1: %

2: %	REVTeX 4

3: %	Global Settings

4: % twocolumn,showpacs,preprint,eqsecnum,aps,eqsecnum,aps,draft,prb

5: %

6: \documentclass[prl,twocolumn,showpacs,superscriptaddress,floatfix]{revtex4}

7: %\documentclass[prl,preprint,showpacs,superscriptaddress,floatfix]{revtex4}

8: 

9: % 

10: %	Packages

11: %

12: \usepackage{amsmath}			% rozsirenie math-prostredia

13: \usepackage{amssymb}			% rozsirenie math-fontov

14: \usepackage{graphicx}			% input obrazkov

15: \usepackage{epic}

16:   \graphicspath{{./Eps/}}

17: %\usepackage{dcolumn}			% zarovnavanie v tabulkach

18: 					% na des. bodku

19: 

20: %

21: %	Bibliography style path

22: %

23: %\bibliographystyle{d:/Pavel/Misc/BibTeX/MyBibStyle}

24: 

25: %

26: %	My macro

27: %

28: \newcommand{\ua}{$\uparrow$}

29: \newcommand{\da}{$\downarrow$}

30: \newcommand{\singlet}{\ua \da}

31: \newcommand{\triplet}{\ua \ua}

32: \newcommand{\tlt}{\tilde t}

33: 

34: \newcommand{\myvec}[3]{ 

35: \left(

36:   \begin{array}{c}

37:    #1 \\

38:    #2 \\

39:    #3

40:   \end{array}

41: \right)

42: }

43: 

44: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

45: %	document begin

46: %

47: \begin{document}

48: 

49: \title{Fermion nodes and nodal cells of noninteracting and interacting fermions

50: }

51: 

52: \author{Lubos \surname{Mitas}}

53: %\email{lmitas@ncsu.edu}

54: \affiliation{Department of Physics, North Carolina State University, Raleigh,

55: NC 27695}

56: 

57: 

58: \date{\today}

59: 

60: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

61: %	Abstract

62: %

63: \begin{abstract}

64: A fermion node is subset of fermionic configurations for which a real wave 

65: function vanishes due to the antisymmetry and the node divides the configurations

66: space into compact nodal cells (domains). We analyze the properties of fermion 

67: nodes of fermionic ground state wave functions for a number of systems. For several

68: models we demonstrate that noninteracting spin-polarized fermions in dimension two 

69: and higher have closed-shell ground state wave functions with the minimal two nodal

70: cells for any system size and we formulate a theorem which sumarizes this result. 

71: The models include periodic fermion gas, fermions on the surface of a sphere, 

72: fermions in a box. We prove the same property for atomic states with up to $3d$ 

73: half-filled shells. Under rather general assumptions we then derive that the same 

74: is true for unpolarized systems with arbitrarily weak interactions using 

75: Bardeen-Cooper-Schrieffer (BCS) variational wave function. We further show that pair 

76: correlations included in the BCS wave function enable singlet pairs of particles 

77: to wind around the periodic box without crossing the node pointing towards the 

78: relationship of nodes to transport and many-body phases such as superconductivity.

79: Finally, we point out that the arguments extend also to fermionic temperature

80: dependent/imaginary-time density matrices. The results reveal fundamental properties 

81: of fermion nodal structures and provide new insights for accurate constructions of 

82: wave functions and density matrices in quantum and path integral Monte Carlo methods.

83: \end{abstract}

84: 

85: \pacs{02.70.Ss, 03.65.Ge}

86: %\pacs{61.46.+w, 73.22.-f, 36.40.Cg}

87: %\keywords{Suggested keywords}

88: 

89: \maketitle

90: 

91: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

92: %

93: %\section{Introduction}

94: \section{\label{sec:level1} Introduction}

95: %The properties of fermionic many-body systems belong to the most

96: %fascinating challenges in condensed matter quantum physics.

97: Let us consider a system of fermions described 

98: by a real wave function $\Psi(R)$ where $R$ denotes 

99: fermions coordinates 

100:  $R=({\bf r}_1, ...,{\bf r}_N)$. Due to the antisymmetry, there

101: exists a subset of fermion configurations for which the 

102: wave function vanishes and

103: this subset is called the fermion node.

104: The fermion node can be  

105: implicitly defined by 

106: $\Psi(R)=0$, assuming that the nodal set does not include

107: configurations for which the wave function vanishes because

108: of other reasons, eg,

109: boundary conditions.

110: In general, for $N$  spin-polarized fermions in  

111:   a $d$-dimensional space,  the fermion node is

112:  a $(dN-1)-$dimensional manifold

113: (hypersurface). 

114: It is a well-known fact that for $d>1$ the

115: antisymmetry alone does not specify 

116: fermion nodes completely. This is not difficult to understand

117: since antisymmetry

118: fixes only lower-dimensional 

119: coincidence hyperplanes with dimensionalities $(dN-d)$.

120: Therefore the fermion nodes and their properties are determined   

121: by interactions and many-body effects.

122: 

123: The fermion nodes play an important role in 

124:  quantum Monte Carlo (QMC) methods which belong to the

125:  most promising and productive

126: approaches for studying quantum many-body

127: systems.

128: Let us briefly describe the basic idea of QMC and its relationship 

129: to fermion nodes. 

130: Consider a  Hamiltonian $H$ and  a

131: trial function $\Psi_T(R)$ which approximates the ground state

132: $\Psi_0(R)$ of $H$ within a given

133: symmetry sector. It is straightforward to show that 

134: $\lim_{\tau\to\infty}\exp(-\tau H)\Psi_T\propto \Psi_0$ 

135: where $\tau$ is a real parameter 

136: (imaginary time). This imaginary time projection can be conveniently

137: carried out by simulating a stochastic process 

138: which maps onto 

139: the imaginary time  Schr\"odinger equation.   

140: The wave function is represented by an

141: ensemble  of $R$-space sampling points which are propagated

142:  according to  $G(R,R', \tau)=

143: \langle R |\exp(-\tau H)|R'\rangle$ and for large $\tau$

144: the ensemble becomes distributed according to $\Psi_0$.

145: Unfortunately, the straightforward application of this idea

146: to fermionic systems encounters a fundamental complication

147: in the fermion sign problem \cite{qmchistory, hammond, jbanderson}.

148: %In the simplest extension of the QMC

149: %approach to fermionic wave functions

150: The fermion sign problem makes QMC studies of large fermionic systems

151: difficult since the statistical errors 

152: of fermionic expectation values grow exponentially in the projection

153: time $\tau$ and also in the number of particles $N$. 

154: %The difficulty stems from the fact that, in general,

155: %probabilistic methods are difficult to apply to

156: %%distributions which are not strictly

157: %nonnegative and finds and one encounters an

158: %exponential loss the fermionic "signal" on the top

159: %of bosonic "noise".

160: 

161:  One possibility for avoiding the fermion sign problem

162: is to employ the fixed-node approximation 

163: which restricts the fermion node

164:  of the solution $\Psi_0(R)$

165: to be identical to the fermion

166: node of an appropriate trial function $\Psi_T(R)$.

167:  The fixed-node approximation introduces an energy bias which

168: scales as the square of the difference between the exact and approximate

169: nodes and therefore the accuracy of 

170: fermion nodes becomes very important. In particular,

171: for the exact node one can obtain the exact energy in computational

172: time proportional to a low-order polynomial in $N$. The fixed-node

173: approximation has been very successful even for rather approximate

174: nodes of 

175: commonly used trial wave functions which are based on

176: Hartree-Fock (HF) determinants or a few-determinant post-HF expansions.

177: %The approximate nodes of these trial functions

178: % allowed to obtain a number of valuable results which

179: %become established benchmarks

180: %and produced predictions which would be difficult to obtain by

181: %other approaches. Let us mention calculations of homogeneous electron gas,

182: %quantum liquids, lattice models and electronic structure of

183: %atoms, molecules, surfaces, solids and other systems

184: % over the past two decades

185: %% \cite{qmchistory,hammond,qmcrev}.

186: %Quantum many-body systems belong to the most intriguing and

187: %exciting challenges in physics. 

188: %Perhaps the most successful approach to investigate the 

189: %properties of quantum systems in recent years has been 

190: %the quantum Monte Carlo method. 

191: The fixed-node QMC electronic structure calculations

192: using HF or post-HF nodes usually recover

193: between 90 and 95 \%

194: of the correlation energy 

195: and energy differences 

196: agree with experiments typically within a few percent.  

197: Such encouraging results have been observed across many 

198: systems such as atoms, molecules, clusters and solids \cite{qmcrev}.

199: It is remarkable that

200: QMC methods have enabled us to "corner" the correlation energy problem

201: into the last few percent of the correlation energy using algorithms which 

202: polynomial scaling.

203: 

204: %Perhaps the most unique aspect of the QMC method is that

205: %it enables us to combine the known analytical properties

206: %of wave functions, such as 

207: %electron-electron cusps, with the

208: %robustness of stochastic approaches and provides thus 

209: %a unique capability to attack challenging quantum many-body

210: %problems. In addition, so far there is no fundamental obstacle

211: %on improving the wave functions and finding better functional

212: %%forms such as Bardeen-Cooper-Schrieffer pairing wave functions, 

213: %pfaffians  and backflow wave functions. One gets not only more accurate

214: %%results which in some cases start to reach the spectroscopic 

215: %accuracy but also revealing insights into the character of many-body effects.

216: 

217: %In this approach one forces the nodes of the solution (ie, places whwer

218: %of configurations for which the wave function vanishes due to the 

219: %antisymmetry) of the solution to be identical to the fermion

220: %node of the trial wave function $\Psi_T$. 

221: 

222: %It is now well

223: %understood that if there are any near-degeneracies present one needs to include 

224: %the corresponding excitations into the determinantal part of the wave function 

225: %post-Hartree-Fock methods such as Configuration Interaction or similar

226: %multi-determinant approaches. Although the analysis of the wave functions 

227: %to find out the important excitations might be tedious once this is done the

228: %%fixed-node QMC is very effective. 

229: 

230: Unfortunately,

231: for many problems even a few percent of the correlation energy can be 

232: significant. Typical

233: examples are  transition metal systems where the fixed-node error

234: can be of the order of several eVs. 

235: It is therefore clear that better understanding of fermion nodes

236: could be an important step forward for the QMC methodology and beyond.

237:  In addition, the fermion nodes are

238: related to other physical quantities and could shed light on other many-body phenomena 

239: which currently are not completely understood.

240: %fermion nodes to other physical quantities and effects which are not completely

241: %understood, some of which will be addressed also in this paper. 

242: 

243: %{\it Exact nodes.}

244: %So far our knowledge about the properties of fermion nodal hypersurfaces is quite limited. 

245: In order to introduce the basic properties of the fermion nodes 

246: and to illustrate the problems 

247: involved we will first mention a few unique cases of interacting systems

248: for which the nodes are known exactly.

249: These examples include a few two- and three-electron atomic states, namely,

250: %It is interesting that

251: %the exact nodes are known for a few special cases such as 

252: triplets of He atom  $^3S(1s2s)$, $^3P(2p^2)$

253: \cite{dariohe}

254: and the exact node of a three-electron atomic state

255:    $^4S(p^3)$ \cite{bajdichnode}. 

256:  The wave functions of these

257:  high symmetry states can be parametrized by appropriate 

258: coordinate  maps in which the node is described

259: by a single variable. For example, in the case of He triplet $^3S(1s2s)$ state the 

260: corresponding "nodal coordinate" is 

261: $\cos\beta={\bf r}^+_{12}\cdot {\bf r}_{12}/( r^+_{12} r_{12})$ where

262: ${\bf r}^+_{ij}={\bf r}_{i}+{\bf r}_{j} $, ${\bf r}_{ij}={\bf r}_{i}-{\bf r}_{j} $.

263: For the triplet $^3P(2p^2)$ the relevant variable is

264: $\cos\omega'={\bf z}_0\cdot ({\bf r}_{1}\times {\bf r}_{2})$ assuming 

265: that the

266: $P$ state is oriented along the $z$-axis which is specified by

267: the unit vector ${\bf z}_0$ \cite{dariohe,bajdichnode}.

268: For the quartet state 

269:  $^4S(p^3)$

270: the node is captured by the variable 

271: $\cos\omega={\bf r}_{1}\cdot ({\bf r}_{2}\times {\bf r}_{3})$.

272: The node in these 

273: systems is encountered whenever the corresponding variable,

274:  $\cos\beta,\cos\omega'$

275: or $\cos\omega$,

276: vanishes.

277: 

278: The nodal surface divides the space of fermion configurations into

279: nodal cells, sometimes also called domains.

280: (The name nodal cell was introduced in the previous paper \cite{davidnode}.

281: In algebraic geometry and topology

282:  nodal cells are usually called nodal domains, see, for example,

283: Ref. \cite{berger}. We will use

284: both of these two expressions interchangeably.)

285:  The few known examples of exact nodes mentioned in the preceding paragraph help to illustrate an 

286: important point, which has been conjectured for the fermionic ground states

287: for some time \cite{davidnode}, namely,

288: that the ground state node divides the configuration space into

289: the minimal number of two nodal cells: a "plus" cell with $\Psi>0$ and

290:  a "minus" cell with $\Psi<0$.  From the node equation 

291: $\cos\beta=0$  of the $^3S(1s2s)$ state we easily find

292: that the "plus" and "minus" domains are given by the 

293: conditions $r_1>r_2$ and $r_1<r_2$, respectively.

294: For the node given by $\cos\omega'=0$ the nodal domains are given

295: by the orientation of ${\bf z}_0, {\bf r}_{1},{\bf r}_{2}$: the three 

296: vectors are either left- or right-handed, corresponding to either "plus" or "minus"

297: cell, respectively. Similarly, for the node $\cos\omega=0$

298: the domains are given by the left- or right-handedness of

299: the vectors ${\bf r}_{1},{\bf r}_{2}, {\bf r}_{3}$.

300: The minimal number of

301: two nodal domains

302: was found  

303: also for $2D$ and $3D$ noninteracting spin-polarized homogeneous

304: electron gas with periodic boundary conditions for up to 200 particles using

305: a numerical proof \cite{davidnode}.

306: 

307: Understanding the fermion nodes and their properties has become a 

308: challenge which might help to advance both practical calculations

309: but also open a deeper insights into the properties of fermionic systems.

310: One can envision two key problems:

311: 

312: a) {\it Topology of the fermion nodes, ie, how many nodal cells

313: are actually present} since this is of high importance for the fixed-node 

314: QMC approaches. Note that if the number 

315: of nodal cells is incorrect (typically {\it higher} than it should be since 

316: mean-fields such as HF 

317: have tendency to divide the configurations space into too many domains) then the QMC sampling

318: around the artificial nodes will be very sparse. This could lead to 

319:  large statistical fluctuations from poor sampling, 

320: and, possibly, to an 

321: effective non-ergodicity due to the finite-time projection time in practical calculations.

322: 

323: b) Once the topology is correct, {\it the accuracy of the manifold 

324: shape becomes important.} This is an area

325: where our insights are particularly limited since the exact nodes for 

326: large interacting systems are virtually unknown except

327: for a few-particle special cases mentioned above. 

328: 

329: In our recent paper \cite{mitasshort},

330: we have made some encouraging steps forward in trying to understand the 

331: topological issues

332: and we have 

333: analytically derived a number of new results regarding the number of nodal domains.

334: In particular, we have shown that spin-polarized 

335: closed-shell ground states of noninteracting harmonic 

336: fermions in $d=2$ and higher 

337: have the minimal number of two nodal domains.

338: % for {\it an arbitrary size of the system}.

339: We have proved the same for

340: %spin-unpolarized harmonic fermions with interactions, 

341: spin-polarized atomic states with several electrons,

342: both for noninteracting and HF wave functions.

343: We have also shown that

344: by imposing additional symmetries one can generate more than two

345:  nodal domains but that

346: interactions  can relax this "nodal degeneracy" to the minimal number of two 

347: domains such as in the case of the $^4S(1s2s3s)$ atomic state.

348: 

349: For noninteracting spin-unpolarized systems, ie, with both spin channels

350: occupied, the number of nodal cells is four since the wave function is

351: a product of spin-up and spin-down Slater determinants ($2\times 2=4$).

352: (Here and later on we assume that the Hamiltonian does not include spin-dependent

353: terms so that the particle spins are conserved. 

354: The wave function is then a product of two determinants which depend 

355: only on the spatial degrees of freedom \cite{qmcrev}.)

356: In the last few years, studies of a few-particle systems have revealed

357: that interactions and many-body effects

358: do affect the nodal topologies

359: and can change the number of nodal cells \cite{dariobe,cmt28}.

360: In particular, for the case of Be atom it has been found that

361: the noninteracting/HF four nodal cells of the singlet ground state

362: change to the minimal number of two due to the electron correlation

363: \cite{dariobe} and qualitatively the same has been observed

364: in  other systems \cite{cmt28}. 

365: In our recent paper \cite{mitasshort} we have found that this is a rather

366: generic property of ground states in interacting systems.

367:  With some conditions, 

368:  we have explicitly demonstrated that interactions lift

369: the "nodal cell degeneracy" in spin-unpolarized systems

370: and smooth out the four noninteracting cells

371: into the minimal two. We have shown this for

372:  $2D$  harmonic fermions in closed-shell states {\it of arbitrary size}

373: using a variational Bardeen-Cooper-Schrieffer wave function.

374: 

375: In this work we further advance these ideas.

376: % and apply it to more types of systems.

377: We analyze the fermion nodes in a number of other paradigmatic fermionic

378: models: homogeneous electron gas with periodic

379: boundary conditions, particles in an infinite well and on the surface of a sphere.

380: For all these systems

381: we prove that in the spin-polarized noninteracting closed-shell ground state the fermion nodes 

382: divide the configuration space into minimal two cells for arbitrary number of

383: particles.

384: We also extend our previous proof for spin-unpolarized systems

385: demonstrating 

386: that interactions smooth out multiple nodal cells of noninteracting/mean-field

387: wave functions into the minimal two for more systems such as homogeneous

388: electron gas and $3D$ harmonic oscillator. 

389: %We show that by introducing the repulsive/attractive interaction

390: %this property does not change.

391: %For spin-unpolarized or partially polarized systems we show that in

392: %the noninteracting case the number of nodals cells is trivially four since

393: %the wave function is a product of determinants in

394: %the spin-up and -down channels. This is also the case of

395: %wave functions for unpolarized systems in Hartree-Fock(HF) theory.

396: % We show that arbitrarily small interaction removes

397: %the spin-up and -down degeneracy and the consequence of this is that the

398: %nodal partitioning changes from four to two.

399: These results

400:  contribute to our understanding

401: of the fermion nodes and their impacts on the accuracy 

402: of wave functions with direct implications for the QMC methods. 

403: 

404: In the last sections we show how in periodic

405: spin-unpolarized interacting fermion gas the pairs of particles 

406: can wind around the box without crossing the node

407: what points towards the connection of nodes 

408: to the transport and to the 

409: existence of other quantum phases such as superconductivity.

410:  Finally, we then

411: generalize the results to the temperature density matrices 

412: with the implications

413: for path integral Monte Carlo methods \cite{davidhe}.

414: 

415: 

416: \section{II. General properties of fermion nodes.}

417: 

418: Let us introduce the basic properties of fermion nodes

419: %The fermion node of a wave function $\Psi(R)$

420: %is implicitly defined as 

421: %the set of configurations for which $\Psi(R)=0$ (here we assume that we have omitted

422: %from the nodal set the configurations for which the wave function vanishes because of 

423: %other reasons such as singular potential or boundary conditions). Assuming that we have 

424: %%$N$ spin-polarized fermions 

425: % in $d$-dimensional space the node is a $(dN-1)-$dimensional manifold 

426: %(hypersurface). 

427: as they were studied

428: by Ceperley \cite{davidnode} some time ago. 

429: 

430: a) Nondegenerate ground states wave functions fulfill the so-called 

431: tiling property.  Let us define a nodal cell

432:  $\Omega(R_0)$ as a subset of configurations which can be reached

433: from the point $R_0$ by a continuous path without crossing the node.

434: %(The name nodal cell was introduced in the previous paper \cite{davidnode}.

435: %In algebraic geometry and topology  

436: % nodal cells are called nodal domains, see, for example,

437: %Ref. \cite{berger}. We will use 

438: %both of these two expressions interchangeably.)

439: The tiling property

440: says that by applying all possible particle permutations to an arbitrary

441: nodal cell

442: one covers the complete configuration

443: space. Note that this {\em does not} specify how many nodal cells are there.

444: 

445: b) If $m$ nodal surfaces cross then the angle of crossing is

446: $\pi/m$.

447:  Furthermore, the symmetry of the node is the same as the

448: symmetry of the state.

449: 

450: 

451: c) 

452: It is possible to show that there are only two nodal cells using the following

453: argument based on triple exchanges. Let us first 

454: introduce the notion of particles connected by triple exchanges. 

455: % (continuously connected) by the following construction. Consider

456: % a spin-polarized system

457: %with the wave function $\Psi(R)$ where $R=({\bf r}_1, ..., {\bf r}_N)$ are particle

458: %coordinates.

459: We will call the three particles $i,j,k$  {\it connected}

460: if there exists a triple exchange path $i\to j, j\to k, k\to i$ which does 

461: not cross the node. If more than three particles are connected then  

462: they can form a connected cluster. An example of 

463: six particles connected into a single cluster 

464: is sketched as follows: 

465: \begin{picture}(12,7.5)(6,4.5)

466: \put(2,5){$\bullet$}

467: \put(6,2){$\bullet$}

468: \put(6,8){$\bullet$}

469: \put(10,5){$\bullet$}

470: \put(14,2){$\bullet$}

471: \put(14,8){$\bullet$}

472: \drawline(3.5,7)(7.7,4)(7.7,10)(3.5,7)

473: \drawline(7.7,10)(15.8,4)

474: \drawline(7.7,4)(15.8,10)(15.8,4)

475: \end{picture}.

476: If there exists a point $R_t$ such that {\it all} the 

477: particles are connected  into a single cluster then $\Psi(R)$ has only two nodal cells.

478: This can be better understood once we realize the following two facts.

479: First, any triple 

480: permutation can be written as two pair permutations. Therefore the connected

481: cluster of triple permutations enables to realize any even parity permutation 

482: without crossing the node. That exhausts all permutations which are available

483: for cell of one sign since the  wave function is invariant to even parity permutations. 

484: Second, the tiling property implies that once the particles are connected for $R_t$

485: the same is true for the entire cell. By symmetry, the same arguments

486: apply 

487: to the complementary "minus" cell which correspond to the odd permutations. 

488: More details on this property can be found in the original Ref. \cite{davidnode}.

489: 

490: % We show that arbitrarily small interaction removes

491: %the spin-up and -down degeneracy and the consequence of this is that the

492: %nodal partitioning changes from four to two.

493: 

494: 

495: \section{\label{sec:sp} III. Noninteracting spin-polarized fermions.}

496: 

497: \subsection{III.a. Homogeneous electron gas. }

498: 

499: %{\it Noninteracting homogeneous electron gas with 

500: %periodic boundary conditions.} 

501: 

502: We consider a

503: system of spin-polarized noninteracting

504: fermions in a periodic box  in $d$ dimensions. The  

505: spatial coordinates are rescaled by the box size

506: so that we can use dimensionless variables and the box

507: becomes

508: $(-\pi,\pi)^d$. 

509: 

510: We first analyze the fermion nodes for $d=1$ since the result will be

511: useful in subsequent derivations. We consider 

512: a system with $N=(2k_F+1)$ particles. In our $1D$ dimensionless

513: units the Fermi momentum becomes an integer,

514: $k_F=1,2 ...$.  The one-particle occupied states are written 

515: as $\phi_n(x)=e^{inx},$ $n=0,\pm 1,...,\pm k_F$ 

516: and the spin-polarized ground state is given by

517: \begin{equation}

518: \Psi_{1D}(1,...,N)={\rm det}\left[ \phi_n(x_j)\right]

519: \end{equation}

520: %\begin{equation}

521: %{\rm det}\{1,e^{\pm ix}, e^{\pm i2x}, ..., e^{\pm ik_Fx}\}

522: %\end{equation}

523:  where

524: $x_j$ is the $j$-th particle coordinate and $j=1,...,N$.

525: We factorize the term 

526: $\exp(-ik_F\sum_jx_j)$ out of the determinant so that the 

527: Slater matrix elements become powers of $z_j=e^{ix_j}$. 

528: The resulting Vandermonde determinant can be 

529: written in a closed form and after some rearrangements we find 

530: \begin{equation}

531: \Psi_{1D}(1,...,N)=

532: e^{-ik_F\sum_jx_j}\prod_{j>k}(z_j-z_k)=\nonumber

533: \end{equation}

534: \begin{equation}

535: =\mu_0\prod_{j>k}

536: \sin(x_{jk}/2)

537: \label{eq:van1d}

538: \end{equation}

539: where 

540: $x_{jk}=x_j-x_k$ and $\mu_0$ is

541: a constant prefactor which is unimportant 

542: for our purposes.  (In the derivations below we will be using

543: the letter $\mu$ for denoting prefactors which are either

544: constants or nonnegative functions in particle coordinates

545: and therefore do not affect the nodes).  

546: %(In order 

547: %to simplify 

548: %the expressions

549: %we will be omitting similar constant prefactors

550: %whenever appropriate.) 

551: %Assuming that in the

552: %expression above the particle positions are all distinct,

553: The derived wave function has the following important properties.

554: 

555: a) The fermion nodes

556: appear at the particle coincidence points. This implies a well-known 

557: result that the ground state wave function in $1D$ have $N!$ nodal cells since any 

558: particle permutation  

559: requires crossing the node at least once. In addition, this also means that 

560: {\it any} fermion configuration which 

561: preserves the particle order is contained within the same nodal domain/cell.

562: %In particular, winding the particle around the box 

563: %without crossing any other particle keeps the configuration in the same

564: %nodal domain. 

565: 

566: % Note that whenever a particle

567: %of some of the terms might need adjustment so that the wave function is continuous.

568: %%For example, a particle with the label $i$ moves in $+x$ direction hits 

569: %$x_i=\pi$ and reenters as $x_i \to x_i-2\pi$

570: %so that the pair terms in the wave function transform to

571: %\begin{equation}

572: %\sin(x_{ij}/2-\pi)=-\sin(x_{ij}/2)=\sin(x_{ji}/2)

573: %\end{equation}

574: %implying exchange of $x_{ij} \to x_{ji}$ in all the

575: % pairs which involve the $i-$th particle.

576: 

577: 

578: b) The wave function is invariant to

579: translations and cyclic exchanges of particles. 

580: Due to the periodic boundary conditions

581: this includes also winding the system

582: around the periodic box, ie, the translation of all particles

583: by $2\pi$. We can formalize 

584: this by introducing an operator $T^{x}_a$ which translates all the particles 

585: along the $x$-axis

586: as $x_j \to x_j+a$ so that the translation invariance can be written as

587: \begin{equation}

588: T_a^x \Psi_{1D}(1,...,N)= \Psi_{1D}(1,...,N)

589: \label{eq:inv1}

590: \end{equation}

591: 

592: Similarly, the wave function  is invariant to 

593: cyclic exchange of all the particles given by

594: $j \to j+1, j=1, ..., N$ and $N+1 \to 1$.

595:  This action is carried out   

596: by $C_{+1}^{x}$ operator where the notation means the 

597: exchange by one site is in the $+x$-direction. 

598: Clearly, the inverse operator

599: $[C_{+1}^x]^{-1}=C_{-1}^x$. Note that here we have assumed that $N$ is odd, in agreement with our 

600: definition.  The invariance holds only for $N$ odd since then

601: the cyclic exchange is equivalent

602: to an even number of pair exchanges. For the sake of completeness, we consider

603: also $N$ even, when the cyclic exchange

604: can be replaced by an odd number of pair exchanges, resulting

605: in the wave function sign flip.

606: In general, for the cyclic exchanges we can therefore write  

607: \begin{equation}

608: C_{\pm 1}^x\Psi_{1D}(1,...,N)=(-1)^{N+1}\Psi_{1D}(1,...,N)

609: \label{eq:inv2}

610: \end{equation}

611: 

612: c) Assuming the particle positions are all distinct, 

613: the cyclic exchange {\it path} can be chosen in such 

614:  a way that it

615: {\it does not cross the node}. 

616: %It is not difficult to imagine 

617: %that the particles move in a concerted manner

618: %and never cross each other along the whole cyclic exchange path. 

619: This is easy to accomplish by maintaining 

620: finite distances between the particles along the path trajectory. 

621: Let us parametrize the exchange path by a parameter $t$ so that the

622: path starts at

623:  $t=0$, the path is completed at $t=1$ and the  

624:  exchange path operator is then denoted as $C^x_{+1}(t)$ with $0\leq t\leq 1$.

625: The fact that the path  

626: does not cross the node (ie, the path is contained within the same 

627: nodal cell) can be then written as

628: \begin{equation}

629: |C_{+1}^x(t)\Psi_{1D}(1,...,N)|>0, \; 0\leq t \leq 1

630: \label{eq:inv3}

631: \end{equation}

632: where, of course, $N$ is assumed to be odd.

633: 

634: Let us now derive the wave functions and generalize the results for 

635: fermions in $2D$. The one-particle states in $2D$ are  

636: $\phi_{nm}(x,y)=e^{i(nx+my)}$.  

637: The states are occupied up to the Fermi momentum $k_F$ so that

638: we have $n^2+m^2\leq k_F^2$, where $k_F$ in $2D$ is not necessarily an integer. 

639: Similarly to our previous paper \cite{mitasshort},

640: we show that the spin-polarized electron gas  

641: for closed-shell

642: ground

643: states have only two nodal cells.

644: %The number of particles is given by

645: %$N=2M(M+1)+1$ where $M$ labels the closed-shell states with the

646: %Fermi surface given by (a rotated) square see Fig. .

647: %$M$ denotes also the highest power of $e^{ix}$ in one-particle states

648: %and we have then $(2M+1)$ lines of particles (vertically or horizontally).

649: 

650: \begin{figure}[ht]

651: \centering

652: \begin{tabular}{ccc} 

653: \includegraphics[width=1.1in,clip]{fig1A.eps} &

654: \includegraphics[width=1.1in,clip]{fig1B.eps} &

655: \includegraphics[width=1.1in,clip]{fig1C.eps} \\

656: (a) & (b) & (c)

657: \end{tabular}

658: \caption{Positions of five fermions in the $2D$

659:  periodic box $(-\pi,\pi)^2$. (a) Particles aligned horizontally,

660: along the lines $y=\zeta_0,\zeta_1,\zeta_2$.

661: (b) Particles aligned vertically, along the lines  $x=\xi_0,\xi_1,\xi_2$.  (c) Particles aligned 

662: in both directions, positioned on a lattice.}

663: %See the text for details. }

664: \label{fig:five2d}

665: \end{figure}

666: 

667: The proof is by induction, therefore 

668: let us first consider $k_F=1$ with five particles

669: occupying $\{1,e^{\pm ix},

670: e^{\pm iy}\}$ one-particle states. 

671: We place the particles as in Fig.\ref{fig:five2d}a so that

672: the coordinates are given by 

673: ${\bf r}_1=(x_1,\zeta_1)$, ${\bf r}_2=(x_2,\zeta_1)$, ${\bf r}_3=(x_3,\zeta_1)$,

674:  ${\bf r}_4=(x_3,\zeta_2)$,  ${\bf r}_5=(x_5,\zeta_0)$.

675: By eliminating the terms with $\zeta_1$ 

676: the wave function can be factorized 

677: as follows  

678: \begin{equation}

679: \Psi_{2D}(1,...,5)=\left| \begin{array}{ccccc}

680:    1  &  1   &  1   &    1  &  1    \\

681:  e^{ix_1}  &  e^{ix_2}   &  e^{ix_3}   &

682:    e^{ix_4}  &  e^{ix_5} \\

683:  e^{-ix_1}  &  e^{-ix_2}   &  e^{-ix_3}   &

684:    e^{-ix_4}  &  e^{-ix_5} \\

685:  e^{i\zeta_1} &  e^{i\zeta_1} &

686: e^{i\zeta_1} &  e^{i\zeta_0} &

687: e^{i\zeta_2} \\

688:  e^{-i\zeta_1} &  e^{-i\zeta_1} &

689: e^{-i\zeta_1} &  e^{-i\zeta_0} &

690: e^{-i\zeta_2} \\

691: \end{array} \right|=\nonumber

692: \end{equation}

693: \begin{equation}

694: \mu_0\left| \begin{array}{ccc}

695:    1  &  1   &  1     \\

696:  e^{ix_1}  &  e^{ix_2}   &  e^{ix_3}  \\

697:  e^{-ix_1}  &  e^{-ix_2}   &  e^{-ix_3}  \\

698: \end{array} \right|

699: \left| \begin{array}{cc}

700:   e^{i\zeta_0}-e^{i\zeta_1} &

701: e^{i\zeta_2} -e^{i\zeta_1}\\

702:  e^{-i\zeta_0} -e^{-i\zeta_1} &

703: e^{-i\zeta_2} -e^{-i\zeta_1} \\

704: \end{array} \right|=\nonumber

705: \end{equation}

706: 

707: \begin{equation}

708: = \mu_0'\Psi_{1D}(1,2,3)\sin(\zeta_{10}/2)\sin(\zeta_{20}/2)\sin(\zeta_{21}/2)

709: \label{eq:five2dzeta}

710: \end{equation}

711: where $\mu_0,\mu_0'$ are irrelevant constant prefactors.

712: We have obtained a product of $1D$ wave function and

713: terms with distances between the

714: {\it lines} $y=\zeta_0,\zeta_1,\zeta_2$.

715: Note that analogous result can be found for the configuration

716: sketched in Fig. 1b so the particles are aligned 

717: in parallel to the 

718: $y$-axis with the wave function given by

719: \begin{equation}

720: \Psi_{2D}(1,...,5)=

721:  \mu_0\Psi_{1D}(5,2,4)\times \nonumber

722: \end{equation}

723: \begin{equation}

724: \times \sin(\xi_{10}/2)\sin(\xi_{20}/2)\sin(\xi_{21}/2)

725: \label{eq:five2dxi}

726: \end{equation} 

727: If the particles are aligned in both directions as in Fig. 1c

728: both expressions apply.

729: % and give the same result. 

730: From the derived wave function it is clear that the node is encountered

731: when particles lying on the same line reorder or 

732: when the lines of particles 

733: cross each other, eg, $\xi_1=\xi_0$.

734: %Note that for all three configurations in Fig. 1 the 

735: %particles occupy the {\it same } nodal domain since the configurations 

736: %can be obtained from each other by repositioning 

737: %the particles without crossing the node.

738: %This is easy to see because

739: %the repositioning can be done without any

740: %reordering of particles along $1D$ directions 

741: %and without any two lines

742: %of particles crossing each other. 

743: The wave functions for the configurations outlined 

744: in Fig.\ref{fig:five2d} possess an  

745: important property. Consider 

746: $\Psi_{1D}(...)$ in Eqs.  \ref{eq:five2dzeta},  \ref{eq:five2dxi} with

747:  groups of particles positioned on the corresponding lines.

748: Assuming the group of particles is allowed to move only along the given line,

749: one can consider this to be an effective $1D$ subspace.

750: The $1D$ wave function for such a subspace  

751: obeys all three conditions derived

752: for $1D$ case, ie, 

753:  Eqs. \ref{eq:inv1}, \ref{eq:inv2} and \ref{eq:inv3}. 

754: For example, for $\Psi_{1D}(1,2,3)$ which appears 

755:  in Eq. \ref{eq:five2dzeta} we have

756:  $|C_{+1}^x(t)\Psi_{1D}(1,2,3)|>0$. This is easy to check also analytically. 

757: We fix the coordinates in 

758:  Fig.1(a) as follows $x_1=-2\pi/3, x_2=0, x_3=2\pi/3$ 

759: and then we can 

760: carry out a "synchronized" cyclic exchange using the translation  

761: by $2\pi/3$ in $x$-direction. The 

762: wave function is constant 

763: along the translation/cyclic exchange path so that we have 

764: \begin{equation}

765: C^x_{+1}(t) \Psi_{1D}(1,2,3)=\Psi_{1D}(1,2,3), \;\; 0\leq t \leq 1

766: \end{equation}

767: because of the translational invariance.

768: 

769: Finally, we are ready to show that there are only two nodal cells in this 

770: five-particle ground state. 

771: Assume that the particles are positioned as in Fig.1c. 

772: Consider the following sequence of 

773: four exchanges, $C^x_{+1}C^y_{+1}C^x_{-1}C^y_{-1}$, where the operators act

774: from the right, ie, $C^y_{-1}$, acts first.  We denote this exchange symbolically as 

775: \begin{equation}

776: C_{\rightarrow\uparrow\leftarrow\downarrow}=C^x_{+1}C^y_{+1}C^x_{-1}C^y_{-1}

777: \end{equation}

778: The exchanges are performed only for  

779: the particles lying on the lines $\zeta_1,\xi_1$ in corresponding directions,

780: eg, $C_{\pm 1}^x$ acts on particles along  $y=\zeta_1$ while  $C_{\pm 1}^y$ 

781: cycles particles along $x=\xi_1$ . 

782: It is easy to check that this results in a triple exchange 

783: $423 \to 342$ while the particles 1 and 5 end up in their original

784: positions. It is therefore clear that particles 2,3,4 are connected by 

785: this triple exchange.

786: It is straightforward to show that 

787: $C^x_{-1} C^y_{-1} C^x_{+1}C^y_{+1}$ exchanges particles 1,2,5 so that 

788: we can conclude that all five particles are connected by triple exchanges    

789: into a single cluster

790: and there are only two nodal domains.

791: 

792: Now we need to extend the arguments and find the wave functions for

793: general cases 

794: with occupied states within the Fermi disk with an arbitrary $k_F$.

795: It is instructive to first derive 

796: the wave functions for $k_F=\sqrt{2}$ with 9 particles

797: occupying

798:  states $\{ 1,e^{\pm ix}, e^{\pm iy}, e^{i(\pm x \pm y)}\}$

799:  and 

800: for $k_F=2$ with 13 particles  with additional states 

801: $\{ e^{\pm i2x}, e^{\pm i2y}\}$.

802: This will enable us to understand how to  

803: write down the wave function

804: for a general case while avoiding rather tedious notations 

805: which would appear in the direct derivations.

806: % for a general case.

807: 

808: \begin{figure}[ht]

809: \centering

810: \begin{tabular}{ccc}

811: \includegraphics[width=1.35in,clip]{fig2A.eps} &

812: \quad &

813: \includegraphics[width=1.35in,clip]{fig2B.eps} \\

814: (a) & \quad & (b)

815: \end{tabular}

816: \caption{Positions of fermions in

817:  the $2D$ periodic box in real space. (a) Positions of the nine-fermion system.

818: (b) Positions of the 13-fermion system.

819: See the text for details. }

820: \label{fig:nine2d}

821: \end{figure}

822: 

823:  Let us position the particles into a pattern which mimics the lattice

824: of the occupied wave vectors in the reciprocal space, see Fig.\ref{fig:nine2d}.

825: Using appropriate algebraic arrangements we subsequently factorize the lines 

826: of particles as given by

827: \begin{equation}

828: \Psi_{2D}(1,...,9)= \nonumber

829: \end{equation}

830: \begin{equation}

831: =\mu_0\Psi_{1D}(1,2,3)\Psi_{2D}(4,...,9) \sin^3(\zeta_{21}/2) \sin^3(\zeta_{10}/2)=\nonumber

832: \end{equation}

833: \begin{equation}

834: =

835: \mu_0'\Psi_{1D}(1,2,3)\Psi_{1D}(4,5,6)\Psi_{1D}(7,8,9)\times \nonumber

836: \end{equation}

837: \begin{equation}

838: \times \sin^3(\zeta_{21}/2) \sin^3(\zeta_{10}/2)\sin^3(\zeta_{20}/2)

839: \end{equation}

840: where $\mu_0,\mu_0'$ are constants.

841: Similarly, for $k_F=2$ we can first factorize

842: the line with the largest number of particles on 

843: $y=\zeta_2$ (see Fig.\ref{fig:nine2d})

844: \begin{equation}

845: \Psi_{2D}(1,...,13)=

846: \mu_0\Psi_{1D}(1,...,5)\sin(\zeta_{02}/2)\sin^3(\zeta_{12}/2)

847: \times

848: \nonumber

849: \end{equation}

850: \begin{equation}

851: \times \sin^3(\zeta_{32}/2)\sin(\zeta_{42}/2)

852: \Psi_{2D}(6,...,13)\nonumber

853: \end{equation}

854: and 

855: then factorize out the particles lying on another line $y=\zeta_3$ to obtain

856: \begin{equation}

857: \Psi_{2D}(1,...,13)=

858: \mu_0\left[\prod_{j\neq 2}\sin^{n_j}(\zeta_{j2}/2)\right]

859: \left[\prod_{j\neq 2,3}\sin^{n_j}(\zeta_{j1}/2)\right]

860: \times

861: \nonumber

862: \end{equation}

863: \begin{equation}

864: \Psi_{1D}(1,...,5)\Psi_{1D}(6,7,8)\Psi_{2D}(9,...,13)

865: \end{equation}

866: where $n_j$ is the number of particles lying on the line $y=\zeta_j$.

867: We have obtained a product of $1D$ wave functions, 

868: terms with distances 

869: between the lines and the $2D$ wave function with lower number of particles

870: positioned in the same type of pattern is in Fig.\ref{fig:five2d}. Obviously, we can

871: further factorize  $\Psi_{2D}(9,...,13)$ using Eq.\ref{eq:five2dzeta} until

872: we end up 

873: with $1D$ factors only.

874: % At this point it has become

875: %clear that we can evaluate 

876: %the determinant by factorizing out the lines of particles

877: %in a recursive manner.

878: Using these insights into the recursive form of the wave function, 

879: for a general case with $M+1$ lines we can write

880: \begin{equation}

881: \Psi_{2D}(1,...,N)=\nonumber

882: \end{equation}

883: %\begin{equation}

884: %=\pm\Psi_{2D}(1,...,N/I_k)\Psi_{1D}(I_k)\prod_{j\ne k} \sin^{n_j}(\zeta_{jk}/2)

885: %\nonumber

886: %\end{equation}

887: \begin{equation}

888: =\mu_0\prod_{k=0}^{M}\left[\Psi_{1D}(I_k)\prod_{j>k } 

889: \sin^{n_j}(\zeta_{jk}/2)\right]

890: \end{equation}

891: where $I_k=i^{(k)}_1, ..., i^{(k)}_{n_k}$

892: denotes the labels of particles

893: lying on the line $y=\zeta_k$.

894: %The sign depends on the number of row exchanges

895: %and on the actual ordering of particles. 

896: In addition, analogous expression can be found if we factorize along the $x=\xi_j$

897: lines, the only difference being replacement $\zeta_{jk}$ by $\xi_{jk}$ 

898: and corresponding replacement of particle sets in $\Psi_{1D}$ wave functions.

899: %Note the translation

900: %invariance of Eq. \ref{eq:recur}.

901: 

902: 

903: \begin{figure}[ht]

904: \centering

905: \begin{tabular}{ccc}

906: \includegraphics[width=1.35in,clip]{fig3A.eps} &

907: \quad &

908: \includegraphics[width=1.35in,clip]{fig3B.eps} \\

909: (a) & \quad & (b)

910: \end{tabular}

911: \caption{Illustrations of the position patterns in $2D$ periodic box for the

912:  size increases from $k_F$ to $k_F+\Delta k_F$.

913: The particle layouts mimic the occupied states in the

914: reciprocal space, however, the particles are positioned in the real space. 

915: The additional particles are in grey/yellow.

916: (a) The additional particles positioned along $x-$ and $y-$axis directions.

917: (b)  The additional particles positioned along diagonals.

918: Lines $x=\xi_k$ and $y=\zeta_l$ are used for the proof that the particles $a,b,c$

919: are connected by a triple exchange.

920: See the text for details. }

921: \label{fig:indu2d}

922: \end{figure}

923: 

924: We are now ready for the induction step. Consider a 

925: spin-polarized closed-shell ground state

926: with a given $k_F$. For this wave function we assume that all the

927: particles are connected by the triple exchanges, ie, there are only two nodal domains.

928: Let us

929: increase $k_F \to k_F+\Delta k_F $ until the Fermi disk

930: includes the next unoccupied star of states $\phi_{nm}(x,y)$ for which

931: $k_F < (n^2+m^2)^{1/2} \leq k_F+\Delta k_F$. The system size increases

932: by the corresponding number of additional particles.

933: Assuming the particles are positioned in the real space 

934: in the same pattern as

935:  the occupied ${\bf k}$-points in reciprocal space (Fig.\ref{fig:indu2d}),

936:  the additional particles will appear at the 

937: borderline of the disk in the real space.

938: % (corresponding to the Fermi disk

939: %in the {\bf k}-space).

940: 

941: The two basic possibilities how the additional particles are positioned 

942: are  given in the Fig.\ref{fig:indu2d}.  

943: %These two are, however, essentially the same in the sense

944: %that the following arguments one how the additional particles 

945: %are connected by exchanges with the original particles will apply to both of them.

946: We need to show that these additional particles are connected 

947: to the original particles

948: by the triple exchanges. 

949:  This can be demonstrated by

950: the sequence of the four cyclic exchanges which we have

951: used above for the five particle case. 

952: It involves particles on

953:   the lines $x=\xi_k$ and $y=\zeta_l$ as schematically drawn in Fig.\ref{fig:indu2d}.

954: Consider the exchange

955: $C_{\rightarrow\uparrow\leftarrow\downarrow}=$

956: $C^x_{+1}C^y_{+1}C^x_{-1}C^y_{-1}$  where the cyclic exchanges in $x$ direction are

957: applied only to particles on the line $x=\xi_k$ and, similarly, 

958: the cyclic exchanges in $y$ direction are applied only to particles along $y=\zeta_l$. 

959:  Note the wave function is invariant to cyclic exchanges

960: since the number of particles along each line 

961: is odd for any closed shell state. It is then straightforward to find 

962: out that $C_{\rightarrow\uparrow\leftarrow\downarrow}$

963:  exchanges particles $a,b,c$ while the rest of the particles remains intact.

964:  Similar exchanges can be carried out for 

965: all additional particles. Finally, this shows

966: that the additional particles are connected 

967: to

968: the particles of the wave function with size $k_F$ and finalizes the 

969: proof.

970: 

971: The proof can be extended into $3D$ and higher dimension by positioning the particles onto

972: an appropriate $3D$ pattern which in real space mimics the occupied states in the Fermi

973: sphere in the reciprocal space. This is possible due to the fact that with proper positioning

974: of particles

975: one can subsequently factorize the Slater

976: determinant along {\it hyperplanes, planes and lines}. Using arguments similar to the $2D$ case 

977: we can perfrom cyclic exchanges without

978: crossing the node in $3D$ and higher dimensions. Therefore the proof for higher dimensions

979: is essentially the same. 

980: 

981: In many cases,

982: the proof can be extended to open shells. If an open-shell state is degenerate

983: it is necessary to fix 

984: the ambiguity in the nodes, for example, by considering

985: wave functions which transform according to an appropriate

986: symmetry subgroup \cite{davidnode, foulkes}. In general, however, depending on symmetries,

987: the number of degenerate

988: states, etc, one cannot rule out possibilities  

989: of states with the number of

990: cells beyond the minimal two  (this will be investigated in the next paper).

991:  

992: This concludes the arguments and the proof that for $d>1$ the 

993: noninteracting spin-polarized

994: fermion gas closed-shell ground states in periodic boundary conditions have only two nodal cells. 

995: 

996: \smallskip

997: \subsection{ III.a. Fermions on the surface of a sphere.}

998: 

999:  Spin-polarized free fermions on the surface of a sphere are 

1000: described by a Slater determinant with 

1001: one-particle states being the spherical harmonics

1002: $Y_{lm}(\vartheta,\varphi)$. The spherical harmonics are polynomials 

1003: in variables $\cos\vartheta$ and $\sin\vartheta e^{\pm i\varphi}$. 

1004:  The Slater matrix elements can be  

1005: rearranged to monomials in these variables and it 

1006: is then straightforward to

1007: factorize the determinant into similar form as obtained

1008: for the homogeneous fermion gas or for the harmonic oscillator

1009: \cite{mitasshort}. Let us assume that the particles are positioned

1010: as sketched in Fig.~\ref{fig:sph}a.  Using familiar expressions for

1011: the first few spherical harmonics we get for the $^5S(sp^3)$ state 

1012: \begin{equation}

1013: \Psi_{sp^3}(1,...,4)=\mu_0(u_1-u_0) v_1^2\prod_{2\leq i<j\leq 4}\sin(\varphi_{ij}/2)

1014: \end{equation}

1015: where we have denoted $u=\cos\vartheta$ and $v=\sin\vartheta$ and $\mu_0$ 

1016: is a constant. Similarly,

1017: for the  $^{10}S(sp^3d^5)$ state with nine fermions we obtain

1018: \begin{equation}

1019: \Psi_{sp^3d^5}(1,...,9)=

1020:  \mu'_0\Psi_{sp^3}(1,...,4) \times \nonumber

1021: \end{equation}

1022: \begin{equation}

1023: \times (u_2-u_0)(u_2-u_1)^3 v_2^6 \Psi_{1D}(5,...,9)

1024: \end{equation}

1025: where we have denoted 

1026: $\Psi_{1D}(1,...,N)=

1027: \prod_{j<k}\sin(\varphi_{jk}/2)$ in agreement with Eq. \ref{eq:van1d}.

1028: 

1029: 

1030: \begin{figure}[ht]

1031: \centering

1032: \begin{tabular}{ccc}

1033: \includegraphics[width=1.35in,clip]{fig4A.eps} &

1034: \quad &

1035: \includegraphics[width=1.35in,clip]{fig4B.eps} \\

1036: (a) & \quad & (b)

1037: \end{tabular}

1038: \caption{

1039: Positions of the particles on the sphere surface for the state $^{10}S(sp^3d^5)$

1040: using coordinate system $u=\cos(\vartheta)$

1041: and $0\leq \varphi \leq 2\pi$.

1042: The particles are positioned along the lines with constant $u=u_j$.

1043: For larger systems the patterns are the same with particles corresponding to increasing

1044: angular momentum $l$ lying on the lines with increasing $u_l$. (a) Particle positions

1045: used for derivation of the wave function factorization. (b) Particles after appropriate

1046: shifts along $\varphi$ and a subsequent rotation.  }

1047: %See the text for details. }

1048: \label{fig:sph}

1049: \end{figure}

1050: 

1051: For an arbitrary size closed-shell $S$ symmetry state the one-particle 

1052: states are occupied up to the angular

1053: momentum $L$ with the corresponding number of particles  

1054: $N=\sum_{l=0}^L(2l+1)=(L+1)^2$. 

1055:  Using the previous two examples 

1056: %the $1D$ wave function for $(2L+1)$ particles 

1057: %lying on the line $u=u_l, -1<u_l<1$  and occupying the states 

1058: %$(u_l)^l,$ $(u_l)^{l-1}(1-u_l^2)^{1/2}e^{\pm i\varphi},$ $..., (1-u_l^2)^{l/2} 

1059: %e^{\pm il\varphi}$

1060: %as

1061: %\begin{equation}

1062: %\Psi_{1D}^{sph}=C(u_l)\prod_{j>i}\sin(\varphi_{ij}/2)

1063: %\end{equation}

1064: %where $C(u_l)$ is a nonnegative factor which depends only on $u_l$.

1065: we can directly write the factorized wave function assuming  

1066: the

1067: particle positions follow the pattern in Fig.~\ref{fig:sph}a. 

1068: % We occupy states up to the angular

1069: %momentum $L$ so that for spin-polarized system we have $N=(L+1)^2$ particles

1070: %in a closed-shell $S$ symmetry state.

1071: We first factorize the line $u=u_L$ and then recursively the rest of the lines

1072: so that we can write

1073: \begin{equation}

1074: \Psi(1,...,N)= \mu_0\Psi_{2D}(1,...,N/I_L)\times

1075: \nonumber

1076: \end{equation}

1077: \begin{equation}

1078: \times \Psi_{1D}(I_L)v_L^{L(L+1)}\prod_{0\leq j<L}(u_L-u_j)^{n_j}=\nonumber

1079: \end{equation}

1080: 

1081: \begin{equation}

1082: =\mu(v_1, ...,v_L)\prod_{k=0}^{L-1}\left[

1083: \Psi_{1D}(I_{L-k}) 

1084: \prod_{j<L-k}(u_{L-k}-u_j)^{n_j} 

1085: \right]

1086: \end{equation}

1087: %v_{L-k}^{(L-k)(L-k+1)} 

1088: where $I_k$ is the subset of particles lying on the line $u=u_k$.

1089:  In the final expression the powers of $v_j$, 

1090: such as $v_L^{L(L+1)}$ in the preceding line, are absorbed

1091: into $\mu(v_1, ..., v_L)$.  This prefactor is nonnegative

1092: and does not affect the nodes or the wave function rotational invariance.

1093: 

1094: %Besides the spherical symmetry the wave function has the property

1095: %that as soon as the particles in $\Psi_{1D}(I_k)$

1096: % do not cross each other along a given line the configuration remains within the

1097: %same nodal domain. 

1098: We use the properties of $1D$ wave functions and 

1099: the rotational invariance to build the proof of the two nodal domains. First,

1100: note that for four particles in $^5S(sp^3)$ state it is easy to show \cite{mitasshort}

1101: that there are only 

1102: two domains. A little bit of algebra shows that

1103:  the node is encountered when all four particles lie on a circle resulting from a

1104: plane cutting the sphere.  Clearly, it is easy to position the four particles 

1105: on the sphere so that the triple exchanges do not cross this node.  For the induction

1106: step assume that for the size $L$ with $N=(L+1)^2$ the particles are connected. We need to show

1107: that for the size with $L \to L+1$ with additional $2(L+1)+1$ particles, the additional 

1108: particles are connected as well. Using the fact that  particles can be shifted along

1109: the $\varphi$ coordinate and rotated, one can reposition the particles as illustrated

1110: on the $^{10}S(sp^3d^5)$ state,  Fig.~\ref{fig:sph}b. 

1111: By applying the factorization to this particle arrangement we see that

1112: the additional particles are connected and the argument applies to an arbitrary

1113: size closed-shell state.

1114:  

1115: %Using these results as well as the insights from the previous paper we can write

1116: %the following theorem.

1117: 

1118: \subsection{\label{sec:boxsp} III.c. Fermions in a box.}

1119: Let us assume a system of fermions in

1120: a box $(0,\pi)^d$ with the condition that the wave function vanishes

1121: at the boundaries. For $d=1$ the one-particle

1122: states are $\phi_n(x)=\sin(nx), n=1,2,... $. Note that this can be written as 

1123: $\phi_n(x)=\sin(x)U_{n-1}(\cos x)$ where $U_n$ is the $n$-th degree Chebyshev polynomial of 

1124: the second kind. 

1125: We map the variables $x_i$ to $p_i=\cos(x_i)$  so that 

1126: $p_i\in (-1,1)$ for $i=1,...,N$. Note that the map $x \to p$ 

1127: is a homeomorphism (ie, it is bijective and continuous with its inverse). 

1128: Homeomorphisms

1129:  preserve topologies, eg, 

1130: the ordering of points, so that $x_a<x_b$ $\Leftrightarrow$ $p_a<p_b$.

1131: Using this map we can

1132: write the wave function for $N$ fermions in the $1D$ box directly as 

1133: \begin{equation}

1134: \Psi_{1D}(1,...,N)={\rm det} [\phi_n(x_i)] 

1135: =\mu_0\prod_k \sin(x_k) \prod_{i>j}(p_i-p_j) 

1136: \end{equation}

1137: where $\mu_0$ is a constant.

1138: 

1139: In the $2D$ box the one-particle states are given by 

1140: $\phi_{nm}(x,y)=\sin(x)\sin(y)U_{n-1}(p)U_{m-1}(q),

1141: \; n,m=1,2, ...$ where we denoted $q=\cos(y)$.

1142: If we absorb $\sin(x)\sin(y)$ into the common prefactor $\mu$,

1143: the Slater matrix

1144:  elements become monomials of the type $p^{n-1}q^{m-1}$. 

1145: The states which are occupied lie within the quarter of 

1146: the Fermi disk $n^2+m^2 \leq k_F^2$ and $ n,m>0$. 

1147: Assume that the particles lie on a lattice in the space of variables $p,q$ 

1148: so that they are positioned 

1149: on $M$  horizontal and vertical lines (see, eg, the upper right quadrant in Fig.3c).

1150: Using the techniques presented above and in the

1151: previous paper \cite{mitasshort} we can write down the wave function as 

1152: \begin{equation}

1153: \Psi_{2D}(1,...,N)= 

1154: \nonumber

1155: \end{equation}

1156: %\pm \Psi_{2D}(1,...,N/I_k)\times

1157: %\begin{equation}

1158: %\times \Psi_{1D}(I_k)\prod_{i<j}(p_i-p_j)^{n_j}=\nonumber

1159: %\end{equation}

1160: \begin{equation}

1161: = \mu(x_1,y_1,...,x_N,y_N)\prod_{k=1}^{M-1}\left[\Psi_{1D}(I_k) \prod_{j>k} (p_k-p_j)^{n_j}\right] 

1162: \label{fig:box2d}

1163: \end{equation}

1164: where the prefactor is a nonnegative function.

1165: Similar expression can be written down by factorizing along the lines $q=q_i$.

1166: The wave function therefore

1167: factorizes in a manner very similar to  the harmonic oscillator

1168: and periodic fermion gas. Therefore the proof of the two nodal domains

1169:  above can be constructed in a similar manner.

1170: First, it is not too difficult to show that

1171: the there are only two domains in the system with three particles.

1172: Next, one assumes that this is the 

1173: case for a system with $M$ lines and then it is straightforward to show

1174:  that the same applies to the system with $M+1$ 

1175: lines.

1176: Note that Eq.\ref{fig:box2d} suggest

1177: that apparently the wave function 

1178: has  both rotation and translation invariance. This is strictly not true,

1179: due to the nonlinearity of the map $(p,q) \to (x,y)$.

1180: What is true, however, is that node is not crossed during 

1181: rotations and translations since  $(x,y)\to (p,q)$ is homeomorphic.

1182: Therefore rotations and translations 

1183: change the wave function values but not the sign since

1184: there is no reordering of either the lines or the particles along the lines.

1185: That enables us

1186: to use rotations and translations to prove the 

1187: connectedness assuming that we take additional care

1188:  to keep the paths contained 

1189: within the box. 

1190: %However, this is not a significant 

1191: %restriction. 

1192: The proof then follows similar line of arguments as presented before.

1193: It is easy to show that the three-particle state has only two nodal cells.

1194: Place the particles on a lattice and position the system into the center of the box.

1195: We assume that

1196: the lattice constant is sufficiently 

1197: small

1198: so that translations by one lattice constant 

1199: would not push the particles out of the box. 

1200: It is then straightforward to show that the four translations 

1201: $T^x_{-a}T^y_{-a}T^x_aT^y_a$

1202: exchange three particles at the edge of the system and 

1203: by similar exchanges one can show connectedness

1204: of all particles for arbitrary size. Therefore the ground state closed-shell wave functions

1205: for particles in a box have the same

1206: generic properties as the fermionic models studied in preceding sections.

1207: 

1208: \section{\label{sec:theorem}

1209:  IV. Minimal number of nodal cells theorem for spin-polarized noninteracting systems.} 

1210: Using the results and proofs derived in this section

1211: and  also in the previous paper \cite{mitasshort} we can write

1212: the following theorem.

1213: 

1214: {\it Theorem.} Consider noninteracting or mean-field spin-polarized 

1215: fermions in $d>1$  with an exact wave function given by a 

1216: Slater determinant of one-particle states times a prefactor which does affect

1217: the fermion nodes.

1218:  Let the one-particle states are  

1219: such that the Slater matrix

1220: elements can be rearranged to monomials in particle coordinates or in coordinates transformed

1221: by a homeomorphic map. Then for an arbitrary size closed-shell ground state 

1222: the corresponding 

1223: wave function has the minimal

1224: number of two nodal domains.

1225: 

1226: The theorem covers a number of paradigmatic models and it is quite suggestive to think about this 

1227: as being 

1228:  related to general mathematical properties of zeros of functions defined through

1229: determinants. In fact, the 

1230: factorizations which enabled us to prove the two nodal cells property is directly

1231: related to the properties of multiple hyperplane configurations and to the multi-variate  

1232: Vandermonde  determinant theorem which can be found in mathematical literature in

1233: various contexts  \cite{varchenko, lagr}.

1234: %alos to interacting systems. 

1235: %For example it automatically covers exactly $delta({\bf r}_i-

1236: %{\bf r}_j)$ interaction which has no effect on the nodes although it changes the 

1237: %wave function in the vicinity of the nodes. Other short-range interaction will

1238: %also leave the nodal structure intact. 

1239: %Clearly, there must be a deeper connection  

1240: %related to the many-dimensional Vandermonde  determinant theorem which can be found

1241: 

1242: Considering that we restricted the proofs to noninteracting systems and we relied

1243: on the fact that the matrix elements are monomials, it is useful to consider

1244: cases

1245: which go beyond such a framework. 

1246: This line of thought leads us to the following interesting questions:

1247: 

1248: i) Is the {\it two nodal cell property} valid for noninteracting cases with Slater 

1249: {\it matrix elements

1250:    not reducible to monomials} ?

1251: 

1252: ii) What is the {\it impact of interactions} ?

1253: 

1254: A tentative answer to the first question is given in the next section where we show that

1255: atomic states in Coulomb potential, which cannot be reduced to monomials due to the

1256: shell structure, exhibit the same property.  We will not investigate the impact of interactions

1257: for spin-polarized systems in this  

1258: paper. However, we will study and prove the two nodal cells for 

1259:  perhaps even more important cases

1260: of interacting spin-unpolarized systems in the 

1261: section VI.

1262: 

1263: \section{\label{sec:spdsp} V. Spin-polarized atomic states.} 

1264: In the previous paper \cite{mitasshort}

1265:  we have proved that the atomic

1266: spin-polarized state $1s2s2p^3$ has two nodal cells for both 

1267: noninteracting and HF wave functions. The proofs for 

1268: atomic states are more 

1269: involved  since for Coulomb potential it is more

1270: difficult to find appropriate factorizations.

1271: The main complication is that orbitals in 

1272: different subshells and 

1273: angular momentum channels have, in general, 

1274: different radial dependences which cannot

1275: be all factorized out of the determinant into a common prefactor. 

1276: Nevertheless, it is possible to demonstrate the two nodal cells for atomic 

1277: states for several spin-polarized (half-filled) main subshells 

1278: (and possibly, for all the states relevant

1279: for the periodic table of elements). We will

1280:  illustrate the idea of the proof on

1281: the spin-polarized $^{15}S(1s2s2p^33s3p^33d^5)$ state with 14 electrons 

1282: and then point out 

1283: how the proof can be extended

1284: to larger systems.

1285: The one-particle orbitals

1286: are $\rho_{1s}(r),$ $\rho_{2s}(r),$ $\rho_{2p}(r)x,$

1287: $\rho_{2p}(r)y,$ $\rho_{2p}(r)z$,...,$\rho_{3d}(r)(2z^2-x^2-y^2)$, etc,

1288: and we use dimensionless coordinates which are rescaled as

1289: ${\bf r}\leftarrow Z{\bf r}/a_0$ with $Z$ being the nuclear

1290: charge and $a_0$ the Bohr radius.

1291: The wave function is given by

1292: \begin{equation}

1293: \Psi_{at}(1,...,14)=\nonumber

1294: \end{equation}

1295: \begin{equation}

1296: {\rm det}\{\rho_{1s}^{*},\rho_{2s}^{*},x,y,z,

1297: \rho_{3s}^{*}, \phi_{3px}^{*},..., \phi_{3dz^2}^{*}, ...\}

1298: \end{equation}

1299: where $\rho_{1s}^{*}(r)=\rho_{1s}(r)/\rho_{2p}(r)$,

1300:  $\rho_{2s}^{*}(r)=\rho_{2s}(r)/\rho_{2p}(r)$ and

1301: $\phi_{3px}^{*}({\bf r})=x\rho_{3p}(r)/\rho_{2p}(r)$, ..., $\phi_{3dz^2}^{*}({\bf r})

1302: =\rho_{3d}(r)(2z^2-x^2-y^2)/\rho_{2p}(r)$, ...,

1303: etc,  where

1304: we factorized the

1305:  nonnegative radial function $\rho_{2p}(r)$

1306: out of the determinant.

1307: 

1308: We will show the connectedness of all the particles in two steps. 

1309: We first demonstrate that the Slater determinant can be factorized 

1310: into subdeterminants corresponding to subshells $1s$,  

1311: $2s2p^3$ and $3s3p^33d^5$. That will enable us to show

1312: that the particles within each subshell

1313: are connected. In the second step we show, that the particles can exchange

1314: between the subshells.

1315: For this purpose we will specify the explicit paths

1316: and carry out a straightforward numerical check that there is no node crossing

1317: by  tracing the wave function along

1318: the paths. 

1319: 

1320: In order to show the factorization into shells

1321: we position the particles as follows: particle 1 

1322: is in the origin, particles 2 to 5 are on the surface

1323: of a sphere with the

1324: radius $\eta_a$ and particles 6 to 14 are 

1325: are on the surface

1326: of a sphere with the radius $\eta_b$. 

1327: The radii $\eta_a$ and $\eta_b$ are given by the radial nodes of 

1328: the orbitals $\rho_{2s}(r)$ and $\rho_{3p}(r)$, 

1329: ie, $\rho_{2s}(\eta_a)=0$ and  $\rho_{3p}(\eta_b)=0$. 

1330: The Slater determinant can be written in the form

1331: \begin{equation}

1332: \Psi_{at}(1,...,14)=

1333: {\rm det}

1334: \left[ \begin{array}{ccc}

1335: {\bf A } & {\bf D } &{\bf  G} \\

1336: {\bf B } & {\bf E' } & {\bf E}\\

1337: {\bf C } & {\bf F } & {\bf  H}\\

1338: \end{array} \right] 

1339: \label{eq:spd}

1340: \end{equation}

1341: where the block matrices ${\bf A}$ to ${\bf H}$ are given as follows

1342: \begin{equation}

1343: {\bf A}=\left[ \begin{array}{ccccc}

1344: \rho_{1s}^{*}(0)   &  a_1       &  a_1   & a_1  & a_1 \\

1345: \rho_{2s}^{*}(0)   &  0       &  0   & 0  & 0 \\

1346: 0 &  x_2      & x_3 & x_4  &  x_5 \\

1347: 0 &  y_2     & y_3  & y_4  &  y_5 \\

1348: 0 &  z_2     & z_3 & z_4 &  z_5 \\

1349: \end{array} \right]  

1350: \end{equation} 

1351: \begin{equation}

1352: {\bf B}=\left[ \begin{array}{ccccc}

1353: \rho_{3s}^{*}(0)   &  a_3       &  a_3   & a_3  & a_3   \\

1354: 0 &  \phi_{3px}^*(2)      &  ...  & 

1355:  &  \phi_{3px}^*(5) \\

1356: 0 &  \phi_{3py}^*(2)      & ...  &  &  \phi_{3py}^*(5)\\

1357: 0 &  \phi_{3pz}^*(2)       &  &  &  \phi_{3pz}^*(5) \\

1358: \end{array} \right]    

1359: \end{equation}   

1360: \begin{equation}

1361: {\bf C}=\left[ \begin{array}{ccccc}

1362: 0 &  \phi_{3dz^2}^*(2)      &  ...  & &  \phi_{3dz^2}^*(5) \\

1363: 0 &  \phi_{3dx^2}^*(2)      & ...  &  &  \phi_{3dx^2}^*(5)\\     

1364: 0 &  \phi_{3dxy}^*(2)       & ...  &  &  \phi_{3dxy}^*(5) \\

1365: 0 &  \phi_{3dyz}^*(2)       & ... &  &  \phi_{3dyz}^*(5) \\

1366: 0 &  \phi_{3dxz}^*(2)       & ... &  &  \phi_{3dxz}^*(5) \\

1367: \end{array} \right]       

1368: \end{equation}   

1369: \begin{equation}      

1370: {\bf D}=\left[ \begin{array}{cccc}

1371:  b_1       &  b_1   & b_1  & b_1   \\

1372:  b_2       &  b_2   & b_2  & b_2   \\

1373: x_6 &  ...        & ...  &  x_9 \\

1374: y_6 &  ...        & ...  &  y_9 \\

1375: z_6 &  ...        & ...  &  z_9 \\

1376: \end{array} \right]

1377: \end{equation}

1378: \begin{equation}

1379: {\bf E}=\left[ \begin{array}{ccccc}

1380:  b_3       &  b_3   & b_3  & b_3 & b_3  \\

1381: 0 &  ...        &  & ... &  0 \\ 

1382: 0 &  ...        &  & ... &  0 \\   

1383: 0 &  ...        &  & ... &  0 \\

1384: \end{array} \right] 

1385: \end{equation} 

1386: \begin{equation}

1387: {\bf F}=\left[ \begin{array}{cccc}

1388:   \phi_{3dz^2}^*(6)      &  ...  & &  \phi_{3dz^2}^*(9) \\

1389:   \phi_{3dx^2}^*(6)      & ...  &  &  \phi_{3dx^2}^*(9)\\

1390:   \phi_{3dxy}^*(6)       & ...  &  &  \phi_{3dxy}^*(9) \\

1391:   \phi_{3dyz}^*(6)       & ... &  &  \phi_{3dyz}^*(9) \\

1392:   \phi_{3dxz}^*(6)       & ... &  &  \phi_{3dxz}^*(9) \\

1393: \end{array} \right]

1394: \end{equation}

1395: \begin{equation}          

1396: {\bf G}=\left[ \begin{array}{ccccc}

1397:  b_1       &  b_1   & b_1  & b_1 & b_1  \\

1398:  b_2       &  b_2   & b_2  & b_2 & b_2  \\

1399: x_{10} &  ...        & ...  &  ... & x_{14}\\

1400: y_{10} &  ...        & ...  &  ... & y_{14}\\

1401: z_{10} &  ...        & ...  &  ... & z_{14}\\

1402: \end{array} \right]

1403: \end{equation}

1404: \begin{equation}

1405: {\bf H}=\left[ \begin{array}{ccccc}

1406:   \phi_{3dz^2}^*(10)      &  ...  & &  &  \phi_{3dz^2}^*(14) \\

1407:   \phi_{3dx^2}^*(10)      & ...  &  & &  \phi_{3dx^2}^*(14)\\

1408:   \phi_{3dxy}^*(10)       & ...  & &  &  \phi_{3dxy}^*(14) \\

1409:   \phi_{3dyz}^*(10)       & ... &  & &  \phi_{3dyz}^*(14) \\

1410:   \phi_{3dxz}^*(10)       & ... &  & &  \phi_{3dxz}^*(14) \\

1411: \end{array} \right]

1412: \end{equation}

1413: and $a_1=\rho_{1s}^{*}(\eta_a)$, $a_3=\rho_{3s}^{*}(\eta_a)$,

1414: $b_1=\rho_{1s}^{*}(\eta_b)$, $b_2=\rho_{2s}^{*}(\eta_b)$, $b_3=\rho_{3s}^{*}(\eta_b)$.

1415: The block matrix ${\bf E'}$ is the same as  ${\bf E}$ except that it has only four columns 

1416: instead of five. 

1417: 

1418: The following three row additions 

1419: in the Slater matrix given by Eq.\ref{eq:spd} 

1420: allow for factorization into subdeterminants.

1421: We first eliminate elements $a_1$ in the first row   

1422: by adding an appropriate multiple of the fifth row. Similarly, 

1423: we eliminate the terms with $b_1$ in the first row  

1424: by adding a multiple of the second row. This

1425: leads to the first row having only one nonzero element

1426: $A'_{11}=\rho_{1s}^{*}(0)$

1427: $-a_1\rho_{3s}^{*}(0)/a_3$

1428: $-(b_1-a_1b_3/a_3)\rho_{2s}^{*}(0)/b_2$ and reduces the Slater matrix

1429: by the first row and the first column.

1430: Finally,

1431:  by adding a multiple of the second row to the fifth row, 

1432: both ${\bf E'}$ and ${\bf E}$ become zero matrices and we can write

1433: \begin{equation}

1434: \Psi_{at}(1,...,14)=A'_{11}\, {\rm det} {\bf B'}\,

1435: {\rm det}

1436: \left[ \begin{array}{cc}

1437:  {\bf D' } &{\bf  G'} \\

1438: {\bf F } & {\bf  H}\\

1439: \end{array} \right] 

1440: \end{equation}

1441: where  ${\bf B'}$ is the matrix ${\bf B}$ without the first column,

1442: while ${\bf D' }, {\bf  G'}$ are the matrices ${\bf D}, {\bf  G}$ without the first row.

1443: The three factors in the expression above correspond to the three main 

1444: subshells namely, $1s$, $2s2p^3$ and $3s3p^33d^5$ with the dependence

1445: on positions of particles 1, 2-5 and 6-14,

1446: respectively.

1447: %Note that so far we have not specified the actual positions in these subshells except that they

1448: %are

1449: %on the same spherical surface: the $1s$ particle is trivially in the origin, 

1450: %$2s2p^3$ on the sphere with

1451: %radius $r=\eta_a$ 

1452: %and  $3s3p^33d^5$ on the sphere with $r=\eta_b$. 

1453: The key point is that each of the subshells represents a system of particles on a sphere in

1454: the $S$-symmetry state and for such cases 

1455: we proved that the particles are connected.

1456: That concludes the argument that the particles within each subshell are connected.

1457: 

1458: \begin{figure}[ht]

1459: \centering

1460: \includegraphics[width=2.0in,clip]{3spdCsphere.eps} 

1461: \caption{Positions of 14 electrons for the proof that triple exchanges connect the particles 

1462: in the $^{15}S(1s2s2p^33s3p^33d^5)$  state.

1463: Particles 2 to 5 lie on the spherical surface with 

1464: the radius equal to the radial node of $\rho_{2s}(r)$ orbital.

1465: Particles 6 to 14 lie on the spherical surface with the radius equal to the radial node of $\rho_{2p}(r)$.}

1466: \label{fig:3spd}

1467: \end{figure}

1468: 

1469: 

1470: More difficult part of the proof is to show that one can exchange the particles between the subshells.

1471: We were able to prove analytically the exchange between $1s$ and $2s2p^3$ in our previous paper. For

1472: larger cases the analytic proof becomes very tedious and it is much more efficient

1473: to evaluate numerically the determinant along the following 

1474: exchange paths.

1475: The particles are positioned as sketched in Fig.\ref{fig:3spd}

1476: with the coordinates given by

1477: ${\bf r}_1=(0,0,0)$, ${\bf r}_2=(\eta_a,0,0)$, ${\bf r}_2=(0,\eta_a,0)$,

1478: ${\bf r}_{4,5}=(0,0,\pm \eta_a)$, ${\bf r}_6=(\eta_b/\sqrt{2},\eta_b/\sqrt{2},0)$,

1479: ${\bf r}_{7,8}=(\pm\eta_b/\sqrt{2},\mp\eta_b/\sqrt{2},0)$, ${\bf r}_{9,10}=(0,0,\pm\eta_b)$,

1480: ${\bf r}_{11,12}=(\eta_b/\sqrt{2},0,\pm\eta_b/\sqrt{2})$ and ${\bf r}_{13,14}=(0,\eta_b/\sqrt{2},\pm\eta_b/\sqrt{2})$.

1481: There exists enough of triple 

1482: exchanges of neighbouring particles, along the sides of corresponding triangles, to connect 

1483: all the particles into a single cluster. 

1484: For example,

1485: the wave function values for exchanges $123\to 231$, $236\to 362$ (and other exchanges for 

1486: illustration) are given in Fig.~\ref{fig:atexch}.

1487: 

1488: \begin{figure}[ht]

1489: \centering

1490: \includegraphics[width=2.4in,clip]{atexch.eps}

1491: \caption{

1492: Wave function of the $^{15}S(1s2s2p^33s3p^33d^5)$ 

1493: state (arb. u.) along triple exchange paths parametrized by $t$. Particles are positioned 

1494: as given in Fig.~\ref{fig:3spd}. The plotted 

1495: exchanges:  1,2,3 (full line); 2,3,4 (dashed line); 2,3,6 (dotted line); 6,7,9 (dashed-dotted line); 

1496: 9,11,13 (double dashed-dotted line). By symmetry, these non-crossing exchanges connect all the particles of this 

1497: state.}  

1498: \label{fig:atexch}

1499: \end{figure}

1500: 

1501: This is enough to show that the three subshells are connected

1502: since many other exchanges are identical due to the symmetries in particle positions 

1503: and the wave function $S$ symmetry. 

1504: %Many other exchanges are possible. 

1505: For the illustration in Fig.~\ref{fig:atexch} we used the noninteracting radial orbitals.

1506:  For HF orbitals one gets the same qualitative picture since

1507: the basic spatial properties of the noninteracting

1508: and HF orbitals are qualitatively the same 

1509: (ie, ordering of radial nodes of $\rho_{nl}(r)$ orbitals,

1510: behaviour at nucleus, tails,

1511:  etc). The proof uses only the fact that some of the radial nodes of the one-particle orbitals

1512: are ordered as in the non-interacting case so it can be extended to HF wave functions as well. 

1513: %In addition, for large $Z$ the

1514: %one-particle states approach the noninteracting ones since the impact of 

1515: %electron-electron interactions is decreasing and orbitals become even more closer to the

1516: %noninteracting ones.

1517: 

1518: One can expand the proof to larger systems, such as for the state with

1519: occupied fourth main subshell  

1520:  $4s4p^34d^54f^7$ and beyond. The factorization is similar: the particles in the

1521: fourth subshell are positioned on the spherical surface with the radius

1522: equal to the radial node of $\rho_{4d}(r)$ orbital. Somewhat long

1523: but straightforward

1524: algebraic rearrangements show

1525: that the Slater determinant of the $30\times 30$ matrix can be 

1526: reduced to product of subdeterminants corresponding to the subshells.

1527: For the purpose of this paper we do not deem necessary to go through the explicit proof.

1528: 

1529: 

1530: \section{\label{sec:hegsu} VI. Interacting spin-unpolarized fermions.}

1531: 

1532: It is straightforward to understand 

1533: the fermion nodes of noninteracting

1534: spin-unpolarized or partially polarized systems with more than one-electron in each

1535: spin-channel.

1536: The wave function is a product of spin-up and -down Slater determinants and the 

1537: number of nodal cells is the product of the number of cells in each subspace.

1538: (As mentioned before, we assume that the Hamiltonian, even with interactions considered

1539: below, does not include terms with spin so that particles can be assigned 

1540: to the spin subspaces.) For the ground

1541: states with two nodal cells in each subspace we get $2\times2=4$ nodal cells.

1542: This is the nodal structure of a single configuration Hartree-Fock wave function

1543: and  as such has been used in many 

1544: fixed-node QMC calculations. However, analyses of small interacting systems

1545: revealed that this is

1546: not correct and interactions do change the nodal topologies

1547: and the number of nodal cells. 

1548: For the first time 

1549: this has been demonstrated 

1550: for the Be atom \cite{dariobe}

1551: and then also for a few other atoms and small molecules \cite{cmt28,bajdichpfaffians}.

1552: 

1553: In the previous paper \cite{mitasshort} we have outlined a proof 

1554: showing the two nodal domains property for $2D$

1555: harmonic fermions in a closed-shell singlet state for arbitrary size. The most interesting feature

1556: of the proof was that the result was very 

1557: robust in the sense that almost any arbitrary weak interaction would induce the change in the 

1558: topology of the nodal surfaces. 

1559: 

1560: We

1561: refresh some of the key notions and arguments here. 

1562: Let us assume a spin-unpolarized system in its

1563: closed-shell singlet ground state of $2N$ particles.

1564: Consider a simultaneous exchange of an odd number of spin-up pair(s)

1565: {\it and} an odd number of spin-down pair(s).

1566: For noninteracting wave functions such simultaneous pair exchanges

1567: imply that the node will be crossed once or multiple times. This must be the case

1568: whenever the spin-up and -down subspaces are independent of each other, such as 

1569: in the mean-field and HF wave functions. However,

1570: if there exists a point $R_f$ such that during the simultaneous

1571: spin-up and -down pair exchanges the inequality

1572:    $|\Psi|>0$ holds along the whole path,

1573:  then the wave function has only two nodal cells.

1574: 

1575: Using this property we can now demonstrate the two nodal cell for several types of systems.

1576: %Consider a singlet ground state of $2N$ interacting particles.

1577: Under rather general conditions, 

1578:  we will show that the correlation included in

1579: the Bardeen-Cooper-Schrieffer (BCS) pairing

1580: wave function \cite{sorella,carlson} given by

1581: \begin{equation}

1582: \Psi_{BCS}(1,...,2N)={\rm det} [\Phi(i,j)] 

1583: \label{eq:bcs}

1584: \end{equation}

1585: smoothes out 

1586:  the noninteracting four nodal cells

1587: into the minimal two. 

1588: Here $\Phi(i,j)=\Phi(j,i)$ is a singlet pair orbital for

1589: $i\negthinspace\negthinspace\uparrow $ and

1590: $j\negthinspace\negthinspace\downarrow$ fermions and we decompose it

1591: into noninteracting and correlated components

1592: $\Phi(i,j)= \Phi_{0}(i,j) + \Phi_{corr}(i,j)$. Using one-particle orbitals

1593: we can write 

1594: $\Phi_{0}(i,j)=\sum_{\alpha}\negthinspace \phi_{\alpha}(i)\phi_{\alpha}(j)$ where

1595: the sum is over HF (or noninteracting) orbitals 

1596: while $ \Phi_{corr}(i,j)=\sum_{\alpha\beta}c_{\alpha\beta}\phi_{\alpha}(i)\phi_{\beta}(j)$

1597: where $\{c_{\alpha\beta}=c_{\beta\alpha}\}$ are variational parameters and 

1598: the sum is over unoccupied or virtual ("correlating") 

1599: orbitals. The BCS wave function was originally introduced for conventional

1600: superconductors, however, it proved to be very successful also in describing 

1601: the correlation effects in electronic structure problems \cite{sorella,bajdichpfaffians}.   

1602: %appropriate expansion coefficients. 

1603: 

1604: \subsection{VI.a. $3D$ harmonic spin-unpolarized fermions with interactions.}

1605: 

1606: For the $3D$ harmonic oscillator one can show that the interactions lead to the

1607: minimal number of nodal cells in a rather simple and elegant way.

1608:   First let us consider

1609: a small system which is easy to analyze. We illustrate

1610: the idea on $2N=8$ particles in the singlet

1611: ground state with the particle positions given in Fig.~\ref{fig:3dho}. 

1612: The corresponding closed-shell

1613: singlet ground state for 

1614: the $3D$ harmonic oscillator is $^1 S(1s^22p^6)$. (Note that for the harmonic potential

1615: the $2p$ state is below the $2s$ state,

1616: unlike for the Coulomb potential.)

1617: For simplicity, we drop the gaussian prefactors since they do not affect

1618: the nodes and 

1619: the pairing 

1620: functions can be then written

1621: as follows

1622: \begin{equation}

1623:  \Phi_{0}(i,j) = 1 + 3{\bf r}_i\cdot {\bf r}_j

1624: \end{equation}

1625: and 

1626: \begin{equation}

1627:  \Phi_{corr}(i,j) = \gamma [3({\bf r}_i\cdot {\bf r}_j)^2 -r_i^2r_j^2]

1628: \end{equation}

1629: where the noninteracting part is constructed from the occupied $1s,2p$

1630:  orbitals while for the correlating 

1631: part the unoccupied $3d$ subshell orbitals were used. 

1632: The constant $\gamma$ is a variational parameter.

1633: Clearly, both the pair orbitals

1634: and the wave function are spherically symmetric.

1635: Assuming the positions as given in the Fig.~\ref{fig:3dho},  which are of high symmetry

1636: and therefore simplify the evaluation of the wave function, we get 

1637: \begin{equation}

1638: \Psi(1,...,8)= 4\gamma\cos\varphi[1-(3+\gamma)\cos^2\varphi]/3

1639: \end{equation}

1640: where $r_1=r_5=0$,  $\; r_i=1, i=2,3,4,6,7,8$

1641: and $\cos\varphi={\bf r}_4\cdot {\bf r_8}$.

1642: 

1643: \begin{figure}[ht]

1644: \centering

1645: \includegraphics[width=1.8in,clip]{3dHO.eps}

1646: \caption{Positions of $2N=8$ fermions of the $3D$ harmonic well system. Fermions 1 to 4 have spins up,

1647: while 5 to 8 have spins down. Fermions 1 and 5 are at the origin while the rest of the particles

1648: lie on the surface of the sphere with $r=1$. Note that rotation of the system

1649: by $\pi$ around the $z-$axis exchanges particles $3,4$ and $7,8$.

1650: See the text for details. }

1651: \label{fig:3dho}

1652: \end{figure}

1653: 

1654: For $\gamma =0$ the wave function vanishes since the particles in both

1655: spin subspaces lie on the noninteracting node (ie, four particles on the same plane). 

1656: The key point is that for

1657: $\gamma\neq 0$ and $\varphi\neq \pi/2$ the wave function

1658: does not vanish.

1659: Since the wave function is spherically symmetric we can rotate the particles 

1660: around the $z-$axis by $\pi$. This transformation 

1661: exchanges particles 3 and 4 in the spin-up 

1662: channel and particles 7 and 8 in the spin-down channel. 

1663:  The wave function is rotationally

1664: invariant and nonvanishing what clearly implies that the BCS wave function has

1665: spin-up and -down subspaces interconnected since simultaneous exchanges in the spin-up  

1666: and spin-down {\it does not } hit the node. 

1667: %This fulfills the condition for 

1668: %the two nodal cells wave functions.  

1669: On the other hand, it is easy to check

1670: that the rotation of the particles in one spin channel only causes a node-crossing

1671: since then $\cos\varphi$ vanishes either at $\pi/2$ or $3\pi/2$. 

1672: %Note that we never specified the interaction and using the BCS variational wave

1673: %function we tacitly assume that it is accurate enough to describe the key 

1674: %effect from the interactions. 

1675: 

1676: It is clear that our argument is correct

1677: whenever the strength of the interaction

1678: is small so that the BCS wave function is accurate enough. 

1679: The variational 

1680: parameter $\gamma$ is related to the interaction strength. This

1681:  implies that the topological 

1682: change from two to four nodal cells takes place for arbitrary small, but 

1683: nonvanishing, interaction strength.

1684: The only assumptions were

1685: that the interaction will induce electron correlation and lead to a nonzero

1686: variational parameter $\gamma$ (or, in general, to a nonvanishing correlation

1687: component) and that the wave function is spherically

1688: symmetric. Our analytic argument therefore shows

1689: that for weak

1690: interactions the nodal cell "degeneracy" is lifted

1691: and the multiple cells are smoothed to the minimal

1692: number of two. 

1693: 

1694: It is important to note that

1695: one can find different pictures in other circumstances. For example,

1696:  for strong or nonlocal interactions, imposed additional symmetries or

1697: large degeneracies, the nodal changes can be different and the resulting

1698: nodal topology might exhibit more than the two nodal cells;

1699: this aspect is further

1700: discussed in the conclusion section and require more investigation. 

1701: 

1702: Coming back to our example of

1703: harmonic fermions with interactions, using similar arguments 

1704: the two nodal cells can be demonstrated for

1705: an arbitrary size closed-shell ground state. 

1706: Consider $2N$ particles in a singlet closed-shell ground state.

1707: The closed-shell states can be labeled by $M$ which represents the "Fermi momentum"

1708: for the harmonic oscillator and $N=(M+1)(M+2)(M+3)/6$.  Let us

1709: decompose $N$ into the maximum 

1710: odd number of 

1711: pairs $N_P$ and the rest. Therefore we write  

1712: $N=2N_P+J$ where $N_P$ is an odd integer while $J$ has one of the values $J=0,1,2,3$, so 

1713: that we can also define $J=N\, {\it mod}\, 4$.

1714: For example, in

1715: the previous example with $2N=8$ particles we have $N_P=1$ and $J=2$. First, we place $J$

1716: particles from each spin channel on the $z$-axis in

1717:   distinct positions.  Next,

1718: we form $N_P$ pairs 

1719:  in each spin subspace

1720: so that, say, the pair $i\negthinspace\uparrow,

1721: (i+1)\negthinspace\uparrow $ is positioned as  

1722: given by $(r_i,\vartheta_i,\varphi_i)=(r_k,\vartheta_k,\varphi_k)$,

1723:  $\,(r_{i+1},\vartheta_{i+1},\varphi_{i+1})=(r_k,\vartheta_k,\varphi_k+\pi)$,

1724: where $r,\vartheta,\varphi$ are the spherical coordinates.

1725: Here $k=1,...,N_P$ labels the pairs

1726: in the spin-up channel; the spin-down particles are 

1727: placed similarly and labeled by the pair index $l=N_p+1,...,2N_P$.

1728: In this configuration the particles lie on the 

1729: noninteracting node since 

1730: \begin{equation}

1731:  {\rm det}\left[\Phi_0(i,j)\right]={\rm det}\left[\sum_{n+m\leq M} \phi_{nm}(i)

1732:  \phi_{nm}(j)\right]=

1733: \nonumber

1734: \end{equation}

1735: \begin{equation}

1736: ={\rm det}\left[\phi_{nm}(i)\right]

1737: {\rm det}\left[\phi_{nm}(j)\right]=\Psi_{HF}^{\uparrow}\Psi_{HF}^{\downarrow}

1738: \end{equation}

1739: The rotation of the system by $\pi$ causes node-crossings 

1740: in both spin channels

1741: so that both spin-up and -down Slater determinants must  vanish due to the rotational

1742:  invariance.

1743: Now, if all the $N_P$ pair distances and angles

1744: $r_k,\varphi_k,\vartheta_k,r_l,\vartheta_l,\varphi_l$ are distinct, 

1745: then each of the matrices $\left[\phi_{nm}(i)\right], \left[\phi_{nm}(j)\right]$

1746: has exactly one linearly dependent row, ie, both have ranks $N_P-1$.

1747: This can be verified directly

1748: for small values of $M$ and then using induction for any $M$.

1749: Consequently, the matrix $\left[\Phi_0(i,j)\right]$ has linear dependence

1750: in one row and one column, ie, it has the rank of $N_P-1$ as well. In general,

1751: adding virtual states through $\Phi_{corr}(i,j)$ provides

1752: independent rows/columns

1753: %(eg, $(M+1)(M+2)/2$ independent rows/columns from the first unoccupied shell)

1754: which eliminate the linear dependency so

1755: that ${\rm det}\left[\Phi_0(i,j)+\Phi_{corr}(i,j)\right]$ is nonzero. Let us now assume 

1756: that the interactions do not break the rotation invariance. Since at this point the

1757: wave function does not vanish and is rotationally invariant,  

1758: it does not vanish for the whole exchange path implying that the correlated 

1759: BCS wave function has only two nodal cells regardless of the size $2N$.

1760: 

1761:  Clearly, the assumption of the rotational invariance for the interacting

1762: case might be too

1763: restrictive since the interactions could possibly break the spherical symmetry.

1764: To demonstrate  the two nodal cells in such a case one would need to show

1765:  that the wave function does not vanish for the complete exchange path.

1766: % small change in a variable makes it nonvanishing.)

1767: 

1768: \subsection{VI.b. $2D$ harmonic spin-unpolarized fermions with interactions for 

1769: $N=2N_P+J$ when $J=2,3$.}

1770: 

1771: In our previous paper we have demonstrated

1772: the two nodal cells for the

1773: ground states with 

1774: the number of particles $2N$ where $N=2N_P+J$ and $J=0,1$. (We use the 

1775: same notation as in $3D$ where $N_P$ is the maximum odd number of pairs and $J=0,1,2$ or 3).

1776: Note that the states with $J=0$ or 1 exist at any size. Here we would like to present 

1777: an alternative

1778: and a little bit longer

1779: proof which covers $J=0,1$ but also remaining cases

1780:  when $J=2,3$, eg, 

1781: for $N=21=2\times9 +3,\;$ $N=28=2\times13 +2$, etc.

1782: In $3D$, the rotation axis can accommodate multiple particles so that one can always

1783: form an odd number of pairs for the exchanges. In $2D$, the

1784: symmetry point (origin) can accommodate only one particle   

1785: from each spin subspace therefore the proof has to be modified.

1786: For this purpose we outline a factorization

1787: which is different from the previous paper \cite{mitasshort}. Let us remind that for $2D$

1788: harmonic oscillator the closed-shell states and the system size are labeled by

1789: $M=1,2, ...$ where $n+m\le M$ with the number of fermions in one spin

1790: channel given by $N=(M+1)(M+2)/2$.  We express the one-particle states 

1791: as polynomials in $r^2$ and $(re^{i\varphi})^m$  where $r, \varphi$ are 

1792: the cylindric coordinates (we again omit the gaussian factors

1793: which are irrelevant for the nodes).  Note that for a given $M$ the 

1794: quantum number $m$ increases or decreases by 2 unlike

1795: in $3D$. Therefore for $M=2$ we have states $re^{\pm i\varphi}$, for $M=3$ the 

1796: additional states are $(r^2-1), r^2e^{i\pm 2\varphi}$, etc. 

1797: We first write down the wave functions for three- and six-particle 

1798: spin-polarized systems with positions given in Fig. \ref{fig:2dho}.

1799: 

1800: \begin{figure}[ht]

1801: \centering

1802: \begin{tabular}{ccc}

1803: \includegraphics[width=1.4in,clip]{figYa.eps} & \quad &

1804: \includegraphics[width=1.4in,clip]{figYb.eps} \\

1805: (a) & \quad & (b) \\

1806:  \quad & \quad & \quad \\

1807:  \quad & \quad & \quad \\

1808: \includegraphics[width=1.4in,clip]{figYc.eps} & \quad &

1809: \includegraphics[width=1.4in,clip]{figYd.eps} \\

1810: (c) & \quad & (d) \\

1811: \end{tabular}

1812: \caption{Alternative positions of three and six spin-polarized fermions for the $2D$ harmonic

1813: potential.

1814: (a) Three particles positioned on the noninteracting node.

1815: (b) Three particles in positions with a nonvanishing wave function.

1816: (c) Six particles positioned on the noninteracting node.

1817: (d) Six particles in positions with a nonvanishing wave function.

1818: See the text for details. }

1819: \label{fig:2dho}

1820: \end{figure}

1821: 

1822: The three particle wave function is given by ${\rm det}\{1,re^{\pm i\varphi}\}$

1823: \begin{equation}

1824: \Psi_{2D}(1,2,3)

1825: =\mu_0r_0^2\prod_{1(2)\leq j <k \leq 3}\sin(\varphi_{jk}/2)

1826: \end{equation}

1827: where $\mu_0$ is a constant.

1828: The product lower bound is 2 if the positions of the particles 

1829: are  as in Fig.~\ref{fig:2dho}a, while it is 1 if the positions  

1830: are as in Fig.~\ref{fig:2dho}b.

1831: Note that in the case of configuration in Fig.~\ref{fig:2dho}a 

1832: rotation by $\pi$

1833: flips the wave function sign so that the wave function vanishes

1834: while for the configuration in  Fig.~\ref{fig:2dho}b it does not.

1835:  For six particles in the positions outlined in  Fig.~\ref{fig:2dho}d the

1836: wave function is clearly nonvanishing since  we can write 

1837: \begin{equation}

1838: \Psi_{2D}(1,...,6)

1839: =\mu_0(r_2^2-r_1^2)r_2^6\prod_{j<k}^6\sin(\varphi_{jk}/2)

1840: \end{equation}

1841: where $\mu_0$ is some constant (in Fig.~\ref{fig:2dho}d we have

1842:  $r_1=0$).

1843: Alternatively,

1844: the six particles can be positioned in pairs as sketched 

1845:  in Fig.~\ref{fig:2dho}c where the particles are positioned 

1846: on the HF node so that the noninteracting wave function vanishes.

1847: 

1848: For the sake of completeness we provide the expression for the wave function for

1849: a general system

1850: with size $M$  

1851: \begin{equation}

1852: \Psi_{2D}(1,...,N) =\nonumber

1853: \end{equation}

1854: \begin{equation}

1855: =\mu(r_1, ..., r_{\tilde M})\prod_{k=1}^{\tilde M}\left[

1856: \Psi_{1D}(I_k)\prod_{1\leq j<k}(r^2_k-r^2_j)^{n_j}\right]

1857: \label{eq:2dho}

1858: \end{equation}

1859: where $\tilde M= {\rm int}[(M+1)/2]$ is the integer part of $(M+1)/2$. The particles with

1860: indices in the subset $I_k$ lie on a circle with the

1861: radius $r_k$ and $n_k=2(M+1-2k)+1$.  The prefactor is again a nonnegative function

1862: of  

1863:  $r_j$ which has 

1864: no impact on the nodes

1865: and is constant for rotations. From the last equation we see that in cases when $M$ is even, eg,

1866: $M=6, N=21$, we end up with factorization of the type $11\times 7 \times 3$ particles 

1867: while for $M$ odd, eg, $M=7,N=28$, we get $13\times 9\times 5 \times 1$. That means that the last

1868: three particles ($M$ even) or the last six particles ($M$ odd) can be always arranged into possibilities 

1869: as sketched in

1870: Fig~\ref{fig:2dho}a-d. 

1871: %The factorization given by Eq.~\ref{eq:2dho} provides an 

1872: %alternative to the expression derived before \cite{mitasshort} where we used cartesian

1873: %coordinates.

1874: It is now clear that we can prove that

1875: the spin-up and -down configuration space are interconnected

1876: for the cases when $J=2,3$ as specified above. 

1877: Using the alternative configurations for three and six particles we can arrange 

1878: the systems into such 

1879: positions that the rotation by $\pi$ exchanges odd number of pairs in both spin channels.

1880: For example, if $N$ is

1881: odd and $(N-1)/2$ is even

1882: %, instead of locating one of the particles of each spin into 

1883: %the origin as we did in the previous paper \cite{mitasshort}, we rather

1884: we position three particles on a triangle as given in Fig.~\ref{fig:2dho}b. 

1885: That takes out one pair from

1886: the $(N-1)/2$ exchanges and the proof then follows using the arguments for $3D$ harmonic fermions.

1887: Similarly, if  $N$ is even and $N/2$ is also even,

1888: we position one particle of each spin at the origin and 

1889: five from each spin channel on a circle. This eliminates three pairs from $N/2$ and again we end 

1890: up with remaining number of pairs being odd so that the rest of the proof follows similarly to the

1891: $3D$ case.

1892: 

1893: \subsection{VI.c. Spin unpolarized interacting $2D$ and $3D$ homogeneous electron gas.}

1894: 

1895: Let us now prove the two nodal cells for the spin-unpolarized closed shells

1896: for $d>1$ homogeneous  gas with interactions. We use the relevant definitions

1897: introduced  previously for the spin-polarized periodic fermion gas (section III.a.).

1898: We assume a $2D$ system of $2N$ particles in the spin singlet ground state

1899: where the one-particle states occupy the Bloch states within the Fermi disk with the

1900: radius 

1901:  $k_F$. Let us first consider $2N=10$ particle 

1902: case and we position the particles 

1903: as in Fig.  \ref{fig:2dhegunpol}. 

1904: The wave function is translationally invariant therefore the action of

1905: $T^{y}_{\pi}$ leaves the wave function unchanged. 

1906:  Assuming first that there is no interaction, we see that the

1907:  translation  exchanges the particles $1,2$ but due to the invariance 

1908: the wave function value does not change: that is possible only if the particles

1909: are sitting on the node.

1910: This also agrees with expressions for the wave functions derived before  

1911: (see Eq. \ref{eq:five2dxi}.) Consider now that we switch-on interactions

1912: and describe the particles by the BCS wave function as given above. 

1913: The pairing orbital for the noninteracting gas is given by

1914: \begin{equation}

1915: \Phi_0(i,j)=\sum_{|{\bf k}|\leq k_F} \phi_{\bf k}(i)\phi_{\bf -k}(j)=1+2[\cos(x_{ij})+\cos(y_{ij})]

1916: \end{equation}

1917: where we used the occupied orbitals $\{1,e^{\pm ix},e^{\pm iy}\}$.

1918: The correlated part is given by

1919: \begin{equation}

1920: \Phi_{corr}(i,j)=

1921: \gamma\sum_{|{\bf k}|> k_F}\phi_{\bf k}(i)\phi_{\bf -k}(j)=

1922: \nonumber

1923: \end{equation}

1924: \begin{equation}

1925: \gamma[\cos(x_{ij})\cos(y_{ij})+\cos(2x_{ij})+\cos(2y_{ij})]

1926: \label{eq:2dcorr}

1927: \end{equation}

1928: where $\Phi_{corr}(i,j)$ includes the next two unoccupied stars of states with $k=\sqrt{2},2$ 

1929: and $\gamma$ is a variational parameter. 

1930: %We position the particles as sketched in Fig. \ref{fig:2dhegunpol}.

1931: %Consider now the translation $T^y_{\pi}$. If the particles 3-5 are positioned in such a way that

1932: %after translation they reach positions related to the original ones by a mirror symmetry around $y$-axis

1933: %% then it is clear that the value of the wave function has to be the same. At the same time the positions

1934: %of particles 4 and 5 are {\it exchanged}. This means that the only way how this would be the case is that

1935: %the wave function is zero. 

1936: 

1937: \begin{figure}[ht]

1938: \centering

1939: \includegraphics[width=1.5in,clip]{figZ2.eps}

1940: \caption{Positions of five spin-up electrons from an interacting ten-fermion system

1941: in the $2D$ periodic box $(-\pi,\pi)^2$.

1942: For clarity, the five down-spin particles  

1943: are not shown. The spin-down particles are positioned in the same pattern

1944: on two lines $x=const$.}

1945: %$\xi_0,\xi_1$. 

1946: %See the text for more details. }

1947: \label{fig:2dhegunpol}

1948: \end{figure}

1949: 

1950: Unlike for the noninteracting wave function (ie, $\gamma=0$) 

1951: the BCS correlated wave function does not vanish and this is indeed the case due to the 

1952: arguments presented above. The Slater 

1953: matrix of uncorrelated wave function has linear dependency while for the correlated case

1954: there always exists a configuration of lines with nonvanishing wave function. This is

1955: due to the elimination of the linear dependency through addition of

1956: virtual orbitals, as explained

1957: for the $3D$ harmonic oscillator.

1958: Note that as soon as the wave function

1959: is translationally invariant the wave function does not vanish

1960: for the whole translation path, implying that the spin-up and -down

1961: nodal domains are interconnected and the wave function has only two nodal cells. 

1962: It is straightforward to extend the proof

1963: to arbitrary size closed-shell system. 

1964: The configuration of particles can be given as follows: an odd number of pairs (eg, one) in each

1965: spin channel

1966: is positioned in a similar way as in

1967: \ref{fig:2dhegunpol} so that the translation by $\pi$ exchanges the particles in the pair. For example,

1968: for a spin-up pair 

1969: $i,i+1$ we specify the coordinates as $y_i=-y_{i+1}=\pi/2$  and $x_i=x_{i+1}=\xi_0$ where $\xi_0$ is

1970: distinct from $x$ coordinates of the rest of particles with the same spin.

1971: The rest of the particles is positioned in such a way so that the $T_{\pi}^x$ brings them into

1972: a symmetric position regarding the reflection around axis $y$. 

1973: The wave function is translationally invariant and therefore uncorrelated wave function

1974: vanishes while for BCS case it is, in general, nonzero. 

1975: The two nodal cells property then 

1976: follows using the same general arguments as in the previous cases.

1977: 

1978: The correlated wave function has another important property  namely that one can wind around the

1979: periodic box a singlet pair of particles (ie, spin-up and spin-down pair) without hitting

1980: the node. Indeed, this is implied by the fact that there are only two nodal cells: 

1981: once this is the case one can then always

1982: find such a path that the pair of particles can wind around the box without

1983: node-crossing. For example, for the ten-particle 

1984: example above we can wind spin-up and -down particle pair around along $y=\zeta_0$ without hitting

1985: the node. We plot the wave function for two types of particle positions: first, the

1986:  spin-down particles are at identical 

1987: positions as the spin-down ones, ie, ${\bf r}_6={\bf r}_1$, ${\bf r}_7={\bf r}_2$, etc 

1988: (Fig.~\ref{fig:wind}a);

1989: second, the particles  in spin-up and -down channel were offset by small displacement

1990: ${\bf r}_{i+5}={\bf r}_i+0.2,$ $i=1,...,5$, see 

1991:  Fig.~\ref{fig:wind}b. 

1992: Consider now the simultaneous translation $T_{2\pi}^y$ of the particles 1 and 6. This 

1993: translation winds

1994: the pair around the box and the wave function values for the paths are  

1995: plotted in Fig.~\ref{fig:wind}a,b. 

1996: The value of the displacement between the particle positions neither the repositioning

1997: of the particles around the box is not crucial and there 

1998: exists a significant size subspace of particle positions for which the winding of the pair

1999: is possible.

2000: We have chosen just two simple examples as qualitative illustrations.

2001: 

2002: Note that the wave function enables to wind also a single particle without the node crossing. 

2003: However, single particle winding is possible both for correlated and uncorrelated wave functions

2004: since, say, the spin-up HF part has two nodal domains and correlated wave function 

2005: only smoothes out the HF nodes. Therefore this property is

2006: not affected by the correlation in any significant manner.

2007: 

2008: 

2009: \begin{figure}[ht]

2010: \centering

2011: \begin{tabular}{cc}

2012: \includegraphics[width=2.6in,clip]{windA.eps} \\

2013: (a)\\

2014:  \quad \\

2015:  \quad \\

2016:  \quad \\

2017: 

2018: \includegraphics[width=2.6in,clip]{windB.eps} \\

2019: (b) \\

2020:  \quad \\

2021: \end{tabular}

2022: \caption{ Winding the particles 1 (spin-up) and 6 (spin-down) around the $2D$ box along 

2023: the line $y=\zeta_0$. The dashed line is the Hartree-Fock wave function (arb.u.) while 

2024: the full line is

2025: the correlated BCS wave function.

2026: a) Particles 1 and 6, 2 and 7, etc, share the same position. The uncorrelated wave function

2027: touches the node quadratically since the spin-up and -down determinants are 

2028: identical. b)  The positions of particles in spin-down channel are offset by 0.2

2029: from the positions in the particle spin-up channel. 

2030: The uncorrelated wave function

2031: crosses the node multiple times since the spin-up and -down determinants are different.

2032: }

2033: \label{fig:wind}

2034: \end{figure}

2035: 

2036: The effect that singlet pair of particles can pass 

2037: through nodal openings between the spin-up and -down subspaces

2038: have been demonstrated for small number of particles before,

2039: for example, in our paper on employing pfaffian pairing functions for electronic structure

2040: problems \cite{bajdichpfaffians}. 

2041: The case here illustrates that this property stems from interconnected

2042: spin-up and -down subspaces and therefore applies to similar two nodal cell wave functions,

2043: in general.

2044: % as soon as the  wave function has 

2045: %only two nodal domains there always exists a path that a pair of particles can wind around the box.

2046: % a pair of spin-up and a pair of spin-down particles 

2047: %can simultaneously exchange

2048: %without crossing the node. Clearly, this effect is generic and does not say much about the phase 

2049: %%in which the electrons are (eg, superconducting state).  

2050: 

2051: It is well known that a Fermi liquid, such as the homogeneous electron gas, becomes unstable to

2052: a weak attractive interaction, develops Cooper pair instability and opens the superconductivity

2053: gap \cite{supercond}. This effect is captured by the BCS wave function which we used in the proof above. 

2054: Cooper pairs can therefore wind around the box without hitting the node although this effect

2055: is not exclusive to Cooper pairs only. 

2056: %Although

2057: % For example, the same effect can be observed in the Fermi

2058: %liquid with repulsive interactions which is typically not a superconductor.

2059:  In fact, the connectedness of  spin-up and -down

2060: subspaces is a rather generic property in the sense 

2061: that it appears also in systems which are not necessarily superconducting, ie, in Fermi liquids

2062: with repulsive interactions. Since superconductivity is characterized by macroscopic phase

2063: coherence and a number of other properties, which might or might not be related to

2064: the nodal topologies, on the basis of the analysis of the BCS wave function above

2065: one expects  that the  nodal opening 

2066: {\it necessarily} appears in the 

2067: superconducting state, 

2068: however,

2069: for the superconductivity to occur this condition is not sufficient.

2070: In addition, here we assume only

2071: the simplest $s-$wave pairing; for the $p-$wave or higher angular momentum pairing the situation

2072: is less clear and needs to be further investigated. 

2073: 

2074: \section{\label{sec:dm} VII. Nodes of fermionic density matrices.}

2075:  

2076: In this part we generalize the ideas presented in preceding 

2077: section to temperature dependent density matrices \cite{davidnode}. Consider

2078: first a system of spin-polarized fermions. The temperature/imaginary time density matrix  

2079: is given by

2080: \begin{equation}

2081: \varrho(R',R,\beta)= \sum_{n} e^{-\beta E_n} \Psi^{*}_n(R')\Psi_n(R)

2082: \label{eq:dm1}

2083: \end{equation}

2084: where $\beta$ is the inverse temperature and  

2085: the sum is over the complete system of eigenstates of a given Hamiltonian $H$. 

2086: It is clear that the density matrix

2087: is antisymmetric in particle exchanges in the same manner as  

2088: the wave function $\Psi(R)$ or $\Psi(R')$. Therefore the notion of fermion

2089: nodes can be generalized also to density matrix 

2090: in $(2dN+1)$ dimensions since there is an explicit dependence on $\beta$ as well. As pointed

2091: out elsewhere \cite{davidnode}, for fixed $R'$ and $\beta$ one can study the nodes and nodal cells 

2092: in the $dN$ dimensional $R-$subspace. Similarly to the wave function, the node then becomes

2093: $(dN-1)$-dimensional manifold with the generalization

2094: that it is dependent on $R'$ and $\beta$. 

2095: For the fixed  $R'$ and $\beta$

2096: the tiling property holds in the same

2097: manner as for the wave functions. The key additional feature of the density matrix is

2098: that once there are only two nodal cells at some initial $\beta_0$ than this property holds for 

2099: any $\beta >\beta_0$ \cite{davidnode}. This is not difficult to understand since the density matrix fulfills

2100: the following linear equation

2101: \begin{equation}

2102: -{\partial \varrho(R,R',\beta)

2103: \over \partial \beta}=H\varrho(R,R',\beta)

2104: \label{eq:dm2}

2105: \end{equation}

2106: with an initial condition

2107: \begin{equation}

2108:  \varrho(R,R',0) = {\cal A} \delta(R-R') ={\det }[\delta({\bf r}_i-{\bf r'}_j)]

2109: \label{eq:dm3}

2110: \end{equation}

2111: where ${\cal A}$ is the antisymmetrizing operator.

2112: 

2113: We now understand that for almost any  Hamiltonian with interactions 

2114: the density matrix will have only

2115: two nodal cells for sufficiently large $\beta$ (ie, at  low temperatures). This is due to the fact

2116: that at sufficiently low temperature  the ground state becomes dominant

2117: (Eq. \ref{eq:dm1}). The key point now is  to show that this

2118: is the case also for {\it high} temperatures.  

2119: The free particle density matrix 

2120: is given by

2121: \begin{equation}

2122:  \varrho (R,R',\beta) = (2\pi\beta)^{-dN/2}{\rm det} \left[\exp(-|{\bf r}_i-{\bf r'}_j|^2/2\beta)\right]

2123: \label{eq:dm4}

2124: \end{equation}

2125:   where we assume atomic units with $\hbar^2/m=1$.

2126:  (For other than free

2127:   boundary conditions, such as for the periodic ones,

2128:  the expression has to be modified accordingly.)

2129: Note that this

2130:  density matrix is universal since   

2131: at sufficiently high

2132: temperatures the interactions become irrelevant. 

2133: 

2134: There are several ways how to prove that 

2135: the high-temperature density matrix has only two nodal cells.

2136: For very small $\beta$ one can use the induction as follows. 

2137: Assume that $N$ particles are described by the density matrix given by

2138: Eq.~\ref{eq:dm4} and the particles are connected by triple exchanges for a fixed $R',\beta$.

2139: We add an additional particle 

2140:  with the label $N+1$ to the system which 

2141: occupies a certain region of the configuration (free) space.

2142: The particle $N+1$ is positioned at the border of the occupied region

2143: and let us assume that the particles with labels 

2144: $N-1$ and $N$ are its closest neighbours. Let us move the three particles $N-1, N, N+1$

2145:  away from the rest without crossing the node (what can be always done

2146: by appropriate positioning). 

2147: Since for small $\beta$ the overlaps of gaussians become small, one can factorize the determinant 

2148: into a product of the three particles $N-1,N,N+1$ determinant and the determinant for the

2149:  rest. 

2150: For the three particles the density matrix has only two nodal cells as one can show easily

2151: \cite{davidnode}

2152: and therefore

2153: the additional particle is connected. This applies to both $R$ and $R'$ subspaces since they must

2154: have identical properties. (In what follows we will show that, in fact, at high temperatures 

2155: the primed and unprimed spaces are connected as well.)  

2156: 

2157: There is also an alternative proof which is interesting also on its own since it provides

2158: a different view on the density matrices. 

2159: Note that the functional form of the free particle density matrix 

2160: has a unique property. Comparing the Eq.~\ref{eq:dm4} with the  

2161:  BCS wave function (Eq.~\ref{eq:bcs}) we see that the

2162: high-temperature density matrix can be identified with a BCS wave function

2163: if we properly define an underlying effective model. Instead of

2164: our original  system of $N$ spin-polarized fermions, let as consider

2165: a model system with $2N$ particles so that

2166:  the configurations $R$ and $R'$ denote positions

2167: of these {\it different } sets of particles, which we will call for simplicity unprimed and 

2168: primed particles. Let us then define a new Hamiltonian ${\tilde H}(R,R')$ 

2169: with

2170: an effective quadratic interaction between the  unprimed and primed particles as

2171: given by

2172: \begin{equation}

2173: {\tilde H}(R,R') =  T + T' + V_0\beta^{-1} \sum_{i,j} |{\bf r}_i - {\bf r'}_j|^2

2174: \label{eq:dm5}

2175: \end{equation} 

2176: where $T$ and $T'$ denote kinetic energy operators for the corresponding sets of particles and 

2177: $V_0$ is a constant with appropriate dimensions. 

2178: Note that particles within the given set, say, unprimed, are antisymmetric but 

2179: otherwise

2180:  do not interact with each other. The interaction appears only between the primed 

2181: and unprimed degrees of freedom as given by the Hamiltonian $\tilde H(R,R')$.

2182: For $\beta \to 0$ the exact wave function for this system is a BCS wave function 

2183: $\Psi(R,R')={\rm det}[\phi({\bf r}_i,{\bf r'}_j)]$ where the indices  

2184: $i$ and $j$ label unprimed and primed particles, respectively. 

2185: The pairing function $\phi({\bf r}_i,{\bf r'}_j)$ is obviously the gaussian given above. 

2186: It is not too difficult to demonstrate that this wave function

2187: (and the density matrix) has only two nodal cells. 

2188: For example, we can expand the gaussian into plane waves

2189: \begin{equation}

2190: \exp(-|{\bf r}_i-{\bf r'}_j|^2/2\beta)=\sum_{\bf k}c_{\bf k}e^{i{\bf k}\cdot({\bf r}_i-{\bf r'}_j)}

2191: \label{eq:dm6}

2192: \end{equation}

2193: where $\{c_{\bf k}\}$ are expansion coefficients.

2194:  The sum is over states within the Fermi sphere of

2195: a periodic box which accommodates $2N$ particles for a given density. Let us specify that

2196: $N$ is large and $\beta$ is such that 

2197: the system can be considered classical so that the actual interaction in the 

2198: original Hamiltonian $H$ is irrelevant.

2199: Then the density matrix corresponds to the Hartree-Fock product

2200: \begin{equation}

2201: {\rm det}[e^{i{\bf k}_i\cdot {\bf r}_j}]{\rm det}[e^{-i{\bf k}_i\cdot {\bf r'}_j}]

2202: \label{eq:dm7}

2203: \end{equation}

2204: However, we have already proved that such system has only two nodal cells. In addition, 

2205: if the sum includes also "excited states"

2206: (ie, beyond the Fermi sphere) due to the primed-unprimed interactions, 

2207:  we find that  the unprimed and primed 

2208: nodal cells are interconnected.  Therefore at classical temperature  $\beta_0$

2209: the density matrix has only two nodal cells and then the same is true for arbitrary

2210: $\beta>\beta_0$. This primed-unprimed

2211: interconnection becomes less pronounced and ceases to exist at $\beta \to \infty$ since

2212: then the density matrix is proportional to the "noninteracting" product

2213: $\Psi_0(R)\Psi_0(R')$ where $\Psi_0$ is the ground state for the original physical system 

2214: of $N$ fermions described by $H$. This proof is therefore based on an interesting duality between the 

2215: $N$ spin-polarized fermions at classical temperatures and the model

2216: system with $2N$ particles with a temperature-dependent,

2217:  harmonic interaction between the unprimed and primed

2218: subspace particles. 

2219: 

2220: Possibly, there might be also a third way how to prove the minimal number of nodal cells for

2221: the high-temperature density matrix  

2222: through diagonalization of the quadratic Hamiltonian $\tilde H$ (we have not 

2223: investigated this  possibility). 

2224: %The derivations above were carried out

2225: %only for  spin-polarized density matrix. However, using basically the same steps

2226: %it is not difficult to show that the same is true also for the spin-unpolarized case

2227: %assuming that the exchange given above was 

2228: The fact that also the density matrices have two nodal cells is important for the path 

2229: integral Monte Carlo methods which for fermions often employ the fixed-node 

2230: approximation adapted for the 

2231: path integrals \cite{davidhe}. 

2232: %The fact that the singlet particle pair can 

2233: %wind around the box without hitting the node indicates that it might be relevant

2234: %for the superconductivity. It is very suggestive that a node-free winding of 

2235: %a singlet pair is 

2236: %% a necessary condition for superconductivity to occur (albeit not sufficient since

2237: %the effect, as we have seen, is rather generic, for example it appears regardless

2238: %whether the interaction is attractive or repulsive). In a transport process the node

2239: %can be considered a scattering potential since the wave function is prescribed to have 

2240: %a particular shape there. In that sense the nodes have probably more deep relationship 

2241: %to transport phenomena and this aspect remains to be investigated.  

2242: 

2243: \section{VIII. Discussion and conclusions.}

2244:  

2245: Inspired by previous

2246: conjectures and numerical studies

2247: \cite{jbanderson,davidnode,lesternode,sorella,foulkes,dariobe,bajdichnode}

2248: the presented analysis and proofs generalize and clarify the properties

2249: of ground states fermion nodes. 

2250: We have employed symmetries, wave function factorizations

2251: and triple exchanges 

2252: to prove that for $d>1$ the closed-shell ground states 

2253: have two nodal cells for arbitrary number of particles

2254: in several paradigmatic models.

2255: In this paper the proofs were carried for closed shells, mainly to avoid 

2256: additional complications from degeneracies and some

2257: of these aspects will be addressed in subsequent papers.

2258:    

2259: It is perhaps more interesting

2260: to discuss viable possibilities when the ground state wave functions might have 

2261: {\it more 

2262: than two} nodal cells. Clearly, it is easy to generate more nodal cells by imposing 

2263: additional symmetry

2264: or boundary conditions. Another possibility comes from very strong interactions,

2265: for example, whenever the effect of interaction

2266: would

2267:  become competitive with the kinetic energy increase from forming additional 

2268: nodes it is possible that more than the minimal two nodal cells can form. 

2269: %Note that the tiling

2270: %property rules out forming of nodal "bubles" and therefore additional nodal cells will

2271: %be all equivalent

2272: %  interaction  electronic systems whenever exchange and correlation

2273: %energy are of the same order of magnitude neither HF nor BCS framework can be considered

2274: %reliable and the nodes in these systems can be different from what we have derived

2275: %above. 

2276: One can also construct nonlocal interactions

2277: which would violate some of the properties mentioned here (eg, the tiling property),

2278: or reorder the states so that, for example, excited states could 

2279: lie below the ground state, etc.

2280: Another candidate for unusual effects in the nodal structure are open-shell systems with

2281: large degeneracies and densities of states at the Fermi level.  Some of these 

2282: interesting issues will be subject of future studies.

2283: 

2284:  

2285: In conclusion,  we have presented a number of new results which reveal the structure

2286: of nodes and nodal cells

2287:  of fermionic wave functions. 

2288: Building upon ideas introduced in previous paper

2289: we were able to demonstrate the minimal number of two nodal cells in several 

2290: spin-polarized models such

2291: as noninteracting fermions in a periodic box, in a box with zero boundary conditions,  fermions

2292: on a spherical surface, etc. This enabled us to formulate a theorem

2293: which states that in $d>1$ the minimal two nodal cells are present for any

2294: Slater determinant with monomial matrix elements for any size which

2295: allows for a closed-shell nondegenerate ground state. We have shown that this 

2296: property extends also to cases which cannot be described by the Slater matrix of monomials

2297: such as the noninteracting and HF atomic states up to the $3d$ shell.

2298: We have studied the effect of interactions on the noninteracting nodal cells

2299: of spin-unpolarized, closed-shell singlets.  Our results show

2300: that, in general, the interactions smooth out

2301: the noninteracting multiple nodal cells into the minimal number of two. 

2302: For interacting homogeneous

2303: electron gas we have demonstrated that the two nodal cells allow 

2304: singlet pairs of particles to wind around 

2305: the periodic box without crossing the node.

2306: Finally, we have 

2307: shown that the temperature/imaginary time

2308:  density matrix has very similar nodal structure and therefore

2309: the minimal two nodal cells property applies also to density matrices with

2310: important implications for path integral Monte Carlo 

2311: simulations. We have demonstrated this by using an appropriate mapping of the density matrix

2312: onto the ground state of a model systems with twice as many particles interacting 

2313: with temeperature-dependent harmonic potentials.

2314:  

2315: 

2316: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2317: %	Bibliography, assumed extension is *.bib

2318: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2319: 

2320: I am grateful to J. Koloren\v c, M. Bajdich and L.K. Wagner for reading the manuscript, 

2321: comments and discussions.  

2322: I would like to acknowledge the support by

2323: %ONR-N00014-01-1-0408, and

2324: NSF DMR-0102668, DMR-0121361 and  EAR-0530110 grants.

2325: 

2326: \begin{thebibliography}{99}

2327: \bibitem{qmchistory} D.M. Ceperley and M.H. Kalos,

2328: in {\sl Monte Carlo Methods

2329: in Statistical Physics}, Ed. K.

2330: Binder, pp.145-194, Springer, Berlin, 1979;

2331: K.E. Schmidt and D. M. Ceperley,

2332:  in {\sl Monte Carlo Methods in Statistical

2333: Physics II}, pp.279-355, Ed.

2334:  K. Binder, Springer, Berlin, 1984.

2335: \bibitem{hammond} B.L. Hammond, W.A. Lester, Jr., and P.J. Reynolds,

2336: {\sl Monte Carlo Methods in ab initio quantum chemistry}, World Scientific,

2337: Singapore, 1994.

2338: \bibitem{qmcrev} M.W.C. Foulkes, L. Mitas, R.J. Needs and G. Rajagopal,

2339:   Rev. Mod. Phys.  {\bf 73}, pp. 33-83 (2001).

2340: \bibitem{jbanderson} J. B. Anderson, J. Chem. Phys. {\bf 65} 4121 (1976);

2341: J. B. Anderson, Phys. Rev. A, {\bf 35}, 3550 (1987).

2342: \bibitem{davidnode} D. M. Ceperley, J. Stat. Phys. {\bf 63}, 1237 (1991).

2343: \bibitem{lesternode} W. A. Glauser, W. R. Brown, W. A. Lester, D. Bressanini, B. L.

2344:   Hammond and M. L. Koszykowski, J. Chem. Phys. {\bf 97}, 9200 (1992).

2345: \bibitem{sorella} M. Casula and S. Sorella, J. Chem. Phys.

2346: {\bf 119}, 6500 (2003)

2347: \bibitem{foulkes} W. M. C. Foulkes, Randolph Q. Hood and 

2348: R. J. Needs, Phys. Rev. B {\bf 60}, 4558 (1999).

2349: \bibitem{dariobe} D. Bressanini, D. M. Ceperley and P. Reynolds, in

2350: {\sl Recent Advances in Quantum Monte Carlo Methods, II}

2351: Ed. W. A. Lester, S. M. Rothstein, and S. Tanaka, World Scientific, Singapore (2002).

2352: \bibitem{dariohe} D. Bressanini,and P. Reynolds, Phys. Rev. Lett. {\bf 95}, 110201 (2005).

2353: \bibitem{bajdichnode} M. Bajdich, L. Mitas, G. Drobny, and L. K.

2354: Wagner,  Phys. Rev. B {\bf 72}, 075131 (2005)

2355: \bibitem{cmt28} L. Mitas, G. Drobny, M. Bajdich, L.K. Wagner,

2356: in {\sl Condensed Matter

2357: Theories, Vol.20}, Eds. J.W. Clark, R.M. Panoff, and H. Li, Nova Science Publishers, New York, 2006;

2358: cond-mat/0409406. 

2359: \bibitem{berger} M. Berger, {\sl A Panoramic View of Riemannian Geometry}, Springer, Berlin, 2000, p.84. 

2360: \bibitem{varchenko} A. N. Varchenko, Math. USSR Izvestiya, N.3, {\bf 35}, 543 (1990)

2361: \bibitem{lagr} C. K. Chui and L. M. Lai, in

2362: {\sl  Nonlinear and Convex Analysis}, 

2363: Eds. B.L. Lin and S.Simons, pp. 23-35, Marcel Dekker, 1987.

2364: %\bibitem{3rdvander} 

2365: \bibitem{bajdichpfaffians} M. Bajdich, G. Drobny, 

2366: L.K. Wagner, and K. E. Schmidt, Phys. Rev. Lett. {\bf 96},

2367: 130201 (2006). 

2368: \bibitem{mitasshort} L. Mitas, submitted, cond-mat/0601485

2369: \bibitem{carlson} J. Carlson, S.-Y. Chang, V. R. Pandharipande, and K. E. Schmidt, Phys. Rev.

2370: Lett. {\bf 91}, 050401 (2003).

2371: \bibitem{supercond} R. Shankar, Rev. Mod. Phys. {\bf 66}, 000129 (1994);

2372: A. J. Leggett, Rev. Mod. Phys. {\bf 47}, 331 (1975);

2373:  H. Bruus and K. Flensberg, {\sl Many-Body Quantum Theory

2374: in Condensed Matter Physics}, Oxford University Press, Oxford, 2004, p. 325.

2375: \bibitem{davidhe} D.M. Ceperley,  Rev. Mod. Phys. {\bf 67}, 279 (1995).

2376: \end{thebibliography}

2377: %XXXXXXX discussion, including abstract, 

2378: %                 End of the document

2379: \end{document}

2380: