1: \documentstyle[prl,aps,multicol,epsf]{revtex}

2: \voffset -0.5cm

3: \begin{document}

4: \draft

5: \title{Spin polarization of strongly interacting 2D electrons: the role of

6: disorder.}

7: \author{S. A. Vitkalov and M.P. Sarachik}

8: \address{Physics Department, City College of the City

9: University of New York, New York, New York 10031}

10: \author{T.~M.~Klapwijk}

11: \address{Delft University of Technology, Department of Applied Physics,

12: 2628 CJ Delft, The Netherlands}

13: \date{\today}

14: \maketitle

15: 

16: \begin{abstract}

17: In high-mobility silicon MOSFET's, the $g^*m^*$ inferred indirectly from

18: magnetoconductance and magnetoresistance measurements with the assumption

19: that $g^*\mu_BH_s=2E_F$ are in

20: surprisingly good agreement with $g^*m^*$ obtained by direct measurement

21: of Shubnikov-de Haas oscillations.  The enhanced

22: susceptibility $\chi^* \propto (g^*m^*)$ exhibits critical behavior

23: of the form $\chi^* \propto (n - n_0)^{-\alpha}$.  We examine the

24: significance of the field scale $H_s$ derived from transport

25: measurements, and show that this field signals the onset of full spin

26: polarization only in the absence of disorder.  Our results suggest that

27: disorder becomes increasingly important as the electron density is

28: reduced toward the transition.  

29: 

30: \end{abstract}

31: 

32: \pacs{PACS numbers: 71.30.+h, 73.40.Qv, 73.50.Jt}

33: 

34: \begin{multicols}{2}

35: 

36: Two dimensional systems of electrons \cite{krav,papadakis,hanein2}

37: and holes \cite{coleridge,hanein,simmons} have been the focus of a great

38: deal of attention during the last few years.  In contrast with

39: expectations for noninteracting \cite{gang} or weakly interacting

40: \cite{altshuler} electrons in two dimensions, these strongly

41: interacting systems exhibit metallic behavior in the absence of a magnetic

42: field: above some characteristic electron (hole) density, $n_c$, their

43: resistivities decrease with decreasing temperature.  Whether there is a

44: genuine metallic phase and a true metal-insulator transition in these

45: materials continues to be the subject of lively debate \cite {AKS}.

46: 

47: Experimental results have been obtained in the 2D system of

48: electrons in silicon MOSFET's that indicate that the response to a

49: magnetic field applied in the plane of the electrons increases

50: dramatically as the electron density is decreased toward $n_c$.  Based on

51: a study of the scaled magnetoconductance as a function of temperature and

52: electron density, Vitkalov {\it et al.}

53: \cite{vitkalov_ferro} have identified an energy scale $\Delta$ that

54: decreases with decreasing density and extrapolates to zero in the limit $T

55: \rightarrow 0$ at a density $n_0$ in the vicinity of $n_c$; this was

56: interpreted as evidence of a quantum phase transition at $n_0$.  

57: From studies at very low temperatures of the magnetoresistance as a

58: function of electron density, Shashkin {\it et al.} \cite{Shashkin}

59: inferred that the two-dimensional system of electrons in silicon

60: inversion layers approaches a ferromagnetic instability at the critical

61: density $n_c$ for the zero-field metal-insulator transition.  From a

62: determination of the enhanced spin susceptibility derived from

63: Shubnikov-de Haas measurements down to low densities, Pudalov

64: {\it et al.} \cite{{pudalov_SdH}} have claimed there is no spontaneous

65: spin polarization for electron densities above $n = 8.34 \times 10^{10}$

66: cm$^{-2} \approx n_c$, although they could not exclude this for lower

67: densities.  The possibility that a magnetically ordered phase exists in

68: the limit $T \rightarrow 0$ in dilute two-dimensional silicon inversion

69: layers is intriguing and bears further investigation.

70: 

71: In this paper we show that there is very good agreement

72: between values reported for $g^*m^*$ as a function of electron density in

73: high-mobility silicon MOSFET's obtained directly from measurements of the

74: Shubnikov-de Haas oscillations \cite{{pudalov_SdH}} and those inferred

75: indirectly from magnetoconductance and magnetoresistance measurements by

76: two different groups using different methods of analysis and the

77: assumption that

78: $g^*\mu_BH_s=2E_F$ \cite{vitkalov_ferro,Shashkin,comparison}.  Here $g^*$

79: is the enhanced $g$-factor, $m^*$ is the enhanced electron mass,

80: $\mu_B$ is Boltzmann's factor, $E_F$ is the Fermi energy, and $H_s$ is a

81: characteristic field scale determined by different methods from in-plane

82: magnetoconductance \cite{vitkalov_ferro} and magnetoresistance \cite

83: {Shashkin} experiments.  The enhanced susceptibility $\chi^* \propto

84: (g^*m^*)$ exhibits critical behavior of the form $\chi^* \propto (n -

85: n_0)^{-\alpha}$.  Data from the three experimental groups yield exponents

86: $\alpha$ of $0.23, 0.24$ and $0.27$, and critical densities between

87: $0.88$ and $1.04 \times 10^{11}$ cm$^{-2}$.  We examine the significance

88: of the field scale $H_s$, and show that this field signals the onset

89: of full spin polarization only in the absence of disorder.  Our results

90: suggest that disorder becomes increasingly important as the electron

91: density is reduced toward the transition.  

92: 

93: Measurements of Shubnikov-de Haas oscillations in high-mobility silicon

94: MOSFET's with high electron densities have shown that the magnetic field

95: required to achieve complete polarization of the electron spins is

96: approximately the same as that required to saturate the magnetoresistance

97: to a constant value \cite{okamoto,pol1,pol2}.  For the relatively high

98: densities used in these experiments, the field $H_\rho$ correponding to

99: saturation of the magnetoresistance is approximately the same as the

100: field $H_\sigma$ above which there is apparent saturation of the

101: magnetoconductance.  As we show below, this equivalence  breaks down at

102: lower densities.  A clear example is illustrated in Fig. \ref{rhosigma},

103: where the resistivity and conductivity are shown as a function of

104: in-plane magnetic field for a silicon MOSFET with electron density near

105: the critical density, $n_c$, for the metal-insulator transition.  The

106: saturation field $H_\rho$ derived from the resistivity is considerably

107: larger than the field $H_\sigma$ above which the conductivity saturates. 

108: This can be understood with reference to the band diagrams shown as

109: insets to Fig. \ref{rhosigma}.  In the absence of disorder, \vbox{

110: \vspace{0.1 in}

111: \hbox{

112: \hspace{-0.4in} 

113: \epsfxsize 3.6 in \epsfbox{rhosigma.eps} 

114: }

115: }

116: \vskip -0.7cm

117: \refstepcounter{figure}

118: \parbox[b]{2.9in}{\baselineskip=12pt FIG.~\thefigure.

119: For a silicon MOSFET with electron density $0.94 \times 10^{11}$

120: cm$^{-2}$, the conductivity (left curve) and resistivity (right curve)

121: are shown as a function of in-plane magnetic field at temperature

122: $T=0.26$ K; the saturation fields deduced from the resistivity and the

123: conductivity are labeled $H_\rho$ and $H_\sigma$, respectively.  The

124: insets show schematic diagrams of the electron bands (see

125: text for discussion).

126: 

127: \vspace{0.3in}

128: }

129: \label{rhosigma}

130: 

131: \noindent all electron

132: states are extended, band-tailing plays a negligible role, and full spin

133: polarization is achieved when the Zeeman energy is sufficient to

134: completely depopulate the minority spin band: 

135: \begin{equation}

136: g^* \mu_B H_{band} = 2 E_F;

137: \label{band}

138: \end{equation}

139: here $g^*$ is the enhanced $g$-factor, $\mu_B$ is Boltzmann's

140: factor, $H_{band}$ is the magnetic field required to fully polarize the

141: system in the absence of disorder, and $E_F$ is the Fermi energy. 

142: Disorder is weak at high electron densities and one expects $H_{band}

143: \approx H_\rho

144: \approx H_\sigma$.  

145: 

146: As the density is decreased and disorder and the band-tails

147: become more important, complete spin alignment requires the

148: application of a larger magnetic field to fully polarize the tail states

149: as well as the extended states:

150: \begin{equation}

151: g^* \mu_B H_{tail+band} = 2 E_F + \delta

152: \label{tail}

153: \end{equation}

154: where we've assumed the band tail has an effective energy

155: width $\delta$ \cite{strongdisorder}.  

156: 

157: Except very near the transition, the number of states in the

158: band tails in the case of samples of reasonably high mobility is much

159: smaller than the number of extended states; at the same time, the energy

160: width $\delta$ becomes appreciable as the density decreases and the

161: disorder increases.  The field required to align the electrons in the

162: higher mobility band states can thus differ substantially from the

163: magnetic field needed to polarize

164: $all$ the electrons

165: \cite{pudalov2}.  While the (small number) of  tail states make a minor

166: contribution to the conductivity, the resistivity is considerably more

167: sensitive to the low-mobility states in the tail of the distribution, and

168: consequently

169: $H_\rho > H_\sigma$ as is evident in Fig.\ref{rhosigma}.  We suggest that

170: $H_\sigma \approx H_{band}$ and

171: $H_\rho \approx H_{tail+band}$.

172: 

173: \vbox{

174: \vspace{0.3 in}

175: \hbox{

176: \hspace{-0.1in} 

177: \epsfxsize 3.4 in \epsfbox{difference.eps} 

178: }

179: }

180: \vskip 0.5cm

181: \refstepcounter{figure}

182: \parbox[b]{3.3in}{\baselineskip=12pt FIG.~\thefigure.

183: The fractional difference $(H_\rho - H_\sigma)/H_\sigma$ (open symbols)

184: and $2/\sigma$ (closed symbols) versus electron density; $H_\rho$ and

185: $H_\sigma$ are the saturation fields deduced from resistivity and

186: conductivity curves, respectively.  The open circles and closed symbols

187: refer to data taken at $0.26$ K.  The open squares are data obtained at

188: $0.1$ K on a different MOSFET. 

189: 

190: \vspace{0.10in}

191: }

192: \label{difference} 

193: 

194: 

195: The fractional difference between $H_\rho$ and $H_\sigma$, $\Delta H/H=

196: (H_\rho - H_\sigma)/H_\sigma$, is shown as a function of electron density

197: in Fig. \ref{difference}; $\Delta H/H$ increases rapidly with decreasing

198: electron density when disorder becomes more dominant.  The quantity

199: $2/\sigma$ is plotted for comparison through the following argument.  For

200: weak scattering, the parameter $\delta$ is on the order of the scattering

201: rate: $\delta \sim \hbar/\tau$. With Eqs. \ref{band} and \ref{tail},

202: this gives $\Delta H/H =\delta/2E_F=\hbar/2(E_F \tau)$.   Using the

203: expression for the Drude conductivity $\sigma = ne^2\tau/m^*$, and the

204: Fermi energy $E_F/\hbar = (nh)/g_vg_sm^*$ with a valley degeneracy $g_v=2$

205: and spin degeneracy

206: $g_s=2$, one obtains $\Delta H/H=(e^2/h)(2/\sigma)$.  The correlation

207: between 

208: $\Delta H/H$ and $2/\sigma$ is evident in Fig. \ref{difference}.

209: 

210: In an earlier paper \cite{vitkalov_ferro}, we showed that the

211: magnetoconductance of silicon MOSFET's can be scaled onto a single

212: curve by plotting $[\sigma(H) - \sigma(0)]/[\sigma(H=\infty) -

213: \sigma(0)]$ as a function of $H/H_s$.  The parameter $H_s$ obtained by

214: this method is proportional to $H_\sigma$ discussed above.  For high 

215: densities where disorder plays a small role, the magnetic field

216: $H_\sigma$ needed to saturate the conductivity is very nearly equal to

217: the field required to obtain full spin polarization.  At lower densities,

218: the saturation fields deduced from the resistivity and the conductivity

219: are not the same, and we have argued that the difference is associated

220: with the effect of electrons in the states in the band tails. 

221: We've suggested that $H_\sigma$ is the magnetic field required to polarize

222: the band states; the Zeeman energy and $g^*m^*$ are then given by Eq.

223: \ref{band} with $H_{band} = H_\sigma$.  The tail states remain unpolarized

224: in $H=H_\sigma$.  However, except perhaps very near the transition (or in

225: samples of very low mobility), they represent a small fraction of the

226: electrons, so that the system is close to full spin polarization.

227: 

228: 

229: 

230: \vbox{

231: \vspace{0.4 in}

232: \hbox{

233: \hspace{-0.2in} 

234: \epsfxsize 3.4 in \epsfbox{critical.eps} 

235: }

236: }

237: \vskip 0.5cm

238: \refstepcounter{figure}

239: \parbox[b]{3.1in}{\baselineskip=12pt FIG.~\thefigure.

240: The inverse of the enhanced susceptibility $\chi_0/\chi^*$ versus electron

241: density obtained by Vitkalov {\it et al.} \cite{vitkalov_ferro}, Shashkin

242: {\it et al.}

243: \cite{Shashkin}, and Pudalov {\it et al.} \cite{pudalov_SdH}.  Data are

244: normalized to the Shubnikov-de Haas values at high densities.  The curve

245: is a fit to the critical form

246: $\chi_0/\chi^* = A(n-n_0)^\alpha$ for the data of ref.

247: \cite{vitkalov_ferro} (excluding the point shown at $\chi_0/\chi^* = 0$).

248: 

249: \vspace{0.10in}

250: }

251: \label{critical} 

252: 

253: 

254: Fig. \ref{critical} shows $2m_0/m^*g^* = \chi_0/\chi^*$ as a function of

255: electron density $n_s$ obtained from our data \cite{vitkalov_ferro},

256: by Sashkin {\it et al.} \cite{Shashkin}, and Pudalov {\it et al.}

257: \cite{pudalov_SdH}.  Here

258: $\chi^*/\chi_0$ is the enhanced susceptibility normalized to its free

259: electron value, and

260: $\chi_0/\chi^*$ is its inverse.  The closed circles denote values obtained

261: from scaling our data for the in-plane magnetoconductance and the

262: assumption that $g^*

263: \mu_B H_\sigma = 2 E_F$; the open circles were obtained by Shashkin {\it

264: et al.}

265: \cite{Shashkin}from magnetoresistance measurements using a different

266: data-fitting procedure and the same assumption as above; the squares

267: are from direct Shubnikov-de Haas measurements of Pudalov {\it et

268: al.}\cite{pudalov_SdH}.  The data of Shashkin {\it et al.} decrease

269: somewhat more rapidly at low densities than the others.  However, the

270: three sets obtained by different groups using different measurements and

271: different methods of analysis agree surprisingly well.  Again, this

272: indicates that the small number of states in the band tails in

273: high-mobility MOSFET's play a neglibible role.  A fit to the critical form

274: \begin{equation}

275: \chi_0/\chi^* \propto (n-n_0)^\alpha,

276: \label{crit}

277: \end{equation}

278: yields the following values for the three data sets considered: for the

279: Shubnikov-de Haas data of Pudalov {\it et al.} \cite{pudalov_SdH}

280: $\alpha = 0.23$, $n_0=0.96 \times 10^{11}$ cm$^{-2}$; for the

281: magnetoconductance data of Shashkin {\it et al.} \cite{Shashkin}

282: $\alpha = 0.27$,

283: $n_0=1.04 \times 10^{11}$ cm$^{-2}$; and for our data

284: \cite{vitkalov_ferro} $\alpha = 0.24$, $n_0=0.88 \times

285: 10^{11}$ cm$^{-2}$.

286: 

287: We have argued above that for high-mobility samples, the difference

288: $(H_\rho - H_\sigma)$ is associated with the effect of a small fraction

289: of the electrons  in the band tails.  The characteristic field $H_s$

290: obtained in our earlier work was determined from scaling the

291: magnetoconductance, which is a measure of the field required to align the

292: band states while leaving a few electrons in the tail states

293: unpolarized.  Shashkin {\it et al.} determined a field scale by matching

294: magnetoresistance data at $low$ magnetic fields; close examination shows

295: that this procedure does not produce a match at high fields (note that

296: their data is shown on a logarithmic scale, which deemphasizes

297: differences between the curves at high values of magnetic field).  Both

298: methods are sensitive to the contribution of the extended state and

299: minimize the effect of the states in the band tails.  These procedures

300: yield reliable measures for the behavior of the system at high

301: electron densities where disorder does not play an important role.  This

302: accounts for the surprisingly good agreement between the $g^*m^*$

303: obtained from transport experiments and those found by direct measurement

304: of the Shubnikov-de Haas oscillations.  At densities very near the

305: transition (and for very low mobility MOSFET's) one should expect this

306: correspondence to break down as disorder becomes more dominant.  We

307: suggest that an understanding of any phase transition that occurs in this

308: regime must incorporate the effect of disorder in a central way.

309: 

310: We thank M. Gershenson and S. V. Kravchenko for providing data

311: for Fig. \ref{critical}.  This work was supported by the US Department of

312: Energy under Grant No.~DE-FG02-84ER45153.

313: 

314: \begin{references}

315: 

316: \bibitem{krav} S.\ V.\ Kravchenko, G.\ V.\ Kravchenko, J.\ E.\

317: Furneaux, V.\ M.\ Pudalov, \and M.\ D'Iorio, Phys.\ Rev.\ B

318: {\bf 50}, 8039 (1994); S.\ V.\ Kravchenko, W.\ E.\ Mason, G.\ E.\

319: Bowker, J.\ E.\ Furneaux, V.\ M.\ Pudalov, \and M.\ D'Iorio, 

320: Phys.\ Rev.\ B {\bf 51}, 7038 (1995); S. V. Kravchenko, D.

321: Simonian, M. P. Sarachik, Whitney Mason, \and J. E. Furneaux,

322: Phys.\ Rev.\ Lett. {\bf 77}, 4938 (1996).

323: 

324: \bibitem{papadakis} S. J. Papadakis and M. Shayegan, Phys. Rev.

325: B {\bf 57}, R15068 (1998).

326: 

327: \bibitem{hanein2} Y. Hanein, D. Shahar, J. Yoon, C. C. Li, D. C. Tsui,

328: \and H. Shtrikman, Phys. Rev. B {\bf 58}, R13338 (1998).

329: 

330: \bibitem{coleridge} P.~T.~Coleridge, R.~L.~Williams, Y.~Feng, \and

331: P.~Zawadzki, Phys.\ Rev.\ B {\bf 56}, R12764 (1997).

332: 

333: \bibitem{hanein} Y.~Hanein, U. Meirav, D. Shahar, C. C. Li, D. C. Tsui, 

334: \and H. Shtrikman, Phys.\ Rev.\ Lett. {\bf 80} 1288 (1998).

335: 

336: \bibitem{simmons} M.~Y.~Simmons, A. R. Hamilton, M. Pepper, E. H.

337: Linfield, P. D. Rose, D. A. Ritchie, A. K. Savchenko, \and T. G.

338: Griffiths, Phys.\ Rev.\ Lett. {\bf 80} 1292 (1998).

339: 

340: \bibitem{gang} E. Abrahams, P. W. Anderson, D. C. Licciardello, and

341: T. V. Ramakrishnan. Phys. Rev. Lett. {\bf 42} 673 (1979).

342: 

343: \bibitem{altshuler} B. L. Altshuler, A. G. Aronov, and P. A. Lee,

344: Phys. Rev. Lett. {\bf 44}, 1288 (1980).

345: 

346: \bibitem{AKS} For a review see E. Abrahams, S. V. Kravchenko, and M. P.

347: Sarachik, Rev. Mod. Phys.{\bf 73}, 251 (2001).

348: 

349: \bibitem{vitkalov_ferro} S. A. Vitkalov, H. Zheng, K. M. Mertes, M. P.

350: Sarachik and T. M. Klapwijk, Phys. Rev. Lett. {\bf 87}, 086401, (2001).

351: 

352: \bibitem{Shashkin} A. A. Shashkin, S. V. Kravchenko, V. T. Dolgopolov,

353: and T. M. Klapwijk, Phys. Rev. Lett. {\bf 87}, 086801, (2001).

354: 

355: \bibitem{pudalov_SdH}  V. M. Pudalov, M. Gershenson, H. Kojima, preprint

356: cond-mat/0110160 (2001).

357: 

358: \bibitem{comparison}  A similar comparison was presented in a comment by

359: S. V. Kravchenko, preprint cond-mat/0106056 (2001).

360: 

361: \bibitem{strongdisorder}  For low disorder, one expects that the density

362: of states is much smaller in the band tail than for the extended states,

363: and small variations of $E_F$ can be neglected in Eq. \ref{tail}.

364: 

365: \bibitem{pudalov2}  The possible role of

366: localized or bound states was considered by V. M. Pudalov, G. Brunthaler,

367: A. Prinz, and G. Bauer, to be published in Phys. Rev. Lett. (2002);

368: preprint cond-mat/0004206 (2000).

369: 

370: \bibitem{okamoto} T. Okamoto, K. Hosoya, S. Kawaji, and A. Yagi, Phys.

371: Rev.Lett. {\bf 82}, 3875 (1999).

372: 

373: \bibitem{pol1} S. A. Vitkalov, H. Zheng, K. M. Mertes, M. P. Sarachik

374: and T. M. Klapwijk, Phys. Rev. Lett. {\bf 85}, 2164 (2000).

375: 

376: \bibitem{pol2} S. A. Vitkalov, M. P. Sarachik, and T. M.

377: Klapwijk,  Phys. Rev. {\bf B} {\bf 64}, 073101, (2001).

378: 

379: \end{references}

380: 

381: \end{multicols}

382: 

383: \end{document}

384: 

385: 

386: