1:
2: %\documentclass[preprint,aps]{revtex4}
3: %\documentclass[aps,prl,preprint,groupedaddress]{revtex4}
4: \documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
5:
6:
7: %\newcommand{\mwcd}[1]{\marginpar{\protect\scriptsize \protect\setlength
8: % {\baselineskip}{8pt} to editor- MS revision: #1}}
9: %\setlength {\marginparwidth}{7.0cm}
10:
11: \usepackage{graphicx}% Include figure files
12:
13: \begin{document}
14:
15: \title
16: {Spin and Valley dependent analysis of the two-dimensional low-density
17: electron system in
18: Si-MOSFETS.
19: }
20:
21: %
22: \author
23: {
24: M.W.C. Dharma-wardana\cite{byline1}
25: and Fran\c{c}ois Perrot$^*$
26: }
27: %
28: \affiliation{
29: Institute of Microstructural Sciences,
30: National Research Council of Canada, Ottawa, Canada. K1A 0R6\\
31: }
32: %\date{07-Nov-2002}
33: \date{\today}
34: %
35: \begin{abstract}
36: The 2-D electron system (2DES) in Si metal-oxide field-effect transistors
37: (MOSFETS) consists of two distinct electron fluids interacting
38: with each other. We calculate the total energy as a function of the density
39: $n$, and the spin
40: polarization $\zeta$ in the
41: strongly-correlated low-density
42: regime, using a classical mapping to
43: a hypernetted-chain (CHNC) equation inclusive of bridge terms.
44: Here the ten distribution functions,
45: arising from spin and valley indices, are self-consistently calculated to obtain
46: the total free energy, the chemical potential, the
47: compressibility and the spin susceptibility.
48: The $T=0$ results are compared with the 2-valley Quantum
49: Monte Carlo (QMC) data of Conti et al. (at $T=0$, $\zeta=0$) and found to be
50: in excellent agreement. However, unlike in the
51: one-valley 2DES, it is shown that
52: {\it the unpolarized phase is always the stable phase in the 2-valley system},
53: right up to Wigner Crystallization
54: at $r_s=42$. This leads to the insensitivity of
55: $g^*$ to the spin polarization and to the density.
56: The compressibility and the spin-susceptibility enhancement
57: calculated from the
58: free energy confirm the validity of a simple approach to the
59: two-valley response
60: based on coupled-mode formation.
61: This enables the use of
62: the usual (single-valley) exchange-correlation
63: functionals in quantum calculations of MOSFET properties
64: provided mode-coupling effects are taken into account.
65: The enhancement of the spin susceptibilty calculated
66: from the coupled-valley response and directly from the
67: 2-valley energies is discussed. The three methods, QMC, CHNC, and
68: Coupled-mode theory agree closely. Our results
69: contain no {\it ad hoc} fit parameters. They agree with
70: experiments and do not invoke impurity effects or metal-insulator
71: transition phenomenology.
72: %
73: \end{abstract}
74: \pacs{PACS Numbers: 05.30.Fk, 71.10.+x, 71.45.Gm}
75: %\vspace{0.5in}
76: %
77: \maketitle
78: %
79: \section{Introduction.}
80: %{\it Introduction.}---
81: The 2-D electron systems in GaAs-based structures
82: as well as those found in Si MOSFETs are being
83: intensely studied owing to the accessibility of a
84: wide range of electron densities under controlled conditions,
85: leading to a wealth of the experimental observations\cite{krav}.
86: The nature of the physics depends very much on the
87: ``coupling parameter''
88: $\Gamma$ = (potential energy)/(kinetic energy).
89: The $\Gamma$ for the 2DES at $T=0$, and the mean density $n$ is
90: equal to the mean-disk radius $r_s=(\pi n)^{-1/2}$ per electron,
91: usually expressed in effective atomic units which
92: depend on the bandstructure mass and ``background''
93: dielectric constant.
94: The 2DES in GaAs/AlAs structures will be
95: called a simple 2DES or
96: one-valley 2DES
97: to distinguish it from the two-valley system found in
98: Si-MOSFETS.
99: The inversion layer adjacent to an
100: oxide layer grown on the Si-(001) surface contains two equivalent valleys
101: which host the two valleys of the electron fluid.
102: Various aspects of such multi-valley systems were studied\cite{afs} by
103: Sham and Nakayama, Rasolt et al, and others, mainly in the high density limit.
104: The simple 2DES is also a two component
105: system because of the two spin species, while the 2-valley
106: system involves 4-components and ten pair-distribution functions (PDFs).
107:
108: In a recent study of the effective mass $m^*$ and the effective
109: Land\'e-$g$ factor of the 2DES, the Coulomb coupling between
110: the electrons of the two valleys was shown to
111: have a dramatic effect at
112: low densities when the coupling becomes large\cite{cmm}. In effect, the
113: elementary excitations of the two fluids in the two valleys
114: interact to form coupled modes,
115: giving rise to new effects.
116: It was shown experimentally\cite{sash} that $m^*g^*$ rises
117: rapidly with decreasing density in Si-MOSFETS, and that
118: this rise is due to a dramatic increase in $m^*$, independent of the
119: spin polarization, while $g^*$
120: remains essentially constant. Calculations for the
121: Si system which account for the inter-valley Coulomb coupling
122: quantitatively predict\cite{cmm} the sharp increase in
123: $m^*g^*$.
124: It was also shown that
125: $g^*$ remained essentially constant, in strong contrast to the behaviour
126: found for the simple one valley 2DES\cite{cmm}. The effective mass was also shown to
127: be practically independent of the spin polarization $\zeta$,
128: in excellent quantitative agreement with the data of
129: of Shashkin et al.\cite{sash} for Si-MOSFETS. This
130: result leads to the view that the rapid
131: rise in $m^*$ near $n=1\times10^{11}$ e/cm$^2$ is {\it not caused}
132: by localization effects, impurity effects, etc., usually
133: associated with the phenomenology of the 2-D metal-insulator transition.
134: The enhancement of $m^*g^*$ in the 1-valley 2DES of GaAs/AlAs systems
135: is found to be strongly dependent on the spin polarization\cite{jun},
136: in strong
137: contrast to the Si-MOSFET case. Our calculations\cite{cmm} show that
138: the physics of the 1-valley system is dominated by the presence of a
139: transition to a fully polarized state which makes $g^*$ increase
140: rapidly with $r_s$ as the transition density is approached.
141: The 2-valley system shows {\it no such
142: transition to the spin-polarized state} and is insensitive to the
143: spin polarization.
144:
145: Given that perturbative many-body theory becomes questionable
146: for $r_s>1$, within the present context
147: there are two approaches to evaluating the susceptibility
148: enhancement. The second derivative of the total
149: free energy $F(n,\zeta,T)$
150: with respect to the spin-polarization $\zeta$ gives a direct
151: value for $m^*g^*$ of the two-valley system. This requires the
152: $\zeta$-dependent 2-valley energy which is not yet available from
153: quantum Monte Carlo (QMC) simulations.
154: However, we can evaluate $F(n,\zeta,T)$
155: using CHNC, and also show that the CHNC results agree with QMC
156: data (available at $\zeta=0$ and $T=0)$. Another approach,
157: which avoids the need for a full 4-component calculation is to
158: build up the 2-valley susceptibility
159: by noting that the one-valley energy $F(n,\zeta,T)$ is
160: available at $T=0$
161: from QMC, and at any $T\ne 0$ from CHNC.
162: The coupling of the excitations of the
163: two electron fluids in the two valleys can be included in the
164: coupled response function in a standard way.
165: Then we find that the increase in $g^*$ in the GaAs 2DES
166: is associated with the "blow-up"
167: of the single-valley spin response, while the behaviour of the
168: Si-MOSFET 2DES is related to the "blow-up" of the coupled-mode
169: response. The static small-$q$ limit of the
170: spin response function provides the needed $\chi_s/\chi_p$.
171:
172: The objective of this paper is: (i) Present $F(n,\zeta)$ data
173: for the two valley system by a full CHNC calculation involving the
174: 10 pair distributions that are needed in the 2-valley system and
175: establish the close agreement of the CHNC results
176: with the QMC calculations.
177: (ii) Construct the coupled-mode response functions using the well-established
178: one-valley
179: data and show that these results are also validated by QMC and full CHNC
180: results.
181: That is, we present the CHNC results and compare with QMC data
182: where available, and establish close agreement. We will not present detailed
183: finite-T calculations (and hence $m^*$ calculations, as detailed in
184: ref.~\cite{cmm}) in this paper, partly to limit the phase space
185: that has to be studied,
186: and since finite-T
187: data are not presently available from QMC for comparison.
188:
189: In section~\ref{resp} we discuss the construction of the coupled-mode
190: response functions which use {\it only the one-valley exchange-correlation data}
191: to obtain the 2-valley behaviour via a physically motivated approximation
192: to the inter-valley correlations. The compressibility predicted via the
193: small-$q$ limit of the so constructed
194: coupled-mode response is found to agree very
195: well with that from QMC or the full 4-component CHNC energy calculation.
196: This validates the coupled-mode model
197: used in the calculation (ref.\cite{cmm}) of the $m^*g^*$ enhancement in
198: Si-MOSFETS. The full 2-valley energy calculations enable us
199: to examine the
200: usual 1-valley local-density approximation (LDA)
201: in Si-MOSFETS and the corrections
202: arising from coupled-mode effects. Finally we discuss the
203: spin susceptibility enhancement obtained from these calculations, and
204: the question of relating the electron-disk radius $r_s$ used in
205: these calculations to the experimental densities.
206:
207: %
208: \section{CHNC calculations for the 2-valley electron fluid.}
209: \label{chnc}
210: We consider a 2-DES in a Si-MOSFET at a total density $n$,
211: with $r_s^2=1/\pi n$,
212: while the density in each valley $v=a$ or $b$,
213: is taken to be $n_v=n/2$. Hence the
214: $r_s$ parameter in each valley becomes
215: $r_{sv}=r_s\surd2$. Thus we do not consider density polrizations
216: leading to $n_a\ne n_b$. Also the electrons in both valleys have the
217: same spin polarization $\zeta$ and the same temperature $T$.
218: This is consistent with recent studies which show that the
219: valley splitting is very slight\cite{valpud}.
220: If the two spin species
221: be denoted by $i=1,2$, we have a four-component 2-DES with 10
222: independent pair distribution
223: functions (PDFs), viz., $g_{ij,vw}(r)$. We define $k=1,2$ for
224: the two spins in valley $a$,
225: and $k=3,4$ for the two spins in valley $b$, and write the PDFs as $g_{kl}(r)$.
226: The CHNC method for 2-DES has been described fully in
227: ref.~\cite{prl2}, where the quantum fluid at $T=0$ is considered to be equivalent to a
228: classical fluid at a quantum temperature $T_q(r_s)$.
229: Here, for the two valley system we use the same $T_q(r_s)$ as before,
230: and embodies the essential ``many-body'' input to the problem.
231: In brief outline, in CHNC we assume that the 2-D electrons are
232: mapped on to a classical system where the distribution functions are given by
233: a finite-$T$ classical density functional form:
234: \begin{equation}
235: \label{chnceq}
236: g_{kl}(r)=e^{-\beta\{P(r)\delta_{ij}\delta_{vw}+ V_{cou}(r) + V_c(r:[g_{kl}])\}}
237: \end{equation}
238: Here $\beta P(r)$ is a "Pauli exclusion potential" which acts only for parallel
239: spins, i.e, if $k=l$. It is
240: constructed such that $g_{kl}(r)$ becomes identical with the non-interacting
241: PDF, viz., $g^0_{kl}(r)$ which is known from quantum mechanics
242: when the Coulomb interaction
243: $ V_{cou}(r)$ and the associated correlation corrections $V_c(r)$ are zero. The Coulomb
244: interaction between two electrons in the equivalent classical picture
245: involves a correction arising
246: from their mutual diffraction effects.
247: Thus $ V_{cou}(r)$ is obtained by solving a two-electron
248: Schrodinger equation. The result is parametrized by the form\cite{prl2},
249: \begin{eqnarray}
250: \label{vcou}
251: V_{cou}(r)&=&(1/r)[1-exp(-k_{th}r)] \\
252: k_{th}/k_{dBr}&=&1.1587T_{cf}^{0.103} \\
253: k_{dBr}&=&(2\pi m^*T_{cf})^{1/2}, \; T^2_{cf}=(T^2_q+T^2)
254: \end{eqnarray}
255: Here $k_{dBr}$ is the de Broglie momentum of the scattering
256: pair with the effective pair mass $m^*=1/2$, and $T_{cf}$ is the
257: classical fluid temperature which reduces to $T_q$ at $T=0$. The correlation
258: potential $V_c(r)$ occurring in Eq.~\ref{chnceq} is taken to
259: be the sum of hyper-netted-chain
260: diagrams inclusive of a bridge term. Thus $V_c$ is nonlocal and is a
261: function of the $g_{kl}(r)$ which have to be self-consistently
262: calculated. The bridge term mimics the higher-order correlations
263: which are {\it not} captured by the simplest HNC equations.
264: These were shown to be important in
265: 2-D electron systems in reference ~\cite{prl2}.
266: Particles having identical indices ($k=j$)
267: are restricted from close approach by the Pauli
268: exclusion effect modeled by $P_{kl}(r)$. However,
269: singlet-pairs of electrons, or electrons in two different
270: valleys contribute to strong Coulomb
271: correlations, and hence a bridge term is included in all
272: such ``off-diagonal'' PDFs. The bridge term
273: $B_{kl}(r)$ for $k\ne l$ applies to 6 different PDFs, and we have taken this
274: to be given by the usual hard-disk functional form discussed in ref.\cite{prl2}.
275: (Totsuji and coworkers\cite{hard} have studied a more detailed
276: implementation of the
277: hard-disk bridge function in CHNC, while Bulutay et al.\cite{bulut} have
278: studied the 2-D CHNC without a bridge correction).
279: It should be emphasized that both the HNC approximation as well as the need for
280: a bridge function can be avoided by using the classical mapping to
281: a quantum fluid (CMQF) where we use classical molecular dynamics (MD)
282: to generate the PDFs etc., of the classical fluid under consideration.
283: In such a scheme we use the pair-potential given by Eq.~\ref{vcou} plus
284: the Pauli potential in an MD simulation for a
285: classical plasma
286: at the temperature $T_{cf}$. Such a CMQF-MD scheme would be numerically more
287: demanding than the CHNC, much simpler than the full QMC simulations,
288: and have the
289: advantage of not making the HNC+bridge approximations. However, the 2-valley
290: (4-component) system examined here has been studied by QMC and we use those
291: results to confirm the validity of our methods.
292:
293:
294: The main difference in the physics of the 1-valley system, and the
295: 2-valley system arises from the preponderence of direct
296: Coulomb interactions (from 6 PDFs in the 2-valley, one in the 1-valley)
297: over the exchange
298: interactions (from 4 PDFs in the 2-valley, two PDFs in the 1-valley).
299: This is the main reason for the lack
300: of a transition to a stable $\zeta=1$ state at low density.
301: Since the transition to a $\zeta=1$ state does not occur as
302: $r_s$ increases, the $g^*$
303: remains insensitive to increasing $r_s$,
304: as found theoretically (\cite{cmm})
305: and experimentally\cite{sash}.
306:
307: The 10 coupled equations for $g_{kl}(r)$ are
308: self-consistently solved for many
309: values of the coupling constant $\lambda$ applied to
310: the Coulomb interaction, and the
311: $g_{kl}(r:\lambda)$ are used in the adiabatic
312: connection formula to determine the
313: exchange-correlation free energy of the 2-valley 2DES.
314: While our calculations are easily
315: carried out for any value of $\zeta$, $T$ and $r_s$,
316: the 4-component QMC calculations
317: at finite $T,\zeta$ are
318: a major computational undertaking which has not been attempted.
319: However, at $T=0, \zeta=0$,
320: Conti and Senatore\cite{cs} have presented QMC results
321: for 2-D electron bilayers separated by
322: a distance $d_L$. They give total energies and also a fit to the
323: correlation energy/electron,
324: $\epsilon_c(r_s,\zeta=0,T=0)$ at $d_L=0$, i.e, the case where
325: both electron gases reside in the same layer.
326: In table~\ref{csdata} we compare the 4-component CHNC with
327: the available 4-component QMC data at $d_L=0$.
328: The energies $\epsilon_{c}(r_s)^{\scriptscriptstyle QMC}$,
329: are from the Rapisarda-Senatore fit-formula\cite{rapi} with
330: the parameters quoted in Table I of Conti et al\cite{cs}.
331:
332: \begin{table}
333: \caption{ Comparison of the total energy $\epsilon_{tot}(r_s)$
334: and and the correlation energy $\epsilon_{c}(r_s)$, in atomic units
335: at $T=0,\zeta=0$ for the 2-valley 2DES obtained from CHNC, with the
336: QMC data of Conti et al\cite{cs}.}
337: %see home go2d/Nfp2 xc4rz code which uses 90% k_th
338: \begin{ruledtabular}
339: \begin{tabular}{cccccc}
340: $r_s$ & $\epsilon_{tot}(r_s)^{\scriptscriptstyle QMC}$&
341: $\epsilon_{tot}(r_s)^{\scriptscriptstyle CHNC}$ &
342: $\epsilon_{c}(r_s)^{\scriptscriptstyle QMC}$ &
343: $\epsilon_{c}(r_s)^{\scriptscriptstyle CHNC}$\\
344: \hline \\
345: 2.0 & -0.29302 & -0.29172 &-0.14315 &-0.14202 \\
346: 10.0 & -0.08611 & -0.08647 &-0.04607 &-0.04649 \\
347: 20.0 & -0.04641 & -0.04643 &-0.02577 &-0.02581 \\
348: 30.0 & -0.03196 & -0.03183 &-0.01806 &-0.01795 \\
349: \label{csdata}
350: \end{tabular}
351: \end{ruledtabular}
352: \end{table}
353: %table 1
354:
355: These results show that the CHNC method provides a simple and
356: accurate approach to the treatment of exchange and correlation in
357: the 4-component system. We exploit the simplicity of CHNC for calculations
358: at finite-$T$ and $\zeta$ which are currently too expensive for
359: QMC simulations. However, in situations where QMC results are
360: available for the correlation energies, we adopt the QMC parametrizations.
361: Thus one may use the parametrization of $\epsilon_c$ given by Conti and
362: Senatore for the $T=0, \zeta=0$ case. For $r_s>1$ applications the
363: Tanatar-Ceperley\cite{tc} form may also be used for the 2-valley
364: system as well, with the parameter values
365: $a_0=-0.40242,\, a_1=1.1319, \, a_2=1.3945, \, a_3=0.67883$, fitted
366: to a data base from $r_s$ =1 to 30.
367:
368: The $T=0,\,\zeta=1$ case is particularly interesting since this system
369: (i.e, 2-valley system at density $n$) is
370: mathematically identical to the 1-valley system at the same density
371: but with $\zeta=0$, for the Hamiltonian considered by Conti et al.,
372: and by us in this study. In this case the 2-valley system has a two-fold symmetry
373: since the energy is the same irrespective of the orientation of the spin
374: in each valley. That is, $\zeta=1$ means all the spins in valley $a$ are oriented,
375: while all the spins in valley $b$ are also oriented, but independently of
376: the orientation of the spins in $a$. This degeneracy would be resolved in
377: real Si-MOSFETS but not in the model used here, or in Conti and Senatore.
378: For instance, the three-body correlations for inter-valley interactions
379: may be slightly different from those in the {\it intra}-valley interactions,
380: and hence may require two different bridge parameters, to be determined
381: variationally by an energy minimization using the hard-disk
382: reference fluid approach. We have not done this, and simply used
383: the same bridge parameter as in ref.~\cite{prl2} for all interactions.
384: In the QMC calculation this would require independent optimization of the
385: model for back-flow corrections. Finally,
386: the correlation energy of the fully spin-polarized (degenerate) 2-valley system
387: can be parametrized using the Tanatar-Ceperley form with
388: $a_0=-0.19162,\, a_1=3.6123, \, a_2=1.9936, \, a_3=1.4714$, in atomic units.
389: %
390: \subsection{The energy of unpolarized and polarized phases.}
391: %
392: The $T=0$ correlation energy at finite values of $\zeta$ were
393: calculated using
394: the CHNC procedure and compared with the
395: values predicted from the polarization factor used for the
396: one-valley 2DES. This has the form\cite{prl2}
397: \begin{eqnarray}
398: \label{spinpol}
399: p(r_s,\zeta) &=& \frac{\epsilon_c(r_s,\zeta)-\epsilon_c(r_s,0)}
400: {\epsilon_c(r_s,1)-\epsilon_c(r_s,0)}
401: \,\, = \frac{\zeta_+^{\alpha(r_s)}+\zeta_-^{\alpha(r_s)}-2}
402: {2^{\alpha(r_s)}-2}\nonumber\\
403: \alpha(r_s)&=&C_1-C_2/r_s+C_3/r_s^{2/3}-C_4/r_s^{1/3}
404: \end{eqnarray}
405: Here, $\zeta_{\pm}$=$(1\pm\zeta)$.
406: It turns out that the coefficients $C_1-C_4$ obtained
407: for the one-valley 2DES, i.e.,
408: 1.5404, 0.030544, 0.29621, and 0.23905 respectively, work quite
409: well for the 2-valley system as well,
410: even at high $r_s$.
411: Thus, using the
412: TC-type fit formulae for the 2-valley $\epsilon(r_s,\zeta=1)$ and
413: $\epsilon_c(r_s,\zeta=0)$, the estimated values of $\epsilon_c(r_s,\zeta)$
414: and the CHNC values are given
415: in Table~\ref{polfit}.
416: %
417: \begin{table}
418: \caption{ The correlation energy $\epsilon_{c}(r_s,\zeta)$
419: per electron, as a function of $\zeta$, estimated using the 1-valley
420: polarization factor of Eq.~\ref{spinpol}, and from the full 2-valley
421: CHNC calculation}
422: \begin{ruledtabular}
423: \begin{tabular}{cccccc}
424: $r_s$ & $\epsilon_{c}^{fit}$&
425: $\epsilon_{c}^{\scriptscriptstyle CHNC}$ &
426: $\epsilon_{c}^{fit}$ & $ \epsilon_{c}^{\scriptscriptstyle CHNC}$\\
427: \hline \\
428: $\;\;\;\;\zeta \to$&0.25 & 0.25& 0.75 & 0.75 &\\
429: \hline \\
430: 5.0 & -0.07686 & -0.07757 &-0.06286 & -0.06434\\
431: 10.0 & -0.04518 & -0.04562 &-0.03765 & -0.03831 \\
432: 20.0 & -0.02529 & -0.02535 &-0.02134 & -0.02151 \\
433: 30.0 & -0.01772 & -0.01764 &-0.01477 & -0.01504 \\
434: \label{polfit}
435: \end{tabular}
436: \end{ruledtabular}
437: \end{table}
438: %table 3
439:
440: The finite-$\zeta$
441: calculations show that there is {\it no ferromagnetic phase transition}
442: in this system at $T=0$, since the total energy of the $\zeta=0$ phase
443: is always more negative than any polarized phase. This is expected from the
444: dominance of the many off-diagonal terms contributing direct
445: Coulomb interactions, but no exchange interactions.
446: This was pointed
447: out in ref.~\cite{cmm} where the insensitivity of $m^*g^*$ to $\zeta$
448: obtained from the theory was in excellent agreement with the experiments
449: of Shashkin et al\cite{sash}. The
450: stabilization energy $\Delta E$ of the $\zeta=0$ phase with respect to
451: the fully polarized phase is 0.12734$\times 10^{-3}$ a.u.
452: at $r_s$=25, and diminishes
453: to 0.89929$\times 10^{-4}$ a.u. at
454: $r_s=40$. These are very small energy differences
455: and within the error of the CHNC method and possibly of the 2-valley QMC calculations.
456: However, the pattern of
457: stability of the $\zeta=0$ phase holds for
458: all $r_s$ investigated. Conti and Senatore\cite{cs} find a Wigner crystal
459: phase at $r_s=42$ and claim that ``in the intermediate regime the 4-component
460: 2-DES can be expected to mimic the 2DES $\cdots$, with the appearance of
461: a spin-polarized ground state''. We arrive at the opposite conclusion.
462: %
463: \subsection{LDA-type calculations for Si-MOSFETS.}
464: %
465: In most density-functional calculations of Si/SiO$_2$
466: quantum wells, the
467: Kohn-Sham exchange-correlation potential $V_{xc}(r)$
468: is calculated using the local
469: density approximation (LDA) where the total density
470: $n(r)$ is considered without
471: taking account of the valley degeneracy. In effect, the electron gas is assumed
472: to be a {\it single} electron gas at a density $r_s$ and its exchange-correlation
473: energy $\epsilon_{xc}(r_s)$ and the Kohn-Sham potential $V_{xc}(r_s)$ are calculated
474: at the given density.
475: (We recall that $V_{xc}(r_s)$ is simply the xc-contribution to the
476: chemical potential $\mu_{xc}$).
477: Results for $\epsilon_{xc}$ and $\mu_{xc}$ for electrons in a Si-MOSFET calculated
478: correctly, i.e., taking account of its degenerate valley structure, and in the usual
479: LDA approach are compared in Table~\ref{lda}.
480: The full chemical potential $\mu$, as well as the compressibility ratio $K^0/K$
481: calculated for the 2-valley system, and that obtained within the LDA approach,
482: are also shown in Fig.~\ref{muk}. It is noteworthy that
483: the total chemical potential has a minimum near $r_{sm}\sim 1$. In effect, a low density
484: electron fluid ($r_s > r_{sm}$) whose chemical potential is
485: equal to that of a high density
486: gas ($r_s < r_{sm}$) exists and this could lead to spontaneous
487: density inhomgenieties in these systems.
488: %The figure shows $K^0/K$ which is the samll-$k$ limit of the LFF of the
489: %coupled response (2-valley response), here determined directly from the
490: %total energy of the 2-valley system.
491: %
492: %
493: \begin{table}
494: \caption{ Comparison of the exchange-correlation energy $\epsilon_{xc}(r_s)$
495: per electron, and the Kohn-Sham potential $V_{xc}(r_s)$, in atomic units
496: at $T=0,\zeta=0$ for the 2-valley 2DES obtained from QMC-fit or CHNC, and from
497: the LDA.}
498: \begin{ruledtabular}
499: \begin{tabular}{cccccc}
500: $r_s$ & $\epsilon_{xc}(r_s)^{LDA}$& $\epsilon_{xc}(r_s)^{QMC}$ &
501: $V_{xc}(r_s)^{LDA}$ & $ V_{xc}(r_s)^{QMC}$\\
502: \hline \\
503: 2.0 & -0.38213 & -0.35535 &-0.55189 & -0.50271 \\
504: 10.0 & -0.09007 & -0.08851 &-0.13133 & -0.12835 \\
505: 20.0 & -0.04742 & -0.04699 &-0.06957 & -0.06874 \\
506: 30.0 & -0.03241 & -0.03221 &-0.04769 & -0.04730 \\
507: \label{lda}
508: \end{tabular}
509: \end{ruledtabular}
510: \end{table}
511: %table 3
512:
513: When the actual electron densities in Si-MOSFETS are converted to effective $r_s$
514: units (see below), the $r_s$ range 1-6 is the most important for device
515: applications, and hence overestimates in $V_{xc}$ contained in the usual LDA
516: approach could be significant.
517:
518: %
519: \begin{figure}
520: %\includegraphics*[width=9.0cm, height=10.0cm]{fig1v2.ps}
521: \includegraphics*[width=9.0cm, height=10.0cm]{muk.ps}
522: \caption
523: {Left panels: Comparison of the total chemical potential $\mu$
524: i.e, the total Kohn-Sham potential, calculated at
525: the total density $n$ and $r_s$, for the 2-valley system,
526: and if the LDA were used (labeled 1-valley), ignoring the 2-valley
527: nature. Right panels: Same for the compressibility ratio}
528: \label{muk}
529: %fig 1
530: \end{figure}
531: %
532: %
533: \subsection{Pair-distribution functions in the Si-MOSFET system}
534: %
535: The PDFs, denoted by $g_{kl}(r)$, embody the detailed
536: particle correlations in
537: the system. In Fig.~\ref{pdf} we display an illustrative set of
538: pair-distribution
539: functions. QMC-based
540: PDFs have not been reported in the literature and hence we do not have a
541: direct comparison. However, good agreement between CHNC and QMC-
542: based PDFs have been found in other systems (e.g, 2DES, 3DES, and fluid
543: hydrogen)\cite{prl2,prl1,hyd}.
544: For $\zeta=0$ the four
545: diagonal PDFs $g_{kk}$ are
546: identical, and similarly, all the six
547: off-diagonal PDFs are also identical.
548: Hence there are actually only two distinct PDFs, just
549: as in the single-valley case where $g_{11}$ and $g_{12}$ define the $\zeta=0$
550: case. These are shown for $r_s=2$, 5 and 10 in the top
551: panel of Fig.~\ref{pdf}.
552: It is also clear that the correlation effects are mainly
553: determined by the off-diagonal PDFs. At finite $\zeta$ there are five
554: independent PDFs. There are two independent diagonal PDFs
555: $g_{11}=g_{33}$ and $g_{22}=g_{44}$. The three independent off-diagonals
556: are $g_{12}=g_{23}=g_{34}=g_{14}$, $g_{13}$ and $g_{24}$.
557: These are shown for the case $r_s=10$ and $\zeta=0.5$ in the
558: lower panel of Fig.~\ref{pdf}.
559: %
560: %
561: \begin{figure}
562: %\includegraphics*[width=9.0cm, height=10.0cm]{fig1v2.ps}
563: \includegraphics*[width=9.0cm, height=10.0cm]{gr.ps}
564: \caption
565: {Pair distribution functions of the two-valley system. For $\zeta=0$
566: the 10 PDFs reduce to two indeependent PDFs. For $\zeta \ne. 0$
567: there are 5 independent PDFs.
568: }
569: \label{pdf}
570: %fig 1
571: \end{figure}
572: %
573: %
574:
575: %
576: \section{Response functions.}
577: %
578: \label{resp}
579: %
580: In the following we discuss the linear response functions
581: since the static small-$k$ limit can be related to the
582: derivatives of the total energies that were calculated from CHNC or QMC
583: if available.
584: This enables us to verify a simple procedure for the construction
585: of the 2-valley response functions using {\it only} single-valley
586: exchange correlation data\cite{cmm}. The density-density response
587: function $\chi(k,\omega)$ will to be called the {\it d-d} response for brevity.
588: It is expressed
589: in terms of a reference ``zeroth-order''
590: $\chi^0_R(k,\omega)$ and a local-field factor (LFF),
591: denoted by $G(k,\omega)$.
592: \begin{equation}
593: \label{lffdef}
594: \chi(k,\omega)=\chi^0_R(k,\omega)/[1-V_k\{1-G_d(k,\omega)\}\chi^0_R(k,\omega)]
595: \end{equation}
596: The {\it s-s} response (or ``spin susceptibility'') is written as
597: \begin{equation}
598: \label{spindef}
599: \chi_s(k,\omega)=-\mu_B^2\chi^0_R(k,\omega)/[1-V_k\{1-G_s(k,\omega)\}\chi^0_R(k,\omega)]
600: \end{equation}
601: where $\mu_B$ is the Bohr magneton. Note that our definition of
602: the spin-LFF differs somewhat from a commonly used
603: definition\cite{santoro}.
604: Our form makes the {\it d-d} and
605: {\it s-s} LFFs appear formally similar, at least at this stage.
606: Hence a single discussion applies
607: to both, and we drop the subscripts $d$ and $s$.
608: The reference function $\chi^0_R(k,\omega)$ has the form:
609: \begin{equation}
610: \chi^0_R(k,\omega)=\sum_{\vec{k},\sigma} \frac{n_{\sigma,\vec{k}}
611: -n_{\sigma,\vec{k}+\vec{q}}}
612: {\omega+\epsilon_{\vec{k}}-\epsilon_{\vec{k}+\vec{q}}}
613: \end{equation}
614: Here $\vec{k},\vec{q}$ are two-dimensional vectors, while the corresponding
615: single-particle energies are denoted by $\epsilon_{\vec{k}}$ etc. The
616: Fermi occupation number $n_{\sigma,\vec{k}}$ may be chosen to be the
617: non-interacting value, in which case $\chi^0$ is the 2-D Lindhard
618: function\cite{afs}. An alternative choice is to use the
619: fully-interacting density, evaluated from the fully-interacting
620: chemical potential, as
621: in density-functional theory (DFT).
622:
623: %The local-field factors $G_d(k,\omega)$ and $G_s(k,\omega)$, appearing in
624: %the two response functions can be written in terms of the
625: % static structure factors
626: %of the 2DES using CHNC theory.
627: The small-$k$ limits of the local-field factors $G_d(k,\omega)$ and $G_s(k,\omega)$
628: can be obtained
629: from the second derivatives
630: of the exchange-correlation free-energy functional $F_{xc}(n,\zeta)$ of DFT,
631: with respect to
632: the charge densities or the spin densities\cite{iwamoto}.
633: These second-order derivatives, together
634: with the second derivative of $F_{xc}(n,\zeta)$ with respect to $T$
635: for a two-valley system were used in our $m^*$ and $g^*$ calculations
636: reported in ref.\cite{cmm}, using the CHNC technique.
637: Here we look at the compressibility and spin-susceptibility
638: of the 2-valley system obtained from
639: the small-$k$ limit of the coupled-mode response function (built up
640: from 1-valley data) and compare it
641: with that obtained directly from the 2-valley CHNC and QMC.
642: %We also report the $k$-dispersion of the static LFFs of the two valley system
643: %and compare them with the $G(k)$ of the simple 2DES\cite{euplfc}.
644: %%
645: \subsection{Response functions of the two-valley system.}
646: %The simple 2DES in GaAs structures are themselves 2-component systems since
647: %there are two spin species present. Thus, in our CHNC calculations, we
648: %determined the pair distribution functions (PDFs), i.e.,
649: % $g_{ij}(r)$, where $i$, or $j$
650: %are spin indices (1,2). In a one-valley system we need
651: %the three PDFs $g_{11},\,g_{12}=g_{21}$ and $g_{22}$,
652: % and the corresponding free energy contributions $F_{11}, F_{12}$,
653: %and $F_{22}$. The energy evaluation requires the evaluation of $g_{ij}(r)$
654: % as a function of
655: %the Coulomb coupling constant, for each $r_s,\zeta$, and $T$, as
656: %discussed in
657: %Ref.\cite{prl3}.
658: %Once the $g_{ij}(r)$ are obtained, CHNC theory also affords the possibility
659: %of extracting the corresponding two-component $\zeta$ and $T$ dependent
660: %LFFS quite straight-forwardly, since the classical response functions
661: %are directly related to the PDFs. Explicit results are displayed in a
662: %later section.
663: %
664: The theory of the one-valley fluid can be used for the two-valley
665: (4-component) fluid if there
666: is no valley polarization (i.e, the two valleys are assumed equivalent
667: although distinct), as in
668: ref.\cite{cmm}. As this may not be completely
669: clear from the abbreviated discussion in ref.\cite{cmm}, we
670: present some details here.
671:
672: In the theory of classical fluids, the response functions are
673: simply related to the structure factors, while the LFFs are
674: simply related to the direct correlation functions
675: of Ornstein-Zernike(OZ) theory. Since this paper is directed
676: more towards electron-fluid studies, we follow the language
677: of the LFFS and the related response methodology rather than the
678: OZ presentation.
679:
680: Let us indicate the species
681: (which may be a valley index or a spin index)
682: by $u$ or $v$, taking the values 1,2, and let
683: us consider a weak external potential $\phi_v(\vec{k},\omega)$ which
684: acts only on the electrons of species $v$. The external potential induces
685: density deviations $\delta n_v(\vec{k},\omega)$ such that\cite{vashista}:
686: \begin{equation}
687: \label{linres}
688: \delta n_u(\vec{k},\omega)=\sum_v
689: \chi_{uv}(\vec{k},\omega)i\phi_v(\vec{k},\omega)
690: \end{equation}
691: These equations define the linear {\it d-d}
692: response functions involving the
693: species $u$ and $v$. The longitudinal
694: dielectric function $\varepsilon(\vec{k},\omega)$ is now given by
695: \begin{equation}
696: 1/\varepsilon(\vec{k},\omega)=1+V_k\sum_{u,v}\chi_{uv}(\vec{k},\omega)
697: \end{equation}
698: Here $V_k$ is the 2-D Coulomb interaction $2\pi/k$. Note that we are using
699: effective atomic units (Hartrees etc.) such that
700: $e^2$ divided by the background dielectric constant
701: is taken to be unity (see section~\ref{n1n2}).
702: To relate the response functions to the local fields,
703: we consider the effective potentials $U_v(\vec{k},\omega)$ such that:
704: \begin{equation}
705: U_v(\vec{k},\omega)=V_k[1-G_{uv}(\vec{k},\omega)]\delta n_v(\vec{k},\omega)
706: \end{equation}
707: Thus the bare Coulomb interaction between the particles of type $u$ and $v$
708: is
709: modified by the LFFs $G_{uv}$. Hence we can write the density deviations
710: $\delta n_u(\vec{k},\omega)$ in terms of the effective potentials and the
711: zeroth order response functions as follows (we drop the $\vec{k},\omega$ labels for brevity).
712: \begin{equation}
713: \label{effective}
714: \delta n_u=\chi^0_u[\phi_u
715: +V_k\sum_v (1-G_{uv})\delta n_v]
716: \end{equation}
717: Now, by a comparison of the equations \ref{linres} and\ref{effective}, we can
718: write down the response functions of the coupled two-component system in terms
719: of the zeroth-order response functions and the LFFs.
720: \begin{eqnarray}
721: \label{coupled}
722: \chi_{11}&=&\chi_1^0\,d_2/D,\,\,\chi_{22}=\chi_2^0\,d_1/D, \\
723: \chi_{12}&=&V_k\chi_1^0\chi_2^0[1-G_{12}]/D \\
724: %\chi_{21}&=&V_k\chi_1^0\chi_2^0[1-G_{21}]/D \\
725: d_1&=&1-V_k\chi_1^0[1-G_{11}]\\
726: %d_2&=&1-V_k\chi_2^0[1-G_{22}]\\
727: d_{12}&=&V_k\chi_1^0[1-G_{12}]\\
728: %d_{21}&=&V_k\chi_2^0[1-G_{21}]\\
729: D&=&d_1 d_2 -d_{12}d_{21}
730: \end{eqnarray}
731: We have suppressed the $\vec{k},\omega$ dependence in the above
732: equations, and also not explicitly given $\chi_{21}$, $d_2$ and $d_{21}$ for brevity.
733: We now define a total coupled-mode response function $\chi_T(\vec{k},\omega)$ via
734: \begin{equation}
735: 1/\varepsilon(\vec{k},\omega)=1+V_k\chi_T(\vec{k},\omega)
736: \end{equation}
737: Then the total two-component 2DES response is given by:
738: \begin{eqnarray}
739: \label{sumichi}
740: \chi_T&=&[\chi_1^0+\chi_2^0+V_k\chi_1^0\chi_2^0G_{\Sigma}]/D\\
741: G_{\Sigma}&=&G_{11}+G_{22}-G_{12}-G_{21}
742: \end{eqnarray}
743: In the following we some times denote $\chi_T$ by $\chi_{cm}$ to
744: emphasize the coupled-mode nature of the total response.
745: If we are dealing with a simple (one-valley) electron fluid,
746: e.g., a partially spin-polarized electron gas,
747: the species index $v$ is simply the
748: spin index.
749: Notice that the coupling between the two systems (be they spins or valleys),
750: replaces the individual denominators $d_1$ and $d_2$ by a new denominator
751: $D$, common to both systems, and containing the cross terms $d_{ij}$.
752: That is, instead of the two sets of excitations given by the zeros
753: of $d_1$ and $d_2$, we now have a {\it common set
754: of ``coupled-mode'' excitations}
755: defined by the zeros of $D$.
756: We emphasize that in this analysis we have {\it not }
757: used any form of CHNC theory.
758:
759: All the response functions prior
760: to the switching-on of the Coulomb interaction between the
761: two valleys are known. The problem is to determine the
762: cross terms $d_{ij}$, i.e, the inter-valley term, $d_{uv}$, occurring
763: in the coupled-mode denominator $D$, using only the free energy data
764: for a single valley.
765: If we consider the
766: small-$k$ limit, we see that the terms $d_{ij}$ occurring in
767: $D$ are directly related to the second density derivative or magnetization
768: derivative of the free energy contributions $F_{ij}$ arising
769: from the PDFs $g_{ij}$. We know these {\em individual}
770: free energy contributions for the one-valley problem.
771:
772: Let us first consider the small-$k$
773: limit of the static response functions to make contact
774: with the compressibility and susceptibility sum rules.
775:
776: \subsection{Small-$k$ limit of the static response.}
777: \label{smallk}
778: The small-$k$ behaviour of the static response is
779: related to the second derivative with respect to the
780: density (or magnetization) and this provides well known
781: sum rules that we exploit here.
782: For simplicity, and for comparison with
783: the degenerate two-valley case, let us review the
784: one-valley paramagnetic case $\zeta=0$,
785: and consider the calculation\cite{cmm} of the small-$k$,
786: static ($\omega=0$) limit of the simple (one-valley) response function.
787: \begin{equation}
788: \chi_v(n_v)=\chi^0(n_v)/[1-V_k(1-G_{vv})\chi^0(n_v)].
789: \end{equation}
790: The density-density response function is associated with the
791: proper polarization function $\Pi_v$. Dropping the
792: species subscript $v$ for the present, we have
793: \begin{eqnarray}
794: \Pi &=&\Pi^0/(1+V_kG\Pi^0)\\
795: \Pi^0&=&-\chi^0
796: \end{eqnarray}
797: The small-$k$ behaviour of this function states that
798: \begin{equation}
799: \Pi/\Pi_0=\kappa/\kappa_0
800: \end{equation}
801: The compressibility $\kappa$ is calculated via the
802: chemical potential $\mu$, starting from the total free
803: energy per unit volume obtained from the CHNC calculation.
804: \begin{eqnarray}
805: F&=&F_0+F_x+F_c\\
806: F&=&\sum_v n_v[\epsilon_0^v(n_v)+\epsilon_x^v(n_v)]+n\epsilon_c(n)\\
807: \epsilon_0^v&=&(1+\zeta^2)/2r_{sv}^2\\
808: \epsilon_x^v&=&-\frac{2\surd{2}}{3\pi r_{sv}}[(1+\zeta)^{3/2}+(1-\zeta)^{3/2}]\\
809: \mu&=&\frac{dF}{dn}=\mu_0+\mu_x+\mu_c\\
810: 1/\kappa&=&n^2\frac{d\mu}{dn}
811: \end{eqnarray}
812: At $T$=0, the chemical potential is given (in Hartrees) by:
813: \begin{equation}
814: \mu=n_v\pi-2(2/\pi)^{1/2}n_v^{1/2}+\mu_c
815: \end{equation}
816: The compressibility calculated directly from the 4-component calculation
817: should agree with that obtained from the coupled-valley formalism.
818: There, the small-$k$ limit of the denominator of the
819: density-density proper-polarization function is given by
820: $$1+V_kG_d\chi^0=\kappa/\kappa_0$$
821: Here $G_d$ is the LFF of the density-density polarization function.
822: Hence the denominator $d_1$, or $d_2$
823: of the $d$-$d$ response occurring in Eq.\ref{coupled},
824: for any particular species is available for the density-density
825: response function in each valley. But the cross
826: density-density LFFs, e.g., $G_{12}$, and the cross-denominators
827: $d_{12}$ needed to form the coupled-valley forms are not yet specified.
828:
829: In the case of the spin susceptibility $\chi_s$, the role played
830: by the compressibility is taken over by the spin-stiffness, which
831: is the second derivative of the free energy $F$ with respect to
832: the spin polarization $\zeta$. Here we have, for a single
833: valley,
834: \begin{equation}
835: \label{chirat}
836: \chi_s/\chi_P=1+d^2\{r_{sv}^2 F_{xc}\}/d\zeta^2
837: \end{equation}
838: Hence the denominators $d_u$ and $d_v$ of the spin-susceptibilities
839: of the each 2DES are known, at the valley densities
840: $n_v=n_u=n/2$, from a simple CHNC calculation or from a
841: QMC energy parametrization. However, here again the cross terms
842: $d_{12}$ and $d_{21}$, (which are equal)
843: needed to complete the calculation of
844: the coupled susceptibility (Eqs.~\ref{coupled},\ref{sumichi}) are
845: not yet specified.
846:
847: The cross term for the $d$-$d$ response, or for the $s$-$s$ response
848: can be calculated if the free-energy contribution $F_{uv}$ arising
849: from the Coulomb interaction among the electrons in the two
850: valleys is known. The interaction is among the electrons of
851: valley $u$, at density $n/2$, and the electrons of
852: valley $v$ at density $n/2$.
853: The inter-valley free-energy contribution $F_{uv}(n/2,n/2)$
854: is purely Coulombic, and hence it is
855: clearly analogous to the correlation free-energy term
856: arising from the antiparallel-spin PDF, i.e., $g_{12}(n,\zeta=0)$ of the
857: simple one-valley 2DES {\it at density} $n$ with two spin species.
858: The case $\zeta=0$ ensures
859: that the total density is split as $n_u=n_v=n/2$.
860: Thus the $d_{12}$ term needed for the spin-susceptibility calculation
861: and the
862: density-density response calculations are:
863: \begin{eqnarray}
864: \label{crossterm}
865: %\mbox{s-s}\; d_{12}&=&\frac{d^2\{rs^2 F_c[g_{12}]\}}{d\zeta^2} \\
866: %\mbox{d-d}\; d_{12}&=&\frac{d^2 F_c[g_{12}]}{dn^2}
867: %\mbox{s-s}\;
868: s-s \;\; d_{12}&=&d^2\{r_s^2 F_c[g_{12}]\}/d\zeta^2 \\
869: %\mbox{d-d}\;
870: d-d \;\; d_{12}&=& -(2/\pi)d^2F_c[g_{12}]/dn^2
871: \end{eqnarray}
872: Note that $d_1$ is calculated from $F_{xc}(n/2,\zeta=0)$, while $d_{12}$ is
873: from $F_c(n,\zeta=0)$ of the simple one-valley 2DES.
874: Hence, knowing $d_1$, $d_{12}$, (which are equal to $d_2$, and $d_{21}$
875: since the valleys are degenerate),
876: we can calculate the susceptibility enhancement $\chi_s/\chi_P$, as
877: well the compressibility ratio $\kappa/\kappa_0$ of the interacting
878: 2-valley 2DES, without actually solving the coupled system of 10 distribution
879: functions needed in the full CHNC calculation of the 4-component system.
880: %
881: %
882: \begin{figure}
883: %\includegraphics*[width=9.0cm, height=10.0cm]{fig1v2.ps}
884: \includegraphics*[width=9.0cm, height=10.0cm]{eck0k.ps}
885: \caption
886: { (a) Comparison of the compressibility ratio $\Pi_0/\Pi_{cm}=K^0/K$ obtained
887: from the coupled-mode approach, and from the second-density derivative of the
888: total free of the 4-component system. (b) The energy estimate $\epsilon_c(rs,\zeta=0)$
889: from the coupled-mode form, and from the 4-component QMC\cite{cs}.
890: }
891: \label{eck0k}
892: %fig
893: \end{figure}
894: %
895:
896: %
897: \subsubsection{Results for the compressibility ratio and the susceptibility ratio}
898: %
899: we consider the compressibility ratio $K_0/K$
900: obtained by the coupled-mode analysis and from the 2-valley QMC data\cite{cs}, or
901: equivalently, from the 4-component CHNC data, in Fig.~\ref{eck0k}.
902: The excellent agreement shows that the coupled-mode procedure for using the
903: 1-valley data to generate 2-valley data is successful.
904:
905: The calculation of the susceptibility enhancement was given in
906: ref.~\cite{cmm}. We consider the susceptibility
907: enhancement in more detail. Equation~\ref{chirat} involves the
908: second $\zeta$- derivative of the correlation energy.
909: It is well known that the $\zeta$- dependent QMC calculations
910: are more prone to errors since a whole shell of spins need to be reversed.
911: In fact, the 1-valley 2DES calculations of
912: Rapisarda and Senatore\cite{rapi} using Diffusion Monte Carlo (DMC)
913: predict a value of $r_s\sim 20$ for the spin transition,
914: different from that ($r_s\sim 26$)
915: predicted by Attacalite et al.~\cite{atta} who also use very similar DMC
916: methods. This is an indication of the type
917: of uncertainty that may be had in $\zeta$-dependent QMC calculations.
918: No $\zeta$-dependent QMC data are available for
919: the 2-valley system. Using Eq.~\ref{spinpol}
920: we can write an {\it approximate} explicit form at $T=0$:
921: \begin{eqnarray*}
922: \label{epsderiv}
923: \epsilon_c(r_s,\zeta)&=&\epsilon_c(r_s,0)+
924: (\epsilon_c(r_s,1)- \epsilon_c(r_s,0))p(r_s,\zeta) \\
925: \frac{d^2 \epsilon_c(r_s,\zeta)}{d\zeta^2}&=&\Delta E(1,0)
926: \frac{d^2 p(r_s,\zeta)}{d\zeta^2}\\
927: \end{eqnarray*}
928: Thus the energy difference between the polarized phase and the unpolarized
929: phase appears directly. This becomes zero in systems like the one-valley 2-DES
930: that show a spin transition. Even in the 2-valley system where there is
931: no transition to a stable $\zeta=1$ state, we can expect the calculated
932: $\chi_s/\chi_P$ or $\chi_{cm}/\chi_P$ to be quite sensitive to $\Delta E(1,0)$,
933: and to the details of the form of the $\zeta$- dependent function.
934: We find that this is very much the case. In Fig.~\ref{mgtc} we display
935: the coupled-mode evaluation of $m^*g^*=\chi_{cm}/\chi_P$ with the
936: value of $\chi_s/\chi_P$ obtained from the $\zeta$-second derivative of the
937: energy obtained from the full 4-component calculation.
938: We give curves labeled ``4-component (a), and (b)'', where
939: the contribution from the $d^2\epsilon_c/d\zeta^2$- derivative
940: differs by $\sim$5\%. Clearly, this small change has a drastic
941: effect on the $m^*g^*$ evaluation.
942: %This suggests that
943: %the coupled-mode formulation is accurate enough.
944: % Noting that $\epsilon_c(r_s,\zeta=1)$
945: %of the 2-valley system was taken to be identical to the one-valley
946: % $\epsilon_c(r_s,\zeta=0)$, we compare the
947: In the lower panel of Fig.~\ref{eck0k} we compare the 2-valley correlation
948: energy $\epsilon_c(r_s, \zeta=0)$ from the coupled-mode analysis and from the
949: direct 4-component QMC calculation. As expected, a small difference appears
950: at low densities, while the accuracy is good near $r_s\sim 5$
951: where the rapid increase in $m^*g^*$ occurs. This type of
952: error is quite within the
953: errors that are possible in the 4-component QMC or the 4-component CHNC,
954: and in fitting to the polarization dependence of the numerically
955: calculated correlation energies. Coupled-mode formation implies that the
956: excitation spectrum of the system no longer shows the features of the
957: individual valleys, and hence is consistent with the conclusions of
958: ref.\cite{valpud} where no evidence for intervalley scattering was seen.
959: %
960: %
961: \begin{figure}
962: %\includegraphics*[width=9.0cm, height=10.0cm]{fig1v2.ps}
963: \includegraphics*[width=9.0cm, height=10.0cm]{mgtc.ps}
964: \caption
965: {Comparison of the suscepribility enhancement $\chi_{cm}/\chi_P=m^*g^*$ obtained
966: from the coupled-mode approach and from the second-$\zeta$ derivative of the
967: total free of the 4-component system. The curves marked
968: ``4-components (a),~(b)''
969: differ by $\sim$5\% in the value of the
970: second $\zeta$- derivative of the correlation energy,
971: showing the strong sensitivity to this energy derivative.
972: }
973: \label{mgtc}
974: %fig 1
975: \end{figure}
976: %
977: %
978: \section{Relation between density $n$ and the $r_s$ parameter in $Si$-MOSFETS.}
979: \label{n1n2}
980: %
981: In the 2DES of GaAs/AlAs- based structures, the dielectric constants of
982: the two materials are nearly identical. The lattice constants are also well
983: matched and hence the calculation of the effective atomic units needed in
984: converting the experimental density $n$ to the effective electron-disk radius
985: ($r_s$)
986: can be carried out in an unambiguous, transparent way. This is
987: important since many-body theory is formulated
988: within the language of $r_s$ and
989: effective atomic units.
990:
991: Unlike in $GaAs$, the situation for $Si$-MOSFETS is more
992: complicated. The
993: lattice mismatch between crystalline Si and most crystalline varieties
994: (e.g., crystobalite) of
995: SiO$_2$ turn out to be 35-40\%. The dielectric constants
996: are also strongly mismatched,
997: being $\sim 11.5$ and $\sim 3.9$ for Si and SiO$_2$. The large lattice
998: mismatch ensures that there is {\it no} sharp Si/SiOi$_2$ interface. The
999: reconstruction of the $Si$ atomic layers between the crystalline-Si and
1000: the SiO$_2$ bulk-like region still contain tetrahedral-bonding networks, but
1001: with strongly modified bond angles, bond lengths etc., characteristic of the
1002: amorphization of the $Si$ layers immediately adjacent to the oxide layer.
1003: Many decades of experimental and theoretical work has gone into sharpening
1004: our understanding of this interface. More recently,
1005: first principles
1006: density-functional calculations by Carrier et al.\cite{pierre},
1007: starting from tight-binding models\cite{nacir}
1008: have presented
1009: a clearer picture of the atomic arrangements near the interface region.
1010: A series of similar studies by Pasquerello et al.\cite{pasq} establish the
1011: geometry of the Si/SiO$_2$/vacuum interface.
1012: Thus a reliable atomic model of the Si/SiO$_2$ interface
1013: obtained via geometry optimization of the total energy
1014: is now available\cite{pierre}. The essential
1015: point is that
1016: the Si/SiO$_2$ interface contains approximately 5 regions containing
1017: crystalline Si (c-$Si$), amorphized Si (a-$Si$), suboxide layers,
1018: amorphized silicon dioxide (a-$SiO_2$),
1019: and crystalline $SiO_2$ . These are indicated
1020: schematically below:
1021: \begin{equation}
1022: [001]\,(z\to)|\mbox{c-}Si|\mbox{a-}Si| \mbox{suboxides}
1023: | \mbox{a-}SiO_2 |\mbox{c-}SiO_2|
1024: \end{equation}
1025: The amorphous (or bond distorted) regions of a-$Si$
1026: do not have a ``conduction band''
1027: and should be considered as the true insulator that
1028: separates the 2DES which resides
1029: at the interface between c-$Si$ and a-$Si$. Let the
1030: location of this amorphization edge be at $z=z_a$. This edge can be defined
1031: to within a few atomic planes within the first-principles theoretical models.
1032: If we are dealing with a thick
1033: electron gas, then envelope-function methods for describing the form factor may
1034: be reasonable. Otherwise a more detailed atomic description involving
1035: Bloch functions is needed. In any case,
1036: if the electron gas is very thin, its growth-direction
1037: density profile may be considered
1038: to be $\delta(z-z_a)$. That is, crystalline $Si$ and a-$Si$ flank the
1039: two sides of the 2-D electron layer, with a-$Si$ playing the role of the insulator.
1040: The valence bonds of the a-$Si$ still form a quasi-random tetrahedral network,
1041: even though distorted, and hence the ``background'' dielectric constant of
1042: a-$Si$ is essentially that of $Si$ That is, the effective dielectric constant:
1043: \begin{equation}
1044: {\bar\epsilon}=0.5(\epsilon_{si}+\epsilon_{ins})
1045: \end{equation}
1046: often used for the 2DES in the MOSFET positioned at
1047: an abrupt $Si/SiO_2$ interface
1048: should be reconsidered. The second formula of Ando et al.
1049: (see appendix of ref.\cite{afs}) for the conversion $n$ to $r_s$,
1050: using a mean ${\bar\epsilon}$ of 7.7
1051: is not recommended. Instead, the
1052: first formula of ref.\cite{afs}, i.e.,
1053: \begin{equation}
1054: \label{firstformula}
1055: r_s/a^*=1.751\left[\frac{10^{12}\mbox{cm}^{-2}}{n_s}\right]^{1/2}
1056: \left[\frac{11.5}{\epsilon_{sc}}\right]
1057: \left[\frac{m}{0.19m_0}\right]
1058: \end{equation}
1059: is clearly the one consistent with the first-principles
1060: atomic structure of the $Si/SiO_2$ interface referred to above\cite{pierre}.
1061: If we look at the $Si$-MOSFET literature, we find that the formula
1062: which uses the average dielectric constant of 7.7, appropriate for the abrupt
1063: $Si/SiO_2$ interface has been used by a number of authors. These authors use
1064: a value of $r_s$ increased by a factor of $\sim 1.49$ compared to
1065: what we recommend.
1066: Thus Pudalov et al.\cite{pudalov}, and Okamoto et al.\cite{okamoto} have used the
1067: mean dielectric constant of 7.7 for their calculation of $r_s$. However, both
1068: these studies use the $r_s$ parameter mainly as a plotting variable in the
1069: figures, and not for
1070: any many-body calculations. Hence a choice of $r_s$
1071: which differs from that used in our work by a factor of $\sim$1.49
1072: is immaterial.
1073: In the review article by Kravchenko and Sarachik\cite{krav},
1074: values of $r_s$ are further
1075: modified by the experimentally obtained $m^*$ to
1076: discuss the interactions in $Si$ MOSFETS.
1077: Thus their $r_s^*$ is not used in the sense used in standard many-body theory.
1078: Hawang and Das Sarma\cite{dasrs} have also examined $Si$-MOSFET
1079: resistivities, using
1080: an impurity-scattering calculation which requires defining
1081: the effective background
1082: dielectric constant. They point out that their results are qualitative.
1083: Their results
1084: would not be affected by the choice of either formula given by ref.\cite{afs},
1085: i.e, using ${\bar\epsilon}$ =7.7 or 11.5. Altshuler and Maslov\cite{maslov}
1086: actually consider the implications of the suboxide layer and how this could play a
1087: significant role in the theory. However, they too point out that their effort
1088: is essentially to
1089: indicate a ``mechanism'' rather than a theory of
1090: the metal-insulator physics of $Si$-MOSFETS.
1091: Hence, once again, the results are too qualitative to make any difference.
1092: Similarly,
1093: the results of other workers \cite{punnose} also do not discriminate sufficiently
1094: to make the choice of ${\bar\epsilon}$ a significant issue.
1095:
1096: Another class of problems where the choice of the average dielectric
1097: constant is an issue
1098: is in calculating electric subband energies\cite{afs}. Although the eigenvalues
1099: of the Kohn-Sham
1100: equation are not to be considered as effective excitation energies,
1101: such an assumption
1102: is often made. The input dielectric constant
1103: (which decides the effective $r_s$) enters into
1104: the exchange-correlation functions as well as the Poisson potential
1105: used. Most calculations
1106: are in the high density regime, and it turns out that,
1107: given the uncertainties
1108: of the quantum-well potentials and other parameters,
1109: the results can be equally well explained by
1110: a range of values of the effective dielectric constant.
1111:
1112: This situation becomes quite different when it comes to {\it quantitative}
1113: calculations for {\it low-density}
1114: MOSFETs, e.g., in the regime $n=1\times10^{11}$ electrons/cm$^{2}$. Our CHNC calculations
1115: for $m^*$ and $g^*$ presented in ref.\cite{cmm} clearly favour the first formula,
1116: Eq.~\ref{firstformula} of
1117: Ando et al.\cite{afs} as the correct formula.
1118: This is also the formula that is consistent
1119: with the Car-Parrinello optimized atomic structure
1120: of the $Si/SiO_2$ interface obtained
1121: from the calculations of Carrier et al.\cite{pierre}
1122: and also of Pasquerello et al.\cite{pasq}).
1123: % What is
1124: %very important is that the striking quantitative agreement between the experimental data
1125: %and CHNC theory is sharp enough to insist on this choice as the correct conversion of
1126: %experimental densities (e.g., $n$ quoted in electrons per square centimeter) to an $r_s$
1127: %value.
1128: We believe that the problem of the correct dielectric constant at the
1129: Si/SiO$_2$ interface has received little scrutiny within the
1130: 2-D electron community in the past because there
1131: was no analytic many-body theory capable
1132: of giving quantitative results for low-density electron
1133: systems. Also, it is interesting to note that if the second formula of Ando et al. were
1134: used instead of the first formula that we recommend, then the calculated {\it total}
1135: $r_s$ is close to the $n/2$ value (per valley, $\sim 1.414r_s$) of $r_s$
1136: calculated by the first formula. This
1137: is consistent with the calculations that we advocate, at the Hartee-Fock level (i.e,
1138: at the single-electron level). This fact can also lead to some confusion in assessing
1139: the validity of numerical calculations.
1140: %Experimental measurement of the compressibility of 2DES in Si MOSFETS
1141: %and comparison with theory (Fig.~\ref{muk})
1142: %could provide further quantitative information on the correct average dielectric
1143: %constant to be
1144: %used in $Si$-MOSFET physics.
1145:
1146: The CHNC programs for electron gas calculations mentioned here and in
1147: ref.~\cite{prl2} may be accessed via the
1148: internet at the address given in Ref.~\cite{web}.
1149:
1150: \section{Conclusion}
1151: We can derive the following conclusions from this study. The CHNC method applied
1152: to a four-component electron fluid gives results in very close agreement with
1153: available Diffusion Monte Carlo calculations,
1154: without the use of any adjustable parameters
1155: specific to the 2-valley problem. The coupled-mode approach to constructing the
1156: 2-valley properties from 1-valley data is also fully confirmed. The calculation of the
1157: spin-susceptibility enhancement $\chi_s/\chi_P$ from the
1158: second $\zeta$- derivative of the spin
1159: dependent energy is found to be very sensitive to
1160: the energy difference between
1161: the polarized and unpolarized phases and to the form of the
1162: polarization dependence. Nevertheless, the coupled-mode form is very
1163: successful in capturing the required physics. Thus the QMC, the 2-valley CHNC,
1164: and the coupled-mode approach based on the 1-valley data provide
1165: three independent methods for the study of the strongly
1166: coupled 2DES in Si MOSFETS. The three method are found to
1167: be in excellent agreement. Finally, we note that the methods
1168: used in this paper can be used to study electron/hole bilayers
1169: which are separated by a physical distance $d_L$, the present work being
1170: for $d_L=0$.
1171:
1172: %
1173: \begin{references}
1174: \bibitem[\dag]{byline1} electronic mail address: chandre@babylon.phy.nrc.ca \\
1175: $\,^*$ Fran\c{c}ois Perrot - NRC visiting scientist program.
1176: %
1177: \bibitem{krav}
1178: S. V. Kravchenko and M. P. Sarachik, Rep. Prog.Phys. {\bf 67}, 1 (2004)
1179: %
1180: \bibitem{afs}
1181: See T. Ando, B. Fowler, and F. Stern, Rev. Mod. Phys. {\bf 54}, 437 (1982)
1182: %
1183: \bibitem{cmm}
1184: M. W. C. Dharma-wardana, http://xxx.lanl.gov/cond-mat/0307153
1185: %
1186: \bibitem{sash}
1187: A. A. Shashkin, M. Rashmi, S. Anissimova, and S. V. Kravchenko, V. T. Dolgopolov,
1188: T. M. Klapwijk, Phys. Rev. Lett. {\bf 91}, 46403 (2003)
1189: %axXiv:cond-mat/0303004, (2003)
1190: %
1191: \bibitem{jun}
1192: J. Zhu, H. L. Stormer, L. N. Pfeiffer, K. W. Baldwin, and K. W. West,
1193: Phys. Rev. Lett. {\bf 90},56805 (2003)
1194: %\bibitem{euplfc}
1195: %M. W. C. Dharma-wardana and Fran\c{c}ois Perrot, Europhys. Lett.,
1196: %660, (2003) and arXiv:cond-mat/0304034
1197: %
1198: \bibitem{valpud}
1199: V. M. Pudalov, A. Punnoose, G. Brunthalef, A. Prinz, and G. Bauer,
1200: arXiv:cond-mat/0104347vi
1201: %
1202: \bibitem{prl2}
1203: Fran\c{c}ois Perrot and M. W. C. Dharma-wardana, Phys. Rev. Lett. {\bf 87},
1204: 206404 (2001)
1205: %
1206: \bibitem{hard}
1207: N. Q. Khanh and H. Totsuji, Solid state Communications, {\it 129},37 (2004)
1208: %
1209: \bibitem{bulut}
1210: C. Bulutay and B. Tanatar, Phys. Reb. B {\bf 65}, 195116 (2002)
1211: %
1212: \bibitem{cs}
1213: S. Conti and G. Senatore, Europhys. Lett. 36, 695, (1996)
1214: %
1215: \bibitem{rapi}
1216: F. Rapisarda and G. Senatore, Aust. J. Phys. {\bf 49}, 161 (1996)
1217: %
1218: \bibitem{tc}
1219: B. Tanatar and D. M. Ceperley, Phys. Rev. B {\bf 39}, 5005, (1989)
1220: %
1221: \bibitem{prl1}
1222: M. W. C. Dharma-wardana and F. Perrot, Phys. Rev. Lett. {\bf 84}, 959 (2000)
1223: Fran\c{c}ois Perrot and M. W. C. Dharma-wardana, Phys. Rev. B, {\bf 62}, 14766
1224: (2000)
1225: %%
1226: \bibitem{prl3}
1227: M. W. C. Dharma-wardana and F. Perrot., Phys. Rev. Lett. {\bf 90}, 136601 (2003)
1228: %
1229:
1230: \bibitem{hyd}
1231: M. W. C. Dharma-wardana and F. Perrot, Phys. Rev. B, {\bf 66}, 14110 (2002)
1232: %
1233:
1234: %
1235: \bibitem{santoro}
1236: e.g., G. E. Santoro and G. F. Giulliani, Phys. Rev. B {\bf 37}, 4813 (1988)
1237: Thus their ``antisymmetric'' $G_{-}$ is $G_s-1$ in our notation.
1238: %
1239: \bibitem{iwamoto}
1240: N. Iwamoto, Phys. Rev. A {\bf 30}, 3289 (1984)
1241: %
1242: \bibitem{vashista}
1243: P. Vashishta, P. Bhattacharya, and K. S. Singwi, Phys. Rev. B10, 5108 (1974),
1244: S. Ichimaru, S. Mitake, S. Tanaka, X-Z Yan, Phys. Rev. A {\bf 32},1768 (1985)
1245: %
1246: \bibitem{atta}
1247: C. Attacalite, S. Moroni, P. Gori-Giorgi, and G. B. Bachelet, Phys. Rev.
1248: Lett. {\bf 88}, 256601 (2002)
1249: %
1250: \bibitem{pierre}
1251: P. Carrier,L. Lewis, C. Dharma-wardana,
1252: Phys. Rev. B {\bf 65}, 165339 (2002);
1253: {\bf 64}, 195330 (2001)
1254: %
1255: \bibitem{nacir}
1256: Nacir Tit and M. W. C. Dharma-wardana,
1257: J. App. Phys., {\it l86}, 387 (1999)
1258: %
1259: \bibitem{pasq}
1260: A. Pasquarello, M.S. Hybertsen, and R. Car, Appl.
1261: Surf. Sci. \textbf{104/105}, 317 (1996);
1262: A. Pasquarello, M.S. Hybertsen, and R. Car, Appl. Phys. Lett.
1263: {\bf 68}, 625 (1996).
1264: %
1265: \bibitem{pudalov}
1266: V. M. Pudalov, M. E. Gershenson, H. Kojima,
1267: N. Butch, E. M. Dizhur, G. Brunthaler, A. Prinz,
1268: G. Bauer, Rev. Let., {\bf 88}, 196404 (2002)
1269: %
1270: \bibitem{okamoto}
1271: T. Okamoto, K. Hosoya, S. Kawaji, and A. Yagi,
1272: Phys. Rev. Lett.,{\bf 82} (3875)
1273: %
1274: \bibitem{dasrs}
1275: S. Das Sarma and E. H. Hwang, Phys. Rev. Lett., {\bf 83} 164
1276: %
1277: \bibitem{maslov}
1278: B. L. Altshuler and D. L. Maslov, Phys. Rev. Lett. {\bf 82} 145
1279: \bibitem{punnose}
1280: A. Punnoose and A. M. Finkelstein, Phys. Rev. Lett.,{\bf 88}, 16802
1281: %\bibitem{dultz}
1282: %S. C. Dultz, B. Alavi and H. W. Jiang, axXiv:cond-mat/0210584
1283: %\bibitem{illani}
1284: %S. Ilani, A. Yacoby, D. Mahalu, and H. Shtrikman, Science, {\bf 292}, 1354
1285: %(2001)
1286: %\bibitem{eisenstein}
1287: %J. P. Eisenstein, L. N. Pfeiffer, and K. W. West,
1288: %Phys. Rev. B {\bf 50}, 1760 (1994)
1289: %\bibitem{web}
1290: %http://nrcphy1.phy.nrc.ca/ims/qp/chandre/chnc
1291: %\bibitem{fxc3D84}
1292: %
1293: %\bibitem{rk}
1294: %A.K Rajagopal and J. C. Kimball, Phys. Rev. B {\bf 15}, 2819 (1977)
1295: %\bibitem{stls}
1296: %K. S. Singwi,M. P. Tosi, R. H. Land, and A. Sj\"{o}lander,
1297: % Phys. Rev. {\bf 176} 589 (1968);
1298: %
1299: %
1300: %\bibitem{tosi}
1301: %B. Davoudi, M. Polini, G. F. Giuliani, and M. P. Tosi
1302: %Phys. Rev. B {\bf 64}, 153101 (2001)
1303: %
1304: %\bibitem{marinescu}
1305: %J. Moreno and D. C. Marinescu, cond-mat/0206465
1306: %
1307: %\bibitem{teter}
1308: % G. S. Atwal, I. G. Khalil and N. W. Ashcroft, Phys. Rev. B, {\bf 67} 115107
1309: % (2003);
1310: %I. G. Khalil, M. Teter and N. W. Ashcroft, Phys. Rev. B {\bf 65}, 195309 (2002)
1311: %
1312: %
1313: %\bibitem{moroni}
1314: %S. Moroni, D. M. Ceperley, and G. Senatore, as quoted in Ref. 10 of
1315: %$B, Davoudi et al., Ref.~\cite{tosi}.
1316: %
1317: %
1318: %\bibitem{prb00}
1319: %Fran\c{c}ois Perrot and M. W. C. Dharma-wardana, Phys. Rev. B, {\bf 62}, 14766
1320: %(2000)
1321: %
1322: %\bibitem{pd2d}
1323: %M. W. C. Dharma-wardana and F. Perrot, Phys. Rev. Lett. {\bf 90}, 136601 (2003)
1324: %
1325: \bibitem{web}
1326: http://babylon.phy.nrc.ca/ims/qp/chandre/chnc
1327: %
1328: \end{references}
1329: \end{document}
1330: %
1331:
1332: