topmargin}{ 0.2cm}
17: \usepackage{epsfig}
18: 
19: \begin{document}
20: \renewcommand{\thefootnote}{\fnsymbol{footnote}} % set footnote symbols
21: 
22: %\begin{frontmatter}
23:   \title{\emph{Ab initio} Simulations of Fe-based Ferric Wheels}
24:  \author{
25:  A.~V. Postnikov$^1$\footnotemark[1], Jens Kortus$^2$\footnotemark[2]
26:  and Stefan Bl\"ugel$^{1,3}$
27: \\*[2mm]
28: $^1$\emph{Universit\"at Osnabr\"uck -- Fachbereich Physik,}\\
29:     \emph{D-49069 Osnabr\"uck, Germany} \\
30: $^2$\emph{Max-Planck Institut f\"ur Festk\"orperforschung, Heisenbergstr. 1,}\\
31:     \emph{D-70569 Stuttgart, Germany} \\
32: $^3$\emph{Institut f\"ur Festk\"orperforschung, Forschungszentrum J\"ulich,}\\  
33:     \emph{D-52425 J\"ulich, Germany} 
34:         }  % author{ 
35: \date{}  % use {} in standard article class, to suppress printing date
36: \maketitle
37: \footnotetext[1]{Corresponding author. Tel. +49-541-969-2377, fax: -2351; 
38: E-mail: apostnik@uos.de.\\
39: Permanent address: Institute of Metal Physics, 620219 Yekaterinburg, Russia
40:                 }  % footnotetext 
41: \footnotetext[2]{Tel. +49-711-689-1664, fax: -1632; 
42: E-mail: j.kortus@fkf.mpg.de.
43:                 }  % footnotetext 
44: 
45: \begin{abstract}
46: Based on first-principles density-functional theory calculations we investigate 
47: the electronic structure of hexanuclear ``ferric wheels'' 
48: %\linebreak
49: $M$Fe$_6$[N(CH$_2$CH$_2$O)$_3$]$_6$Cl ($M$ = Li, Na) 
50: in their antiferromagnetic ground state. The electronic structure 
51: is presented in form of spin- and site-resolved local densities of states. 
52: The latter clearly indicate that the magnetic moment 
53: is distributed over several sites. The local moment at the iron site is still 
54: the largest one with about 4 $\mu_{\mbox{\tiny B}}$, thus indicating  
55: the valence state of iron to be closer to Fe(II) than to commonly
56: accepted Fe(III). The local spin of $S$=5/2 per iron site, following from 
57: magnetization measurements, is perfectly reproduced if one takes the moments 
58: on the neighbor atoms into account. The largest magnetic polarization 
59: is found on the apical oxygen atom, followed by nitrogen bridging oxygens. 
60: These findings are confirmed by a map of spatial spin density. A further 
61: goal of the present study has been a comparative test of two different DFT 
62: implementations, {\sc Siesta} and NRLMOL. They yield a very good 
63: agreement down to small details in the electronic structure.
64: \end{abstract}
65: 
66: \section*{Introduction}
67: 
68: There has been a remarkable increase in the number of studies 
69: on molecular magnets during the last decade which
70: clearly is driven by the progress in the chemistry of such systems. 
71: Many different families of molecular magnetic systems composed of 
72: all possible transition-metal atoms in various combinations with
73: stabilizing organic ligands have been synthesized, their crystal structure 
74: determined and their magnetic properties analyzed \cite{Mol_Magnets}. 
75: 
76: While there is a large amount of experimental data available,
77: the theoretical description in terms of quantitative -- and often
78: also qualitative -- understanding is missing in some cases, in others it
79: stays behind the experimental progress. The traditional approach 
80: to describe these systems is based on model Hamiltonians 
81: (``exact'' diagonalization  within,  e.g., Heisenberg model).
82: One the other hand, with the advance in computer technology and the 
83: development in theoretical methods, most notably in density-functional theory 
84: (DFT) \cite{RMP71-1253}, first-principles calculations became feasible.
85: Both approaches simplify real systems and are prone to certain
86: shortcomings. The model Hamiltonians neglect the true chemical
87: environment and bonding. All interactions are reduced to just few model
88: parameters. One is dependent on experimental input in order to
89: estimate these parameters, and the accuracy of quantitative predictions
90: depends on the parameters chosen. Even more problematic, it is not 
91: \emph{a priori} clear which interactions are important and should be 
92: included in the model and which are negligible. 
93: %Because the interactions e.g. in the Heisenberg model are reduced to
94: %effective interaction between spins only, all microscopic information about 
95: %different exchange path are lost, which is a limitation if one wants 
96: %to optimize magnetic properties by changes of structure or bonding.
97: In contrast, first-principles calculations based on quantum mechanics
98: use the atomic positions as the only experimental input and, as
99: chemical bonding is included in a straightforward way, they are able to provide 
100: microscopic foundations of the magnetic behavior. Unfortunately, the solution 
101: of Schr{\"o}dinger's equation is limited to a few electrons so that one has 
102: to use approximations. The chemical complexity of molecular magnets 
103: (or, their large number of atoms) excluded so far the 
104: ``straightforward'' simulations using approaches of superior accuracy, 
105: like quantum Monte Carlo. Density-functional based 
106: calculations have become the working horse in today's first-principles 
107: simulations.  DFT has been proven to be very accurate and reliable in many 
108: cases, although electron correlation effects beyond the mean-field 
109: approximation in DFT are neglected, which limits the accuracy and may lead 
110: to failures in the description of a given system.
111: 
112: Both approaches, the model-based and DFT-based, are useful and important 
113: for our understanding, and are best used together. While DFT can provide 
114: the input parameters for models (e.g. the exchange parameters $J$ 
115: for a Heisenberg model),
116: the model Hamiltonians can deliver information on many-body effects,
117: temperature dependence, and quantum fluctuations,
118: which are beyond the reach of DFT. By describing the relevant physics 
119: from different points of view
120: one develops a complete picture where essential features do not remain
121: unexplored.
122: 
123: Concentrating in the following on DFT-based strategies, we note that
124: molecular magnets turn out to be hard cases for computational treatment 
125: -- for different reasons in different established calculation approaches.
126: The openness of the crystal structure creates problems for traditional
127: muffin-tin- or atomic sphere approximation-based tools, 
128: and requires requires a ``full-potential'' formalism instead.
129: The presence of transition metals and/or other constituents with deep
130: semicore states, which can not be neglected, creates problems for 
131: pseudopotential planewave methods. Moreover, the need for a large 
132: (albeit largely empty) ``simulation box'' around a molecule makes 
133: planewave-related methods expensive. An abundance of hydrogen and hence 
134: strong disbalance in atom ``sizes'' makes the linearized augmented plane wave 
135: scheme numerically demanding. The often unusual chemical composition and 
136: coordination reduces the usefulness of \emph{a priori} tuned basis sets 
137: in those tight-binding-type methods which heavily depend on pre-adjusted 
138: basis functions for their efficiency. 
139: Further, the option to treat non-collinear spin-density distribution 
140: and to have accurate forces on atoms  for studying structure relaxation 
141: and dynamics may be advantageous and sometimes necessary.
142: As a consequence, flexible, efficient and accurate methods (and codes) are
143: in demand, of which several have been successfully applied
144: to systems of reasonable complexity. 
145: 
146: \section*{Calculations}
147: 
148: As an example of a system of practical interest we consider the
149: hexanuclear ``ferric wheels'' $M$Fe$_6$[N(CH$_2$CH$_2$O)$_3$]$_6$Cl
150: ($M$ = Li, Na, see Fig.~1), synthesized at the 
151: Institut f\"ur Organische Chemie in Erlangen \cite{AnChIE36-2482} 
152: and labeled as substances {\bf 4} and {\bf 3} in the latter publication.
153: These materials belong to a larger family of Fe-coronates, of which
154: 8- and 10-member rings are also known. A common feature of these
155: substances is the bonding of Fe atoms via doubled oxygen bridges,
156: somehow resembling the 90$^{\circ}$ coupling of magnetic atoms
157: in transition-metal oxides. In the vicinity of each Fe atom
158: one finds a nitrogen atom, connected via C$_2$H$_4$ chains to oxygen atoms. 
159: The Li- and Na-centered molecules have a very similar structure, with only 
160: slightly different Fe--O--Fe angles (101.1$^\circ$ for Li-wheel
161: and 103.3$^\circ$ for Na-wheel).
162: 
163: \begin{figure}
164: %\label{fig:molecule}
165: \centerline{\epsfig{figure=fig1.eps,width=9.5cm}}
166: \bigskip
167: \caption{The Li-centered ``ferric wheel'' molecule. The Li ions
168: in the middle of the ring; the distant Cl ion included in the simulation
169: is not shown; the rest of (electrically neutral) solvent is neglected.}
170: \end{figure}
171: 
172: The interest in the magnetic properties of the system is 
173: the presence of large magnetic moments, traditionally argued
174: to correspond to $S$=5/2 on the Fe site, thus implying
175: a highly ionized Fe(III) state. Such an identification of the iron magnetic
176: moment is consistent with magnetization and torque measurements 
177: \cite{InCh38-5879}, from which Waldmann \emph{et al.} derived 
178: the values of exchange parameters $J$ for the spin Hamiltonian
179: of the Heisenberg model:
180: \begin{equation}
181: H = -J \left(\sum_{i=1}^5 {\bf S}_i{\cdot}{\bf S}_{i+1}
182: + {\bf S}_6{\cdot}{\bf S}_1 \right) +
183: (\mbox{anisotropy term}) + (\mbox{Zeeman term})\,.
184: \end{equation}
185: The obtained values of $J$ are $-$18 to $-$20 K for the Li-wheel 
186: (depending on sample and method) and $-$22.5 to $-$25 K for the Na-wheel,
187: thus implying an antiferromagnetic (AFM) ground state. 
188: While the data from bulk magnetic measurements seem to be well  
189: established by now and reliable, virtually no spatially resolved
190: probing of the magnetic structure has been carried out yet. 
191: Recent X-ray photoelectron and X-ray emission spectroscopy studies 
192: \cite{our_XPS} allow to probe the electronic structure in the valence band 
193: and on the Fe site, albeit without resolution in spin. This is exactly 
194: where \emph{ab initio} calculations may yield a valuable contribution 
195: in providing site- and spin-resolved densities of states, as well as 
196: the spatial distribution of charge and magnetization densities. 
197: The verification of these predictions demands the comparison with 
198: (atom- and/or spin-averaged) spectroscopic data, and eventually comparison 
199: with other calculational approaches  at a different level of approximations.
200: 
201: In the present study, we compare the results of electronic structure
202: calculations by two different methods -- {\sc Siesta}\cite{JPCM14-2745}
203: and NRLMOL \cite{PRB41-7453}. %  ,\cite{PRB42-3276}. 
204: Both methods are based on the DFT; from the various exchange-correlation
205: functionals implemented we used the generalized gradient approximation after 
206: Perdew--Burke--Ernzerhof \cite{PRL77-3865} for the present study. 
207: A further similarity is that their basis sets are atom-centered functions -- 
208: numerical pseudoatomic orbitals in {\sc Siesta} 
209: \cite{PRB64-235111} %  ,\cite{JPCM8-3859} 
210: and contracted Gaussian-type orbitals in NRLMOL \cite{PRA60-2840}.
211: The main difference is that {\sc Siesta} uses norm-conserving pseudopotentials 
212: whereas NRLMOL is an all-electron code. NRLMOL uses the symmetry, which is 
213: relatively high in case of the Fe$_6$-wheels, of a molecule in question 
214: in all numerical tasks. Moreover, all computationally intensive parts 
215: of the code are massively parallelized \cite{PSSB217-197}, which makes DFT 
216: calculations on molecular magnets (with a typical size of 100--200 atoms) 
217: feasible. {\sc Siesta} is competitive due to the compactness and efficiency
218: of its numerical basis sets. Moreover, it can work as an order-$N$ method
219: (albeit not yet for magnetic systems), solving the DFT problem without 
220: explicit diagonalization. 
221: Other features of the two calculational methods, which are potentially important
222: in application to molecular magnets but have not been actually used 
223: in the present study, are the availability of accurate forces in both codes,
224: the spin-orbit coupling treated within second order perturbation theory 
225: in NRLMOL \cite{PRB60-9566}, the possibility of general-form
226: (i.e., non-collinear) magnetization in {\sc Siesta} \cite{Fe-clus}.
227: Being an \emph{ab initio} pseudopotential code, {\sc Siesta} is usually
228: able to provide very good accuracy for the electronic structure 
229: and total energies without fitting to any exterior information, but still 
230: benchmark calculations 
231: make sense for critical cases and/or new systems. The present study
232: serves in part exactly this task, as the accuracy of all-electron
233: NRLMOL method is expected to be superior to that of {\sc Siesta}.
234: As the latter is routinely used for dynamical simulations in large systems
235: with virtually all symmetry lost, no special treatment is provided
236: for the use of possible point group symmetries available in the molecule.
237: For the systems in question, due to high degeneracy of energy levels
238: close to the chemical potential, the neglect of symmetry leads to
239: a certain instability in the convergence, which however can be managed
240: by keeping the mixing parameter very low (10$^{-3}$--10$^{-4}$)
241: and imposing the fixed spin moment scheme (FSM, Ref.~\cite{JPF14-L129}).
242: 
243: \section*{Results}
244: 
245: In the following, we discuss the results obtained by {\sc Siesta}
246: for the Li-centered molecule, and by NRLMOL -- for the Na-centered one,
247: with the coordinates as reported from the crystallographic study
248: of corresponding materials \cite{AnChIE36-2482}. The NRLMOL treatment
249: was restricted to the ground-state AFM configuration
250: (alternating orientations of Fe magnetic moments over the ring);
251: the {\sc Siesta} calculation addressed in addition different magnetic 
252: configurations, that allowed to make an estimate of exchange parameters. 
253: 
254: Fig.~2 displays the partial densities of states (DOS) on Fe
255: and its several neighbors in the AFM configuration, 
256: as calculated by both methods.
257: The discrete levels of the energy spectra are weighted (with the charge density
258: integrated over atom-centered spheres in NRLMOL, or according to Mulliken 
259: population analysis in {\sc Siesta}), and broadened for better visibility,
260: with broadening parameter of 0.15 eV  in {\sc Siesta} and 0.14 eV in NRLMOL. 
261: The local moments corresponding to integrating such partial DOS
262: over occupied states are given in Table \ref{tab:DOS}.
263: 
264: %---------------------- Table ----------------------------------------
265: \begin{table}[b]
266: \caption{Local magnetic moments $M$ at Fe and its neighbors. NRLMOL
267: results correspond to spin density integrated over sphere of radius $R$
268: centered at corresponding atom; {\sc Siesta} values are due to Mulliken
269: popluation analysis.}
270: \begin{center}
271: \begin{tabular}{|l@{\hspace*{1.5cm}}r@{.}l
272:                   @{\hspace*{2.0cm}}r@{.}l
273:                   @{\hspace*{2.5cm}}r@{.}l|}
274: \hline
275: Atom & \multicolumn{2}{c}{\hspace*{-20mm}$R$(a.u.)} & 
276:        \multicolumn{2}{c}{\hspace*{-25mm}$M$($\mu_{\mbox{\tiny B}}$), 
277:                         NRLMOL} & 
278:        \multicolumn{2}{c|}{\hspace*{-10mm}$M$($\mu_{\mbox{\tiny B}}$), 
279:                         {\sc Siesta} } 
280: \\
281: \hline
282: Fe          & 2&19 &      3&85 &      3&91 \\
283: O (apical)  & 1&25 &      0&20 &      0&30 \\
284: O (bridge)  & 1&25 & $\pm$0&01 & $\pm$0&02 \\
285: N           & 1&32 &      0&07 &      0&09 \\
286: \hline
287: \end{tabular}
288: \end{center}
289: \label{tab:DOS}
290: \end{table}
291: %---------------------------------------------------------------------
292: 
293: One notes a remarkable agreement between both calculations up to
294: the finest details in the distribution of state densities at Fe and O
295: sites. This is the more astonishing as the local DOS is a loosely
296: defined property, normally dependent on the choice of region it attributes to, 
297: or, in the present context, to the choice of basis functions which are quite 
298: differently constructed in both calculational approaches. It also indicates 
299: that there is no significant influence of the central atom (Li or Na) 
300: on the electronic structure. Using NRLMOL we calculated the total charge 
301: within a sphere with a radius of 2.97 Bohr units around the Na atom. 
302: There are already 10.98 electrons contained inside the sphere and zero spin 
303: polarization which clearly indicates that the Na atom does not play a role for 
304: the electronic structure of the ferric wheel.
305: 
306: \medskip
307: \begin{figure}[h]
308: \centerline{\epsfig{figure=fig2.eps,width=13.9cm}}
309: \caption{Atom- and spin-resolved partial densities of states
310: as calculated for Li-centered molecule by {\sc Siesta}
311: (left panel) and for Na-centered molecule by NRLMOL (right panel). 
312: The DOS at the Fe site is scaled down by a factor of 2 relative
313: to other constituents. The numbering of atoms which are neighbors
314: to the Fe atom is shown in the inset. See text for details of
315: calculation.}
316: \end{figure}
317: 
318: \newpage
319: Notably, both methods find the local magnetic moments on Fe sites 
320: are very close to 4 $\mu_{\mbox{\tiny B}}$
321: and \emph{not} to 5 $\mu_{\mbox{\tiny B}}$ as is generally
322: assumed, based on the above mentioned magnetization data. The maximal
323: magnetization $S$=5/2 of the Fe atom corresponds to a Fe(III)-ion with
324: in 3$d^5_{\uparrow}d^0_{\downarrow}$ configuration.
325: Our first-principles calculations suggest a somewhat different picture:
326: the minority-spin DOS has a non-zero occupation
327: due to the hybridization (chemical bonding) of Fe$3d$ with O$2p$ states.
328: However, the magnetic polarization in the organic ligand which provides the
329: octahedral coordination for the iron atoms, due to Fe is substantial, 
330: the most pronounced effect being on the apical oxygen atom 
331: (which is not participating in the bonding to the next Fe neighbor).
332: Taken together with the (smaller) polarization of the bridging oxygen atoms
333: and magnetization at the nitrogen site, the distributed
334: magnetic moment \emph{per} Fe atom yields 5 $\mu_{\mbox{\tiny B}}$,
335: recovering the agreement with the magnetization results.
336: 
337: An important consequence  is that the charge state
338: of iron is not Fe(III) but more close to Fe(II), according to our
339: calculations. Moreover, the distributed magnetic moment behaves like
340: a rigid one, in a sense that it can be inverted, following a spin flip
341: on a Fe site. This conclusion has been derived from our analysis of other 
342: magnetic configurations, done with {\sc Siesta} \cite{our_XPS}.
343: The local DOS does not change considerably when switching from 
344: AFM to FM configuration -- only the HOMO/LUMO gap becomes less pronounced,
345: and a slight ferro-magnetic shift between the two spin bands appears. 
346: Further, the change of total energies in the sequence of FSM, from the maximal
347: spin moment of 30 $\mu_{\mbox{\tiny B}}$ to one Fe spin inverted
348: (20 $\mu_{\mbox{\tiny B}}$), and then to two (second-neighbors) Fe spins
349: inverted is linear, supporting the conclusion that the system maps well 
350: onto the Heisenberg model. The corresponding exchange parameter, as estimated 
351: over both 30$\rightarrow$20 and 20$\rightarrow$10 $\mu_{\mbox{\tiny B}}$ flips,
352: is around $-$80 K. This yields a correct sign and correct order of
353: magnitude, relatively to experimental estimates of Ref.~\cite{InCh38-5879}.
354: However, the magnitude is overestimated by a factor of $\sim$4.
355: The exchange parameters $J$ depend on the spatial overlap of the
356: $d$-orbitals on different Fe-sites. It is well known that the $d$-orbitals
357: within DFT are not localized enough compared to experiment, consequently
358: the $J$ values will be overestimated. There are two main reasons for
359: this shortcoming. First, possible on-site correlations as known from
360: atomic physics are underestimated in case of ``conventional'' DFT.
361: Second, DFT is not free from spurious self-interactions due to the
362: replacement of the point-like electrons by corresponding densities.
363: Bringing in the atomic physics in the form of LDA+$U$ (adding a local orbital
364: dependent atomic Coulomb interaction parameter $U$ to DFT, \cite{LDA+U}) or 
365: self-interaction corrections (SIC) \cite{PRB37-9919} % ,\cite{PRL65-1148}
366: will improve the results by lowering the $d$-orbitals in energy and
367: therefore localizing them stronger. SIC only affects
368: occupied states, whereas LDA+$U$ plunges the occupied $d$-states and
369: shifts the unoccupied ones to higher energies.
370: By increasing, on the average, the magnetic excitation energy across
371: the spin majority-minority gap, both mechanisms
372: help to effectively reduce the magnitude of $J$.
373: In what regards the LDA+$U$ approach, this has been shown
374: by Boukhvalov \emph{et al.} for another molecular magnet,
375: Mn$_{12}$ \cite{PRB65-184435}.
376: 
377: \begin{figure}[h]
378: \centerline{\epsfig{figure=fig3.eps,width=12.5cm,clip=true}}
379: \bigskip
380: \caption{Plot of the spin density map calculated with NRLMOL. 
381: The iso-surfaces correspond to $\pm 0.01 e/\mbox{\AA}^3$. While most 
382: of the magnetic moment is localized at the Fe atoms, there is still 
383: some spin polarization on O and N.}
384: \end{figure}
385: 
386: It is worth noting that the inversion of the third Fe spin
387: that yields the AFM configuration does not follow the same linear
388: trend. The origin of this is not clear at the moment and should be
389: tested in further high-precision calculations. 
390: We recently came aware of calculations by E.~Ruiz on similar systems 
391: \cite{Ruiz-private}, done also with the use of {\sc Siesta}.
392: According to them, the $J$ values extracted just from the 
393: FM--AFM total energy difference along the procedure described in
394: Ref.~\cite{JCompCh24-982} yielded the $J$ values of $\approx$ $-$24 K
395: (Li system) and $\approx$ $-$45 K (Na-system). A similar estimate from FM 
396: to AFM transition with our total energy data gives $-$38 K, i.e., 
397: such estimates seem to be in better agreement with the reported experimental 
398: data. It remains to be tested, however, whether the FM to AFM transition 
399: can be reasonably well fitted to the Heisenberg model with a single 
400: (nearest-neighbors) exchange parameter, in view of above mentioned
401: nonlinear dependence of the total energy on the total spin.
402: In V$_{15}$, another molecular magnet, a complex system of six exchange
403: parameters was found to be necessary \cite{PRL86-3400} in an otherwise 
404: conceptually similar fit to the Heisenberg model.
405: 
406: A clear visualization of the above discussed delocalized
407: (or, rather, distributed) magnetic moment associated
408: with the Fe atom comes from the map of spin density, obtained
409: from the NRLMOL calculation (Fig.~3). One should take into account that
410: the volume enclosed by the iso-surfaces is not directly 
411: correlated to the total moment at the site. 
412: One sees moreover an absence of magnetization on carbon and hydrogen
413: sites. The fact that the magnetization is noticeable and changes
414: its sign when passing through bridge oxygen atoms emphasizes 
415: the failure of methods depending the spherical averaging of
416: atom-centered potentials. 
417: {}From the side of experiment, it would be interesting to probe 
418: the spin polarization on oxygen and nitrogen atoms.
419: 
420: \section*{Conclusion}
421: 
422: We presented a study of the electronic structure of a six-membered ferric 
423: wheel with Li or Na as a central atom. The local densities of states 
424: on the iron and neighboring sites clearly indicate that the magnetic moment 
425: is distributed over several sites. The local moment at the iron site is still 
426: the largest one with about 4 $\mu_{\mbox{\tiny B}}$, although indicating that 
427: the iron in the ferric wheels is closer to Fe(II) than Fe(III) as generally 
428: expected. The local spin of $S$=5/2 per iron site as deduced from 
429: magnetization measurements is reproduced if one takes the moments 
430: on the neighbor atoms into account. The largest moment is found on the apical 
431: oxygen atom, followed by smaller moments on nitrogen and the bridging oxygen
432: atoms. This picture has been confirmed by a plot of the spin density map. 
433: A further goal of the present study has been a comparative
434: test of two different DFT implementations. In most cases the local 
435: (energy-resolved) DOS and (spatially resolved) magnetization density
436: show a very good quantitative agreement. The most pronounced deviations
437: are related to nitrogen, indicating less localized magnetic moment there
438: and thus higher sensitivity of numbers to the exact normalization
439: of DOS.
440: 
441: In order to explore the range of systems where the presented first-principles 
442: methods give reliable results further studies on more systems are required. 
443: In particular studies of experimentally well characterized classes of 
444: compounds where only the ligands are changed in a controlled way but not 
445: the magnetic core would reveal if our calculational approach can reproduce 
446: the observed changes in the magnetic behavior. Our current results make us 
447: very confident in the predictive power of the presented methods. This should 
448: allow for a microscopic understanding based on the electronic structure 
449: of single molecule magnets.
450: 
451: \section*{Acknowledgments}
452: The authors thank the Deutsche Forschungsgemeinschaft for
453: financial support (Priority Program ``Molecular Magnetism'')
454: and appreciate useful discussions with J\"urgen Schnack,
455: Oliver Waldmann, Sorin Chiuzb\u{a}ian, Manfred Neumann, Eliseo Ruiz
456: and Mikhail Katsnelson.
457: The crystal structure data have been kindly provided by the group of Prof.
458: Dr. R.~W.~Saalfrank from the Institute for Organic Chemistry of the
459: University Erlangen-N{\"u}rnberg.
460: 
461: %\bibliography{Global,update,add,my_publist}
462: %\bibliographystyle{elsart-num}
463: 
464:  \small
465: 
466: \begin{thebibliography}{10}
467: \expandafter\ifx\csname url\endcsname\relax
468:   \def\url#1{\texttt{#1}}\fi
469: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
470: 
471: \bibitem{Mol_Magnets}
472: W.~Linert and M.~Verdaguer (Eds.), 
473: Molecular Magnets, Springer-Verlag, Wien, 2003,
474: special Edition of Monatshefte f{\"u}r Chemie/Chemical Monthly, 
475: {\bf 134} (2003), No. 2.
476: 
477: \bibitem{RMP71-1253}
478: W.~Kohn, 
479:  Nobel lecture: Electronic structure of matter -- wave functions and
480:  density functionals, 
481: Rev.~Mod.~Phys. {\bf 71} (1999) 1253.
482: 
483: \bibitem{AnChIE36-2482}
484: R.~W. Saalfrank, I.~Bernt, E.~Uller, and F.~Hampel, 
485:  Template-mediated self assembly
486:  of six- and eight-membered iron coronates, 
487: Angew.~Chemie -- Intern.~Edition {\bf 36} (1997) 2482.
488: 
489: \bibitem{InCh38-5879}
490: O.~Waldmann, J.~Sch{\"u}lein, R.~Koch, P.~M{\"u}ller, I.~Bernt, R.~W.
491:   Saalfrank, H.~P. Andres, H.~U. G{\"u}del, and P.~Allenspach, 
492:  Magnetic anisotropy of two cyclic hexanuclear {Fe(III)} clusters 
493:  entrapping alkaline ions,
494: Inorg.~Chem. {\bf 38} (1999) 5879.
495: 
496: \bibitem{our_XPS}
497: For a preliminary report, see cond-mat/0306406, to be published in
498:   J.~Phys.~Chem..~Solids; a more detailed discussion will be reported
499:   elsewhere.
500: 
501: \bibitem{JPCM14-2745}
502: J.~M. Soler, E.~Artacho, J.~D. Gale, A.~Garc{\'{\i}}a, J.~Junquera,
503: P.~Ordej{\'o}n, and D.~S{\'a}nchez-Portal, 
504:  The {SIESTA} method for \emph{ab initio} order-{$N$} materials simulation, 
505: J.~Phys.:~Condens.~Matter {\bf 14} (2002) 2745; 
506: see {\tt http://www.uam.es/siesta}.
507: 
508: \bibitem{PRB41-7453}
509: M.~R. Pederson and K.~A. Jackson, 
510:  Variational mesh for quantum-mechanical simulations, 
511: Phys.~Rev.~B {\bf 41} (1990) 7453; 
512: %
513: %\bibitem{PRB42-3276}
514: K.~Jackson and M.~R. Pederson, 
515:  Accurate forces in a local-orbital approach
516:  to the local-density approximation, 
517: Phys.~Rev.~B {\bf 42} (1990) 3276;
518: see {\tt http://cst-www.nrl.navy.mil/$\sim$nrlmol}.
519: 
520: \bibitem{PRL77-3865}
521: J.~P. Perdew, K.~Burke, and M.~Ernzerhof, 
522:  Generalized gradient approximation made simple, 
523: Phys.~Rev.~Lett. {\bf 77} (1996) 3865. 
524: 
525: \bibitem{PRB64-235111}
526: J.~Junquera, {\'O}.~Paz, D.~S{\'a}nchez-Portal, and E.~Artacho, 
527:  Numerical atomic orbitals for linear scaling, 
528: Phys.~Rev.~B {\bf 64} (2001) 235111;
529: %
530: %\bibitem{JPCM8-3859}
531: D.~S{\'a}nchez-Portal, E.~Artacho, and J.~M. Soler, 
532:  Analysis of atomic orbital basis sets from the projection 
533:  of plane-wave results,
534: J.~Phys.:~Condens.~Matter {\bf 8} (1996) 3859.
535: 
536: \bibitem{PRA60-2840}
537: D.~Porezag and M.~R. Pederson, 
538:  Optimization of {G}aussian basis sets for density-functional calculations, 
539: Phys.~Rev.~A {\bf 60} (1999) 2840.
540: 
541: \bibitem{PSSB217-197}
542: M.~R. Pederson, D.~V. Porezag, J.~Kortus, and D.~C. Patton, 
543:  Strategies for massively parallel local-orbital-based electronic 
544:  structure methods, 
545: Phys. stat. sol. (b) {\bf 217} (2000) 197.
546: 
547: \bibitem{PRB60-9566}
548: M.~R. Pederson and S.~N. Khanna, 
549:  Magnetic anisotropy barrier for spin tunneling in
550:  {Mn$_{12}$O$_{12}$} molecules, 
551: Phys.~Rev.~B {\bf 60} (1999) 9566.
552: 
553: \bibitem{Fe-clus}
554: A.~V. Postnikov, P.~Entel, and J.~M. Soler, 
555:  Density functional simulation of small {Fe} nanoparticles, 
556: Eur.Phys.J.~D {\bf 25} (2003) 261. 
557: %{\tt http://arXiv.org/abs/cond-mat/0109540}.
558: 
559: \bibitem{JPF14-L129}
560: K.~Schwarz and P.~Mohn, 
561:  Itinerant metamagnetism in {YCO$_2$}, 
562: J.~Phys.~F: Metal Phys. {\bf 14} (1984) L129.
563: 
564: \bibitem{LDA+U}
565: V.~I. Anisimov, F.~Aryasetiawan, and A.~I. Lichtenstein, 
566:  First-principles calculations of the electronic structure and spectra 
567:  of strongly correlated systems: the {LDA+$U$} method, 
568: J.~Phys.:~Condens.~Matter {\bf 9}) (1997) 767.
569: 
570: \bibitem{PRB37-9919}
571: A.~Svane and O.~Gunnarsson, 
572:  Localization in the self-interaction-corrected
573:  density-functional formalism, 
574: Phys.~Rev.~B {\bf 37} (1988) 9919;
575: %
576: %\bibitem{PRL65-1148}
577: %A.~Svane and O.~Gunnarsson, 
578:  Transition-metal oxides in the
579:  self-interaction-corrected density-functional formalism, 
580: Phys.~Rev.~Lett. {\bf 65}~(1990) 1148.
581: 
582: \bibitem{PRB65-184435}
583: D.~W. Boukhvalov, A.~I. Lichtenstein, V.~V. Dobrovitski, M.~I. Katsnelson,
584: B.~N. Harmon, V.~V. Mazurenko, and V.~I. Anisimov,
585:  Effect of local {C}oulomb interactions on the electronic structure
586:  and exchange interactions in {Mn$_{12}$} magnetic molecules,
587: Phys.~Rev.~B {\bf 65} (2002) 184435.
588: 
589: \bibitem{Ruiz-private}
590: E.~Ruiz~(eliseo.ruiz@qi.ub.es), private communication.
591: 
592: \bibitem{JCompCh24-982}
593: E.~Ruiz, A.~Rodr{'{\i}}guez-Fortea, J.~Cano, S.~Alvarez, and P.~Alemany, 
594:  About the calculation of exchange coupling constants in polynuclear 
595:  transition metal complexes, 
596: J.~Comput.~Chem. {\bf 24} (2003) 982.
597: 
598: \bibitem{PRL86-3400}
599: J.~Kortus, C.~S. Hellberg, and M.~R. Pederson, 
600:  Hamiltonian of the {V$_{15}$} spin system from first-principles 
601:  density-functional calculations, 
602: Phys.~Rev.~Lett. {\bf 86} (2001) 3400.
603: 
604: \end{thebibliography}
605: 
606: \end{document}
607: