1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %
3: % title: Kinetic modelling and molecular dynamics simulation of
4: % ultracold neutral plasmas including ionic correlations
5: % authors: Pohl, Pattard, Rost
6: % APS code:
7: %
8: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
9: % MPI for the Physics of Complex Systems
10: % Noethnitzer Str. 38
11: % D-01187 Dresden, Germany
12: % phone: +49-351-871-2209
13: % FAX: +49-351-871-2299
14: % email: tpohl@mpipks-dresden.mpg.de
15: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
16: %
17: % 0.99, 13.5.04 TPo
18: %
19: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
20:
21: %\documentclass[aps,pra,twocolumn,groupedaddress,showpacs]{revtex4}
22: \documentclass[aps,pra,preprint,groupedaddress,showpacs]{revtex4}
23: \usepackage{graphicx}
24: \usepackage{psfig}
25: \usepackage{epsf}
26: \usepackage{epsfig}
27: \usepackage{amsmath}
28: \begin{document}
29:
30: \title{Kinetic modelling and molecular dynamics simulation of ultracold neutral
31: plasmas including ionic correlations}
32:
33: \author{T.\ Pohl}
34: \author{T.\ Pattard}
35: \author{J.M.\ Rost}
36:
37: \affiliation{Max Planck Institute for the Physics of Complex Systems,
38: N{\"o}thnitzer Str.\ 38, D-01187 Dresden, Germany}
39:
40: \date{\today}
41:
42: \begin{abstract}
43: A kinetic approach for the evolution of ultracold neutral plasmas including
44: interionic correlations and the treatment of ionization/excitation and
45: recombination/deexcitation by rate equations is described in detail.
46: To assess the reliability of the approximations inherent in the kinetic model,
47: we have developed a hybrid molecular dynamics method. Comparison of the results
48: reveals that the kinetic model describes the atomic and ionic observables of
49: the ultracold plasma surprisingly well, confirming
50: our earlier findings concerning the role of ion-ion correlations [Phys.\ Rev.\
51: A {\bf 68}, 010703]. In addition, the molecular dynamics approach allows one to
52: study the relaxation of the ionic plasma component towards thermodynamical
53: equilibrium.
54: \end{abstract}
55:
56: \pacs{52.20.-j, 32.80.Pj, 52.25.Dg, 52.65.Ww}
57:
58: \maketitle
59:
60: \section{Introduction} \label{intro}
61: Recent experiments have produced ultracold neutral plasmas
62: from a small cloud of laser-cooled atoms confined in a magneto-optical trap
63: \cite{Kil99,Kul00,Kil01,Rob00,Eyl00,Gou03}. In one type of experiments
64: \cite{Kil99,Kul00,Kil01}, a plasma was produced by photoionizing
65: laser-cooled Xe atoms with an initial ion temperature of about $10 \,
66: \mu {\rm K}$.
67: By tuning the frequency of the ionizing laser, the initial electron energy
68: $E_{\rm e}$ could be varied corresponding to a temperature range $1 \, {\rm K}
69: < E_{\rm e} / k_{\rm B} < 1000 \, {\rm K}$, and the subsequent expansion of the
70: plasma into the surrounding vacuum was studied systematically. In a
71: complementary type of experiment \cite{Rob00,Eyl00,Gou03}, ultracold Rb and Cs
72: atoms were laser-excited into high Rydberg states rather than directly ionized.
73: In these experiments, also the spontaneous evolution of the Rydberg gas into a
74: plasma has been observed.
75: The time evolution of several quantities characterizing the
76: state of the plasma, such as the plasma density \cite{Kil99,Kul00}, the degree
77: of ionization \cite{Rob00,Eyl00,Gou03} or the energy-resolved atomic level
78: population \cite{Kil01} have been measured using various plasma diagnostic
79: methods.
80:
81: These experiments, which have paved the way towards an unexplored regime of
82: ionized gases, give rise to new phenomena in atomic physics as well as in
83: plasma physics. Hence, a number of different theoretical approaches have
84: been formulated to cover different aspects of these experiments
85: \cite{Kuz02,Kuz02b,Maz02,Rob02,Rob03,Tka01,PPR03}.
86:
87: An important issue is the question whether the plasma produced would be
88: strongly coupled or not.
89: The correlation strength is determined by the Coulomb coupling
90: parameter $\Gamma=e^{2}/(a k_{{\mathrm{B}}} T)$ with the Wigner-Seitz radius
91: $a$ \cite{Dub99}. A plasma is called ``strongly coupled'' if $\Gamma \gg 1$,
92: i.e.\ if the Coulomb interaction between the plasma particles largely exceeds
93: the thermal kinetic energy. In this case, interesting ordering effects such as
94: Coulomb crystallization can be observed \cite{Ich82,PPR04}.
95: For the initial conditions of the NIST experiments
96: \cite{Kil99,Kul00,Kil01}, however, the development of equilibrium
97: electron-electron
98: correlations leads to a rapid heating of the electron gas, which prevents the
99: electron Coulomb coupling parameter $\Gamma_{\rm e}$ from exceeding unity
100: \cite{Kuz02}. The same has been argued for an ion plasma in \cite{Mur01}.
101: Since the electron dynamics proceeds on a much smaller timescale than the ion
102: motion, in \cite{Kuz02,Kuz02b} the electron heating could be studied for the
103: early stage of the plasma evolution only, where the ionic component does not
104: show dynamical effects. On the other hand, ion heating has only been studied in
105: the framework of a model system, consisting of a homogeneous gas of
106: Debye-screened ions \cite{Mur01}, such that the influence of the subsequent
107: expansion could not be explored.
108:
109: The first
110: quantitative comparison with experimental observations has been given in
111: \cite{Rob02}, with the plasma dynamics modeled within a
112: hydrodynamical approach and ionization, excitation and recombination
113: treated by a separate set of rate equations.
114: Since this model does only account for
115: the mean-field potential created by the charges, it cannot describe effects of
116: particle correlations. However, it has also been shown there that the
117: electronic Coulomb coupling parameter $\Gamma_{\rm e}$
118: does not exceed a value of $\approx 0.2$ during the plasma expansion
119: due to heating by three-body recombination. Thus,
120: the influence of electron-electron correlations on the dynamics of the plasma
121: could be neglected. On the other hand, three-body recombination does not
122: influence the ionic temperature, so that the ions can heat up only through
123: correlation heating (and energy exchange with the electrons, which, however,
124: is very slow). Since the ionic temperature was set to zero in \cite{Rob02},
125: the role of ion-ion correlations could not be explored. In a
126: preliminary study \cite{PPR03}, we showed that they indeed change the
127: evolution of the system quantitatively, though not qualitatively. In the
128: following, we will give a detailed account of the kinetic model used in
129: \cite{PPR03} and of all relevant
130: ingredients. We will also develop a hybrid molecular dynamics (H-MD) approach
131: which treats the electronic plasma component in an adiabatic approximation while
132: the ions are fully accounted for. Such an approach permits the description of
133: situations where the ions are strongly coupled \cite{PPR04,PPR04b}, which is
134: clearly beyond the capabilities of the simple kinetic model. Nevertheless, for
135: the typical situations realized in the experiments \cite{Kil99,Kul00,Kil01},
136: comparison of the two theoretical approaches yields very good agreement,
137: corroborating our findings reported earlier \cite{PPR03} and establishing
138: firmly that one can
139: capture the relevant physics with the relatively simple kinetic approach.
140: \section{Theoretical Approach}
141: Our kinetic approach is similar to the one of \cite{Rob02}.
142: The main difference is the inclusion of ion-ion correlations (IC) which will be
143: described in detail below. Briefly, a set of kinetic equations is formulated for
144: the evolution of the plasma (subsection \ref{za}), while ionization/excitation
145: and recombination/deexcitation are taken into account on the basis of rate
146: equations (subsection \ref{zb}). In order to test the applicability and
147: accuracy of this model, we have developed a less approximative and more
148: flexible but computationally much more demanding approach. It uses
149: molecular dynamics for the ionic motion while the electron component is
150: treated as a fluid assuming a quasi-steady state (subsection \ref{zc}).
151:
152: \subsection{Kinetic description}
153: \label{za}
154: Starting from the first equation of the BBGKY hierarchy, the
155: evolution equation for the one-particle distribution function $f_\alpha
156: (\mathbf{r}, \mathbf{v},t)$ of the free plasma charges is obtained as
157: \begin{equation} \label{kin}
158: m_{\alpha}\left(\partial_t + \mathbf{v}\partial_{\mathbf{r}}\right)
159: f_\alpha (\mathbf{r}, \mathbf{v},t) =
160: \sum_{\beta}\int{\left[\partial_{\mathbf{r}}\varphi_{\alpha\beta}(\mathbf{r},
161: \mathbf{r}^{\prime})\right]
162: \partial_{\mathbf{v}}f_{\alpha\beta}(\mathbf{r},\mathbf{v},\mathbf{r}^{\prime},
163: \mathbf{v}^{\prime},t)\:d\mathbf{r}^{\prime}d\mathbf{v}^{\prime}} \; ,
164: \end{equation}
165: where $\alpha, \beta$ label the particle species (e,i for electrons and ions,
166: respectively), $f_{\alpha\beta}(\mathbf{r},
167: \mathbf{v},\mathbf{r}^{\prime}, \mathbf{v}^{\prime},t)$ denotes the
168: two-particle distribution function for the corresponding particle species and
169: $\varphi_{\alpha\beta}=q_{\alpha}q_{\beta}/|{\bf{r}}-{\bf{r}}^{\prime}|$ is the
170: Coulomb interaction potential between the charges $q_{\alpha}$ and $q_{\beta}$.
171: Electron-electron correlations are very small during the plasma expansion,
172: since the electrons will quickly heat up due to three-body recombination and the
173: additional heating due to correlation effects is small in comparison
174: \cite{Rob02}. Hence,
175: we neglect electron-electron as well as electron-ion correlations, leaving
176: only IC as a possible influence on the plasma dynamics
177: beyond the mean-field level. On this level of approximation the ion kinetic
178: equation can be written as
179: \begin{equation} \label{ionkin}
180: m_{\rm{i}}\left(\partial_t + \mathbf{v}\partial_{\mathbf{r}}-\frac{\partial_{
181: \mathbf{r}}\bar{\varphi}}{m_{\rm i}}\partial_{\mathbf{v}}\right)
182: f_{\rm{i}} = \\
183: \int{\left(\partial_{\mathbf{r}}\varphi_{\rm{ii}}\right)\:
184: \partial_{\mathbf{v}}w_{\rm{ii}}(\mathbf{r},\mathbf{v},\mathbf{r}^{\prime},
185: \mathbf{v}^{\prime})\:d\mathbf{r}^{\prime}d\mathbf{v}^{\prime}}\;,
186: \end{equation}
187: where the function
188: \begin{equation} \label{twodist}
189: w_{{\rm{ii}}}(\mathbf{r},\mathbf{v},\mathbf{r}^{\prime},\mathbf{v}^{\prime})=
190: f_{{\rm{ii}}}(\mathbf{r}, \mathbf{v},\mathbf{r}^{\prime},
191: \mathbf{v}^{\prime})-f_{{\rm{i}}}(\mathbf{r},
192: \mathbf{v})f_{{\rm{i}}}(\mathbf{r}^{\prime}, \mathbf{v}^{\prime})
193: \end{equation}
194: contains the contributions of IC to the two-particle
195: distribution function and $\bar{\varphi}$ is the mean-field potential created
196: by all plasma charges. Since $m_{\rm e}/m_{\rm i} \ll 1$, the
197: relaxation timescale of the electrons is much smaller than the timescale of
198: the plasma expansion under typical experimental conditions \cite{Kil99}. Thus,
199: we may safely apply an adiabatic approximation for the electron distribution
200: function, assuming a local Maxwellian distribution
201: \begin{equation} \label{adi}
202: f_e(\mathbf{r},\mathbf{v}) \propto \exp \left(\frac{\bar{\varphi}(
203: \mathbf{r})}{k_{\rm B} T_{\rm e}} \right) \;
204: \exp \left(- \frac{\mathbf{v}^2}{2 k_{\rm B} T_{\rm e}} \right) \;\;,
205: \end{equation}
206: where $T_{\rm{e}}$ is the electron temperature.
207: Eq.\ (\ref{adi}) together with a quasineutral approximation \cite{Dor98} allows
208: one to express the mean-field potential in terms of the ionic density
209: $\rho_{\rm{i}}=\int{f_{\rm{i}}d{\mathbf{v}}}$, resulting in
210: \begin{equation} \label{pot}
211: \partial_{\mathbf{r}} \bar{\varphi}=k_{\rm{B}}T_{\rm{e}} \frac{
212: \partial_{\mathbf{r}} \rho_{\rm{i}}}{\rho_{\rm{i}}} \; .
213: \end{equation}
214: Using Eq.\ (\ref{pot}), the following evolution equations
215: for the second moments of the ion distribution function are derived from
216: Eq.\ (\ref{ionkin})
217: \begin{subequations} \label{mom}
218: \begin{eqnarray} \label{moma}
219: \partial_t\left<r^2\right>&=&2\left<{\mathbf{r}}{\mathbf{v}}\right>
220: \\
221: \label{momb}
222: \frac{m_{\rm{i}}}{2}\partial_t\left<{\mathbf{r}}{\mathbf{v}}\right>&=&
223: \frac{m_{\rm{i}}}{2}\left<v^2\right>+\frac{3}{2}k_{\rm{B}}T_{\rm{e}}
224: +\frac{1}{2}N_{\rm{i}}^{-1}\int{\rho_{\rm{i}}({\bf{r}}){\mathbf{r}}{
225: \mathbf{F}}_{\rm{ii}}({\mathbf{r}})\:d{\mathbf{r}}} \\
226: \label{momc}
227: \frac{m_{\rm{i}}}{2}\partial_t\left<v^2\right>&=&N_{\rm{i}}^{-1}
228: \left(k_{\rm{B}}T_{\rm{e}}
229: \int{\rho_{\rm{i}}\partial_{\mathrm{r}}{\mathbf{u}}\:d{\bf{r}}}
230: -\int{{\mathbf{v}}(\partial_{\mathbf{r}}\varphi_{\rm{ii}}})
231: w_{\rm{ii}}({\mathbf{r}},{\mathbf{v}},{\mathbf{r}}^{\prime},{
232: \mathbf{v}}^{\prime})\:d{\mathbf{r}}d{\mathbf{v}}d{\mathbf{r}}^{\prime}
233: d{\mathbf{v}}^{\prime}\right)
234: \end{eqnarray}
235: \end{subequations}
236: where $\left<r^2\right> = N_{\rm{i}}^{-1}\int r^2 f_{\rm i}({\mathbf{r}},
237: {\mathbf{v}}) d{\mathbf{r}} d{\mathbf{v}}$ etc.
238: The ``correlation force'' ${\mathbf{F}}_{\rm{ii}}({\mathbf{r}})$ is given by
239: \begin{equation} \label{cforce}
240: {\mathbf{F}}_{\rm{ii}}\left({\mathbf{r}}\right)=-\int{\left(\partial_{\mathbf{r
241: }}\varphi_{\rm{ii}}\right)\rho_{\rm{i}}({\bf{r}}^{\prime})g_{\rm{ii}}\left({
242: \mathbf{r}},{\mathbf{r}}^{\prime}\right)}\:d{\mathbf{r}}^{\prime} \;,
243: \end{equation}
244: where the spatial correlation function $g_{\rm{ii}}$ is defined by
245: $\rho_{\rm i}({\mathbf{r}}) \rho_{\rm i}({\mathbf{r}}^{\prime})
246: g_{\rm ii}({\mathbf{r}},{\mathbf{r}}^{\prime}) \equiv \int w_{\rm{ii}}\left({
247: \mathbf{r}},{\mathbf{v}},{\mathbf{r}}^{\prime},{\mathbf{v}}^{\prime}\right)
248: \:d{\mathbf{v}}d{\mathbf{v}}^{\prime} $
249: and ${\mathbf{u}}({\mathbf{r}})=\int{{\mathbf{v}}f_{\rm{i}}}d{\mathbf{v}}$
250: is the hydrodynamical drift velocity of the plasma.
251: With the help of the second
252: kinetic ion equation of the BBGKY hierarchy, the last term on the right-hand
253: side of Eq.\ (\ref{momc}) can be written as
254: \begin{eqnarray} \label{cen}
255: \frac{1}{N_{\rm{i}}}\int{{\mathbf{v}}(\partial_{\mathbf{r}}\varphi_{\rm{ii}})
256: w_{\rm{ii}}\:d{
257: \mathbf{r}}d{\mathbf{v}}d{\mathbf{r}}^{\prime}d{\mathbf{v}}^{\prime}}&=&-
258: \frac{1}{2N_{\rm{i}}}\partial_{t}\int{\varphi_{\rm{ii}}w_{\rm{ii}}\:d{
259: \mathbf{r}}d{\mathbf{v}}d{\mathbf{r}}^{\prime}d{\mathbf{v}}^{\prime}}
260: \nonumber \\
261: &=&-\partial_{t}U_{\rm{ii}} \; ,
262: \end{eqnarray}
263: where
264: \begin{equation} \label{Uc2}
265: U_{\rm{ii}} = \frac{1}{2 N_{\rm{i}}}\int \varphi_{\rm{ii}} \rho_{\rm{i}}({
266: \mathbf{r}})\rho_{\rm{i}}({\mathbf{r}}^{\prime})g_{\rm{ii}}({\mathbf{r}},{
267: \mathbf{r}}^{\prime}) \:d{\mathbf{r}}d{\mathbf{r}}^{\prime}
268: \end{equation}
269: is the average correlation energy per ion. Hence, Eq.\ (\ref{momc})
270: reflects energy conservation for the ion subsystem.
271: The evolution of the hydrodynamical velocity ${\mathbf{u}}$ is determined by
272: \begin{equation} \label{hyda}
273: m_{\rm{i}}\rho_{\rm{i}}\left[\partial_t{\mathbf{u}}+\left({\mathbf{u}}\cdot
274: \partial_{\mathbf{r}}\right){\mathbf{u}}\right]=-k_{\rm{B}}T_{\rm{e}}
275: \partial_{{\mathbf{r}}}\rho_{\rm{i}}-\partial_{\mathbf{r}}P_{\rm{th,i}}
276: -\rho_{\rm{i}}\mathbf{F_{\rm ii}}
277: \end{equation}
278: where $P_{\rm{th,i}}=\frac{m_{\rm{i}}}{3
279: N_{\rm{i}}}\int{({\mathbf{v}}-{\mathbf{u}})^2f_{\rm{i}}\:d{\mathbf{r}}d{
280: \mathbf{v}}}$ is the thermal ion pressure.
281: As shown in the appendix, in the framework of a local density approximation,
282: i.e.\ by assuming that $g_{\rm ii}$ only depends
283: on the distance $|\mathbf{r}-\mathbf{r}^{\prime}|$ and on the
284: densities at the two coordinates $\mathbf{r}$ and
285: $\mathbf{r}^{\prime}$, and that the ionic density $\rho_i$ varies slowly on the
286: lengthscale where $g$ is significantly different from zero,
287: the total correlation energy can be approximated by the well-known LDA
288: expression
289: \begin{equation} \label{Uc}
290: U_{\rm ii}=N_{\rm{i}}^{-1}\int{u_{\rm{ii}}\rho_{\rm{i}}}\:d{\mathbf{r}}
291: \end{equation}
292: while the correlation force is found to be
293: \begin{equation}
294: \label{Fc}
295: \mathbf{F_{\rm{ii}}}
296: = -\frac{1}{3}\left(\frac{u_{\rm ii}}{\rho_{\rm i}}+\frac{\partial u_{\rm ii}}{
297: \partial \rho_{\rm i}}\right) \partial_{\mathbf{r}} \rho_{\rm{i}} \;\; ,
298: \end{equation}
299: where $u_{\rm ii}(\mathbf{r})$ is the correlation energy of a homogeneous plasma
300: of density $\rho_{\rm i}(\mathbf{r})$,
301: \begin{equation}
302: \label{uc}
303: u_{\rm ii}(\mathbf{r})=\frac{e^2}{2}\rho_{\rm i} (\mathbf{r}) \int
304: \frac{g_{\rm ii}(x; \rho_{\rm{i}})}{x} \, d{\mathbf{x}} \;\; .
305: \end{equation}
306: If IC are neglected in Eq.\ (\ref{ionkin}), the kinetic equation exhibits the
307: following selfsimilar solution
308: \begin{eqnarray} \label{sol}
309: f_{\rm{i}}&\propto&\exp{\left(-\frac{r^2}{2\sigma^2}\right)}\exp{\left(-
310: \frac{m_{\rm{i}}\left({\bf{v}}-{\bf{u}}\right)^2}{2k_{\rm{B}}
311: T_{\rm{i}}}\right)} \nonumber \\
312: {\mathbf{u}}&=&\gamma{\mathbf{r}}
313: \end{eqnarray}
314: which corresponds to the initial state of the experiments under consideration.
315: As soon as IC are taken into account via the correlation
316: pressure $\rho_{\rm i} \bf{F}_{\rm{ii}}$ in Eq.\ (\ref{hyda}), however,
317: Eqs.\ (\ref{sol}) are no longer exact solutions of Eq.\ (\ref{ionkin}).
318: Using Eq.\ (\ref{Fc}), the last term on the right-hand side of Eq.\ (\ref{hyda})
319: can be rewritten as $\rho_{\rm{i}}{\bf{F}}_{\rm{ii}}=-\frac{1}{3}\frac{\partial
320: (u_{\rm{ii}}\rho_{\rm{i}})}{\partial \rho_{\rm{i}}}\partial_{\mathbf{r}}
321: \rho_{\rm i}$.
322: Interpreting this term as a local nonideal pressure, an equation for the
323: parameter $\gamma$ was derived in \cite{PPR03} by averaging
324: the differential equation for $|{\bf{u}}|/r$ obtained by inserting Eq.\
325: (\ref{sol}) into Eq.\ (\ref{hyda})
326: over the plasma volume. Obviously, this treatment is not unique.
327: Since, as discussed above, the ansatz (\ref{sol}) does not solve Eq.\
328: (\ref{hyda}) exactly, multiplying Eq.\ (\ref{hyda}) by different
329: functions of $r$ and averaging over the plasma volume will lead to slightly
330: different evolution equations for the parameter $\gamma$.
331: Here, supported by a comparison with our numerical results from the MD
332: simulations, to be discussed below, we assume as in \cite{PPR03} that the
333: functional form of the hydrodynamical
334: quantities of Eqs.\ (\ref{sol}) is not altered by the inclusion of IC, while
335: the dynamics of the parameters appearing in Eqs.\ (\ref{sol}) is determined from
336: the equations (\ref{mom}) for the moments of the distribution function.
337: Clearly, such an approximation can not be {\sl a priori} justified. Hence, it
338: must be validated {\sl a posteriori} by comparison with more sophisticated
339: methods which do not rely on a reduction of the plasma description to a few
340: macroscopic parameters.
341:
342: With this procedure, we arrive at the following set of equations
343: for the width $\sigma$ of the plasma cloud, its expansion velocity $\gamma
344: \mathbf{r}$ as well as ionic and electronic temperature $T_{\rm i}$ and
345: $T_{\rm e}$:
346: \begin{subequations}\label{par}
347: \begin{eqnarray}
348: \label{para}
349: \partial_{t}\sigma^2&=&2\gamma\sigma^2 \\
350: \label{parb}
351: \partial_t\gamma&=&\frac{k_{\rm{B}}T_{\rm{e}}+k_{\rm{B}}T_{\rm{i}}+\frac{1}{3}
352: U_{\rm{ii}}}{m_{\rm{i}}\sigma^2}-\gamma^2
353: \\
354: \label{parc}
355: \partial_t k_{\rm{B}}T_{\rm{i}}&=&-2\gamma
356: k_{\rm{B}}T_{\rm{i}}-\frac{2}{3}\gamma U_{\rm{ii}}-\frac{2}{3}\partial_t
357: U_{\rm{ii}} \\
358: \label{pard}
359: \partial_t k_{\rm{B}}T_{\rm{e}}&=&-2\gamma k_{\rm{B}}T_{\rm{e}} \; .
360: \end{eqnarray}
361: \end{subequations}
362: The last equation (\ref{pard}) has been derived from the electron kinetic
363: equation by making use of the quasineutrality condition.
364: The set of Eqs.\ (\ref{par}) slightly differs from that presented in
365: \cite{PPR03} where $W_{\rm{c}}=\frac{1}{3N_{\rm{i}}}\int \rho_{\rm{i}}\frac{
366: \partial
367: (u_{\rm{ii}}\rho_{\rm{i}})}{\partial \rho_{\rm{i}}}\:d{\bf{r}}$ had been
368: used instead of $U_{\rm{ii}}$ in Eq.\ (\ref{parb}). A
369: comparison with our numerical MD results shows that Eqs.\ (\ref{par}) yield a
370: slightly better quantitative agreement, while the principal influence of
371: IC on the plasma dynamics, which has been partly discussed in
372: \cite{PPR03}, is the same.
373: Eqs.\ (\ref{par}) provide a transparent physical
374: picture of the expansion dynamics. First, Eq.\ (\ref{pard}) together with
375: (\ref{para}) reflects the
376: adiabatic cooling of the electron gas, i.e.\ $T_{\rm{e}}\sigma^2={\rm{const}}$.
377: The ion temperature, on the other hand, is not only affected by the adiabatic
378: cooling, expressed by the first term in Eq.\ (\ref{parc}), but also changes due
379: to the development of IC, which is taken into account by the last term in
380: Eq.\ (\ref{parc}). Furthermore, these correlations reduce the ion-ion
381: interaction and therefore lead to an effective negative acceleration,
382: expressed by the $U_{\rm{ii}}/3$-term in Eq.\ (\ref{parb}), in
383: addition to the ideal thermal pressure.
384: This contribution, which corresponds to the average nonideal pressure
385: known from homogeneous systems \cite{Ich82,Dub99}, also leads to an effective
386: potential in which the ions
387: move. As they expand in this potential, the thermal energy changes due to
388: energy conservation, as expressed by the second term on the right-hand side of
389: Eq.\ (\ref{parc}). Finally, combining eqs.(\ref{par}) yields a second integral
390: of motion, namely the total energy of the plasma
391: \begin{equation}
392: E_{\rm{tot}}=\frac{3}{2}\left(k_{\rm{B}}T_{\rm{e}}+k_{\rm{B}}T_{\rm{i}}\right)
393: +\frac{3}{2}m_{\rm{i}}\gamma^2\sigma^2+U_{\rm{ii}}\;.
394: \end{equation}
395: Although the set of equations (\ref{par}) determines the time evolution of all
396: relevant macroscopic plasma parameters, namely its width, expansion velocity,
397: electron and ion temperature, it is not a closed set since an evolution
398: equation for the correlation energy $U_{\rm{ii}}$ which enters Eq.\
399: (\ref{parc}) is missing. Initially, the plasma is completely uncorrelated, so
400: that $U_{\rm ii} (t=0) = 0$. However, the initial state corresponds to a
401: non-equilibrium
402: situation, and the plasma will relax towards thermodynamic equilibrium,
403: thereby building up correlations. A precise description of this
404: relaxation process in the framework of a kinetic theory is rather complicated
405: and requires a considerable numerical effort \cite{Sem99}. We therefore employ a
406: linear approximation for the relaxation of the two-particle correlation
407: function, the so-called correlation-time approximation \cite{Bon96a},
408: \begin{equation} \label{CTA}
409: \frac{dw_{\rm ii}({\bf{r}},{\bf{v}},{\bf{r}}^{\prime},{\bf{v}}^{\prime};t)}{dt}
410: \approx-\frac{w_{\rm ii}({\bf{r}},{\bf{v}},{\bf{r}}^{\prime},{\bf{v}}^{\prime};
411: t)-w_{\rm ii}^{\rm{eq}}({\bf{r}},{\bf{r}}^{\prime};t)}{\tau_{{\rm{corr}}}} \; .
412: \end{equation}
413: Here, $\tau_{{\rm{corr}}}$ \cite{Bon96a, Bon96b} is the characteristic
414: timescale for the relaxation
415: of particle correlations and $w_{{\rm{ii}}}^{{\rm{eq}}}$ is the equilibrium
416: pair correlation function, which in our case still depends on time via the
417: evolving one-particle distribution function since the plasma is freely
418: expanding. As shown in \cite{Mor98} the correlation time $\tau_{\rm{corr}}$ can
419: be well estimated by the inverse ionic plasma frequency. Hence, in our
420: calculations we set $\tau_{\rm{corr}}=\omega_{\rm{p,i}}^{-1}=\sqrt{m_{\rm{i}}/
421: (4\pi e^2\bar{\rho}_{\rm{i}})}$, where
422: $\bar{\rho}_{\rm{i}}=N_{\rm{i}}/(4\pi\sigma^2)^{3/2}$ is the average ionic
423: density of the plasma.
424: Such a linear approximation is good only for small
425: deviations of $w_{{\rm{ii}}}$ from its equilibrium form. Clearly, this is not
426: the case in the initial stage of the gas evolution. However, after the initial
427: phase of correlation heating the system stays very close to its slowly
428: changing local equilibrium, and one may expect Eq.~(\ref{CTA}) to yield good
429: results. Under the same conditions that lead to Eqs.\ (\ref{Uc}) and (\ref{Fc}),
430: one easily verifies that Eq.\ (\ref{CTA}) leads to
431: \begin{equation} \label{UCTA}
432: \partial_t U_{\rm{ii}}\approx-\frac{U_{\rm{ii}}-U_{\rm{ii}}^{\rm{eq}}}{\tau_{{
433: \rm{corr}}}},
434: \end{equation}
435: where $U_{\rm{ii}}^{\rm{eq}}=N_{\rm{i}}^{-1}\int{\rho_{\rm{i}}u_{\rm{ii}}^{
436: \rm{eq}} d{\mathbf{r}}}$
437: and $u_{\rm{ii}}^{\rm{eq}}({\mathbf{r}})$ is the correlation energy
438: per particle of a homogeneous one-component plasma in local equilibrium.
439: This quantity has been studied intensively in the past, and approximate
440: analytical formulae are available in the literature
441: \cite{Ich82,Dub99}. Here, we adopt the interpolation formula from \cite{Cha98}
442: \begin{equation} \label{eos}
443: u_{\rm{ii}}^{\rm{eq}}({\mathbf{r}})=k_{\rm B} T_{\rm i} \Gamma^{3/2}\left(
444: \frac{A_1}{\sqrt{A_2+ \Gamma}}+ \frac{A_3}{1+\Gamma}\right) \;,
445: \end{equation}
446: with $A_1=-0.9052$, $A_2=0.6322$ and $A_3=-\sqrt{3}/2-A_1/\sqrt{A_2}$,
447: which yields an accurate interpolation between the low-$\Gamma$ Abe limit and
448: the high-$\Gamma$ behavior obtained by Monte Carlo and MD simulations.
449: It should be noted that in the present situation $u_{\rm ii}$ depends on time
450: since the plasma expands. Hence, $\Gamma$ and with it the thermodynamical
451: equilibrium change in time.
452:
453: The set of equations (\ref{par}) describes the evolution of the plasma
454: part of the system, i.e.\ a system of $N_{\rm i}$ ions and electrons.
455: Due to ionization and recombination events occurring during the plasma
456: expansion (discussed in detail in the following subsection), this number
457: $N_{\rm i}$, and hence also the total mass $M = N_{\rm i} m_{\rm atom}$, is not
458: constant
459: over the course of the evolution. However, such a treatment completely
460: neglects the influence of the bound Rydberg atoms on the dynamics. One may
461: argue that they do not influence the plasma evolution since they do not
462: interact with the ions or electrons by Coulomb interaction. On the other hand,
463: a Rydberg atom may carry a significant amount of kinetic energy, gained from the
464: acceleration by the electron pressure before its formation by three-body
465: recombination. In a simple approximation, we
466: assume equal hydrodynamical velocities and density profiles for the
467: ions and atoms, in order to account for this effect. This implies that the
468: expansion of the neutral Rydberg atoms can be taken into account by replacing
469: the mass
470: $N_{\rm{i}}m_{\rm{i}}$ of the ions by the mass of the {\em total} system
471: $(N_{\rm{i}}+N_{\rm{a}})m_{\rm{i}}$, where $N_{\rm{a}}$ is the number of
472: atoms. We therefore replace the ion mass $m_{\rm{i}}$ by an effective mass
473: $(1+N_{\rm{a}}/N_{\rm{i}})m_{\rm{i}}$ in Eq.\ (\ref{parb}).
474: The quality of this approximation can, of course, also be checked by
475: comparison with the H-MD description, see below.
476: \subsection{Ionization and Recombination}
477: \label{zb}
478: As demonstrated in \cite{Rob02}, a satisfactory description of
479: the dynamics of an ultracold plasma can be achieved by combining a
480: hydrodynamic treatment of the plasma evolution with rate
481: equations accounting for inelastic collisions between the plasma particles and
482: Rydberg atoms.
483: The rate equation for the change of density
484: of Rydberg atoms in a state with principal quantum number $n$ reads
485: \begin{equation}
486: \label{rate}
487: \dot{\rho}_{\rm a}(n)=\rho_{\rm e}\sum_{p}\left[K(p,n)\rho_{\rm a}(p)- K(n,p)
488: \rho_{\rm a}(n) \right]+\rho_{\rm e}\left[R(n)\rho_{\rm e}\rho_{\rm i}-I(n) \rho_{\rm a}(n)\right]
489: \;,
490: \end{equation}
491: where \(K(p,n)\) is the rate coefficient for electron impact (de)excitation
492: from level $p$ to level $n$, and $R(n)$ and $I(n)$ describe
493: three-body recombination into and electron-impact ionization from level $n$,
494: respectively.
495: The rate coefficients $K$, $R$ and $I$ have been taken from the classic
496: work of Mansbach and Keck \cite{Man69}. Additional processes, such as, e.g.,
497: ionization by black-body radiation or from dipolar atom-atom interactions,
498: are easily included in Eq.\ (\ref{rate}) if the corresponding rates are
499: available. Such processes are essential for a description of the early stages
500: of the evolution of a system starting with a Rydberg gas
501: \cite{Rob00,Eyl00,Gou03}, but are of minor importance in situations starting
502: from a pure plasma.
503:
504: In this framework, the evolution of the system is obtained by solving
505: Eqs.\ (\ref{para})-(\ref{parc}) together with Eq.\ (\ref{rate}) while the
506: electron temperature is now obtained from the modified energy conservation
507: relation $E_{\rm{tot}}+E_{\rm{a}}={\rm{const.}}$ instead of Eq.\ (\ref{pard}),
508: where
509: $E_{\rm{a}}=-{\cal{R}}\sum_{n}{N_{\rm{a}}n^{-2}}$ is the total energy of the
510: Rydberg atoms and ${\cal{R}}=13.6\:{\rm{eV}}$.
511: \subsection{Hybrid molecular dynamics treatment}
512: \label{zc}
513: As we will show in section III, the kinetic description of the previous
514: subsections is able to describe the plasma dynamics to a surprisingly large
515: extent. However, one of the main motivations of this work is the study of the
516: role of IC, which are incorporated in the model only in an
517: approximate way. To assess their influence on the dynamics reliably,
518: a more sophisticated approach is required, e.g.\ molecular dynamics simulations
519: which fully incorporate the ionic interactions. However, a full MD simulation of
520: both, electrons and ions, is computationally very demanding, and only the very
521: early stage of the system evolution can be described in this way \cite{Kuz02}.
522: On the other hand, as argued above, electronic correlations are not
523: important for the plasma dynamics, so that only IC have to
524: be accounted for in full while the influence of the electrons on the dynamics
525: may be treated on a mean-field level. Moreover, we have seen that the
526: timescale of equilibration of the electronic subsystem is orders of magnitude
527: shorter than that of the ionic subsystem and the timescale of the plasma
528: expansion. This observation led us to use an adiabatic approximation in
529: subsection \ref{za}, where
530: the electrons are assumed to equilibrate instantaneously, assuming a Maxwellian
531: velocity distribution with a well-defined temperature and a spatial profile
532: determined from the total mean-field potential of the
533: plasma charges. The clear separation of timescales suggests that this
534: adiabatic approximation is well justified, hence we will keep it in the
535: following. Consequently, we have developed a hybrid approach where the electrons
536: are treated on a hydrodynamical level as in the kinetic description above,
537: while the ions are propagated individually with their mutual
538: interaction and the influence of the electrons
539: on the ions enters via the electronic mean-field potential. This hybrid approach
540: permits the use of much larger timesteps in the propagation of the system,
541: since the electronic dynamics needs not to be followed in detail but only the
542: ionic motion
543: has to be resolved in time. Consequently, the evolution of the system can be
544: followed over the experimental timescales.
545: Furthermore, the approximate treatment of IC in the kinetic
546: model of subsection \ref{za} can be tested.
547: Finally, beyond the scope of the present work, we have shown \cite{PPR04} that
548: the present H-MD approach can describe situations where the ionic plasma
549: component is so strongly coupled that crystallization of the ions sets in. Such
550: a scenario is clearly beyond the capabilities of a kinetic approach.
551:
552: As discussed above, the
553: electrons are still treated as a fluid, while we lift the
554: quasineutral approximation by calculating the resulting mean-field potential
555: from the Poisson equation
556: \begin{equation} \label{poisson}
557: \Delta \bar{\varphi}=4\pi e^2\left(\rho_{\rm{e}}-\rho_{\rm{i}}\right).
558: \end{equation}
559: However, using Eq.\ (\ref{adi}) poses a conceptual difficulty \cite{Rob03}
560: since the
561: mean-field potential approaches a finite value at large distances and therefore
562: leads to a non-normalizable electron density. This problem, which has been
563: discussed for a long time in an astrophysical context
564: \cite{Cha43}, reflects the fact that a substantial fraction of the
565: electrons indeed escapes the finite potential barrier at long times during the
566: relaxation process until the total kinetic energy of all electrons is less
567: than the height of the potential well. On the timescales under
568: consideration, however, typically only a small amount of the electrons escapes the
569: plasma volume, until the resulting charge imbalance becomes large enough to trap the
570: remaining electrons, which quickly reach a quasi-steady state forming a
571: temporarily quasineutral plasma in the central region. We account for this
572: electron loss by determining the fraction of trapped electrons from the
573: results of Ref.~\cite{Kil99}.
574:
575: The corresponding steady-state distribution, derived for the
576: study of globular clusters, is of the form \cite{Kin66}
577: \begin{equation} \label{king1}
578: \rho_{\rm{e}}\propto \exp{\left(\frac{\bar{\varphi}}{k_{\rm{B}}T_{\rm{e}}}
579: \right)}\int_{0}^{\chi}\exp{\left(-x\right)}x^{3/2}\:dx,
580: \end{equation}
581: where $\chi=m_{\rm{e}}v_{\rm{esc}}^2/(2k_{\rm{B}}T_{\rm{e}})$ with the velocity
582: $v_{\rm{esc}}(\bf{r})$ necessary to escape from a given position
583: in the plasma. In the present case, the potential can have a non-monotonous
584: radial space dependence and the escape velocity has to be defined as
585: \begin{equation} \label{king2}
586: \frac{m_{\rm{e}}}{2}v_{\rm{esc}}^2\left(r\right)=\max_{r^{\prime}\geq r}{\left[
587: \bar{\varphi}\left(r^{\prime}\right)-\bar{\varphi}\left(r\right)\right]},
588: \end{equation}
589: in contrast to astrophysical problems where one only has a
590: single sign of ``charge'' and $m_{\rm{e}}v_{\rm{esc}}^2/2=-\bar{\varphi}$
591: \cite{Kin66}. For a given electron temperature $T_{\rm{e}}$ and ion density
592: $\rho_{\rm i}$ the electron density is found by numerical iteration of Eqs.\
593: (\ref{poisson}), (\ref{king1}) and (\ref{king2}) until selfconsistency is
594: reached.
595:
596: Knowledge of the electron density then permits
597: propagation of the ions in the electron mean-field
598: $\Delta\bar{\varphi}_{\rm{e}}=4\pi e^2\rho_{\rm{e}}$ and the full interaction
599: potential of the remaining ions,
600: \begin{equation} \label{eom}
601: m_{\rm{i}}\ddot{\bf{r}}_{j}=\partial_{\bf{r}_{j}}\bar{\varphi}_{\rm{e}}+e^2
602: \sum_{k}\frac{{\bf{r}}_{j}-{\bf{r}}_{k}}{\left|{\bf{r}}_{j}-{\bf{r}}_{k}
603: \right|^3} \;.
604: \end{equation}
605: The numerical solution of the ion equations of motion represents the most time
606: consuming part of the plasma propagation. In general, for $N$ propagated
607: particles, the corresponding numerical effort scales with $N^{2}$
608: rendering a treatment of large particle numbers difficult. In order
609: to simulate particle numbers relevant to the experiments, we have adapted a
610: hierarchical treecode originally designed for astrophysical problems,
611: first described in \cite{Bar86}. This method provides a numerically
612: exact solution of the ion equations of motion Eq.\ (\ref{eom}), while the
613: numerical effort grows only as $N\ln N$ with increasing $N$.
614: More details about the numerical procedure can be found, e.g., in
615: \cite{Bar90}.
616:
617: In the framework of the kinetic model introduced in section \ref{za}, the
618: influence of IC on the system evolution can be singled out by
619: comparison with the solution of the corresponding equations with
620: $U_{\rm{ii}}\equiv 0$. In order to make an analogous comparison also for the MD
621: simulations, we have performed calculations propagating the ions in the
622: mean-field potential created by all charges. Technically, the mean-field
623: potential is represented using a test-particle method, widely used for various
624: problems in plasma physics (see, e.g., \cite{Bir95}).
625: \section{Results and Discussion}
626: We will discuss the evolution of a plasma initially consisting of $N_{\rm e} =
627: 37500$ electrons and
628: $N_{\rm i} = 40000$ ions with an average density of 10$^9$ cm$^{-3}$
629: at a rather low electronic kinetic energy $E_{\rm e} = 3 k_{\rm B} T_{\rm e}
630: /2 = 20$ K, comparing the results from the kinetic model and our MD simulation.
631: Thereby, we put special emphasis on the role of IC.
632:
633: \subsection{Global aspects of plasma expansion and recombination}
634: \label{GlobalResults}
635:
636: \begin{figure}[bt]
637: \centerline{\psfig{figure=fig1,width=8cm}}
638: \caption{\label{fig1}
639: Electronic temperature $T_{\rm e}(t)$ for an expanding plasma of $40000$ ions
640: with an initial average density of $10^9$cm$^{-3}$ and an initial electron
641: kinetic energy of $20\:$K, obtained from the H-MD simulation (a) and the
642: kinetic model (b), with (solid) and without (dotted) the inclusion of
643: IC. The inset shows the ratio of the electron temperatures obtained from the
644: H-MD simulation and the kinetic model.}
645: \end{figure}
646:
647: The general macroscopic behavior of the system has been described before in
648: several publications, experimentally as well as theoretically
649: \cite{Kul00,Kil01, Rob02,Rob03}. The
650: plasma cloud slowly expands due to the thermal pressure of the electrons,
651: leading to adiabatic cooling of the electrons as well as partial recombination
652: into bound states (figures \ref{fig1} and \ref{fig2}).
653: The amount of recombination and its influence strongly depends on the initial
654: electron temperature and density. If the electrons are too hot (about $50\:$ K
655: for typical experimental densities of $10^9\:$ cm$^{-3}$), recombination is
656: strongly suppressed and the system dynamics is well described by the results
657: of \cite{Dor98} obtained for the collisionless plasma expansion \cite{Kul00}.
658:
659:
660: \subsubsection{Temporal evolution of the electronic temperature}
661: For the lower electron temperatures considered here, as can be seen in Fig.\
662: \ref{fig1}, there is an initial increase of the
663: electron temperature due to electron heating by three-body recombination and
664: subsequent deexcitation of the formed Rydberg atoms. At low initial electron
665: energies this heating drastically increases the electron temperature and thus
666: accelerates the plasma expansion \cite{Rob02}, which
667: explains the enhanced expansion velocity observed in \cite{Kul00}. In contrast
668: to this recombination heating of the electrons, the inclusion of IC
669: only slightly changes the expansion dynamics, as seen in Fig.\
670: \ref{fig1} by comparing the solid and dotted lines. As shown in the inset of
671: Fig.\ \ref{fig1}a, the electron temperature obtained from the H-MD
672: simulation and the kinetic model differ by at most $8$\% during the first few
673: microseconds of the plasma expansion, while the agreement becomes even better at
674: later times. Moreover, the faster decrease of the electron temperature due to
675: the inclusion of IC, predicted by the particle simulations, is quantitatively
676: reproduced by the much simpler kinetic model.
677:
678: Hence, the simple evolution equations (\ref{par}) are sufficient to clarify
679: the role of IC in the expansion
680: dynamics. According to Eq.\ (\ref{parc}), the development of IC
681: quickly heats up the plasma ions to roughly $-\frac{2}{3}U_{\rm{ii}}$
682: since the expansion of the plasma is still negligible during this initial stage.
683: Thereby, the negative correlation
684: energy term $\frac 13U_{\rm ii}$ in Eq.\ (\ref{parb}) is overcompensated,
685: leading to a faster expansion of the plasma. As a consequence of the quicker
686: expansion, the electron temperature decreases somewhat faster than without the
687: inclusion of IC. With Eq.\ (\ref{parb}), the importance of this effect can be
688: estimated by comparing the thermal electron energy $k_{\rm B} T_{\rm e}$ to
689: the net ion contribution $-\frac{1}{3}U_{\rm{ii}}$ in the numerator
690: of the first term on the right-hand side of Eq.\ (\ref{parb}). Estimating the
691: correlation energy by $e^2/a$, it follows that the
692: total pressure driving the plasma expansion is enhanced by a factor of roughly
693: $1+\Gamma_{\rm e}/3$, which only slightly changes the expansion dynamics since
694: the electrons are known to be weakly coupled over the whole observation time
695: \cite{Rob02}.
696:
697: \subsubsection{Formation of Rydberg atoms in time}
698: The number of recombined atoms is influenced more strongly by IC
699: (Fig.~\ref{fig2}). During the evolution of
700: the system, Rydberg atoms are constantly formed by three-body recombination
701: and re-ionized by the free electrons in the plasma. As shown in Fig.\
702: \ref{fig2}a, for the current set of parameters about $7000$ Rydberg atoms are
703: present in the system after $40\:\mu$s, while the kinetic model yields about
704: $7900$ atoms at the same instant of time (Fig.\ \ref{fig2}b).
705: \begin{figure}[bt]
706: \centerline{\psfig{figure=fig2,width=8cm}}
707: \caption{\label{fig2}
708: Number $N_a(t)$ of recombined atoms obtained from the H-MD simulation (a)
709: and the kinetic model (b) for the same parameters as in Fig.\
710: \ref{fig1}. The solid line shows the result taking into account the
711: IC while the dotted line is obtained from the mean-field treatment.}
712: \end{figure}
713: This number is small compared to the size of the whole system, nevertheless it
714: is large enough that the recombined atoms can be detected in an experiment, and
715: corresponding curves have indeed been obtained experimentally \cite{Kil01}. Due
716: to the strong temperature dependence of the total three-body recombination rate,
717: which is proportional to $T_{\rm e}^{-9/2}$ \cite{Man69}, the slight decrease of
718: the electron temperature due to the faster expansion, caused by the correlation
719: heating of the ions,
720: considerably affects the recombination behavior of the plasma. While there is
721: an overall shift between the atom number obtained from the particle simulations
722: and the kinetic model, both the
723: kinetic model and the H-MD simulation yield an increase of the atom
724: number of about $10\%$ at $t=40\:\mu$s (Fig.\ \ref{fig2}), compared to a
725: mean-field treatment of the ion dynamics. Thus, the
726: H-MD simulation corroborates our previous findings \cite{PPR03}.
727:
728: \begin{figure}[t]
729: \centerline{\psfig{figure=fig3a,width=8cm}}
730: \centerline{\psfig{figure=fig3b,width=8cm}}
731: \centerline{\psfig{figure=fig3c,width=8cm}}
732: \caption{\label{fig3}
733: Population of bound Rydberg states with principal quantum number $n$, after
734: $t=1.5\:\mu$s (a), $t=6\:\mu$s (b) and $t=40\:\mu$s (c). The vertical bars
735: represent the
736: H-MD calculation, the solid line the kinetic model. The dashed curve
737: in (c) shows the kinetic model neglecting IC. Initial-state parameters are
738: the same as in Fig.\ \ref{fig1}.}
739: \end{figure}
740: Additional insight into the recombination process can be gained from a closer
741: look at the distribution of bound Rydberg states. Figure \ref{fig3} shows
742: the population of levels with principal quantum number $n$ for three different
743: times, corresponding to different stages of the plasma expansion. Initially,
744: Rydberg states of moderate excitation are populated, due to a relatively high
745: electron temperature (Fig.\ \ref{fig3}a). At later times, higher excited bound
746: states are formed in the course of the plasma expansion (Figs.~\ref{fig3}b and
747: \ref{fig3}c), since the maximum principal quantum number for recombination
748: $n_{\rm{max}}=\sqrt{{\cal{R}}/(2k_{\rm{B}}T_{\rm{e}})}$ \cite{Man69} increases
749: as the electron temperature drops down.
750: Moreover, the deeply bound states formed at earlier times are also not
751: subject to electron-impact excitation and deexcitation anymore since the
752: thermal velocity of the impacting electrons has become too small. Thus, as
753: becomes apparent by comparing Fig.\ \ref{fig3}b with \ref{fig3}c, the
754: deeply bound states ($n \alt 30$) remain basically untouched, while higher and
755: higher states ``freeze out'' as the plasma expands. As may be anticipated from
756: Fig.~\ref{fig2}, IC mainly affect the later stages of the plasma
757: evolution. Hence, the inclusion of IC alters the
758: population of these higher lying states, as shown in Fig.\ \ref{fig3}c. Since
759: these states have small binding energy, they contribute little to the total
760: kinetic energy of the plasma subsystem. This is the reason why
761: the effect of IC is visible in the distribution of Rydberg states,
762: but not in the macroscopic expansion dynamics of the plasma, reflected, e.g., by
763: the asymptotic expansion velocity measured in \cite{Kul00}.
764: \subsection{Spatially resolved plasma expansion and relaxation}
765: While the time evolution of global, i.e.\ space-averaged, observables of the
766: plasma is very well described by the kinetic model, one may expect
767: discrepancies compared to the MD simulations when looking into the spatially
768: resolved plasma dynamics. We will assess these discrepancies quantitatively in
769: the following.
770: \begin{figure}[bt]
771: \centerline{\psfig{figure=fig4a,width=8cm}}
772: \centerline{\psfig{figure=fig4b,width=8cm}}
773: \caption{\label{fig4}
774: Spatial densities $\rho_{\rm i}$ (solid) and $\rho_{\rm a}$ (dashed)
775: of the ions and atoms, respectively, at $t=3\:\mu$s (a) and
776: $t=31.3\:\mu$s (b), compared to the Gaussian profile assumed for the kinetic
777: model (dotted). Additionally, $\rho_{\rm i}$ obtained from the particle
778: simulation using the mean-field interaction only is shown as the dot-dashed
779: line in (a). Initial-state parameters are the same as in Fig.\ \ref{fig1}.}
780: \end{figure}
781:
782: \begin{figure}[tb]
783: \centerline{\psfig{figure=fig5a,width=8cm}}
784: \centerline{\psfig{figure=fig5b,width=8cm}}
785: \caption{\label{fig5}
786: Hydrodynamic velocity $u(r)$ of ions (full circles) and atoms (open circles)
787: at the same times as in Fig.\ \ref{fig4}, compared to the straight-line
788: assumption of the kinetic model. The dashed line in (a) shows the result of the
789: particle simulation using a mean-field treatment of the ion-ion interaction.}
790: \end{figure}
791: \subsubsection{Evolution of the particle densities}
792: In the derivation of the kinetic equations (\ref{par}), we have assumed that
793: the analytical form of the ionic density $\rho_{\rm i}$ remains invariant
794: during the evolution of
795: the system and, moreover, that the atoms will have the same distribution.
796: As the plasma expands, the spatial profile of the ions must deviate from its
797: original Gaussian shape \cite{Rob03}. This is mainly due to deviations from
798: quasineutrality, e.g.\ deviations from the linear space dependence of the
799: outward directed acceleration, at the plasma edge. The influence
800: of the nonlinear correlation pressure on the density profile
801: is of minor importance, as can be seen by comparing the solid and dot-dashed
802: line of Fig.\ \ref{fig4}a in the inner plasma region. As known from earlier
803: studies of expanding plasmas, based on a mean-field treatment of the particle
804: interactions \cite{Gur66,Sac85,Rob03}, a sharp spike develops at the plasma
805: edge, shown by the dot-dashed line in Fig.\ \ref{fig4}a.
806: At later times, this spike decays again when the maximum of the hydrodynamic
807: ion velocity passes the position of the density peak, so that the region of the
808: peak is depleted. Ultimately, at long times, the plasma approaches a
809: quasineutral selfsimilar expansion \cite{Sac85}.
810: From Fig.~\ref{fig4}a it becomes apparent that with IC the peak structure is
811: less pronounced than in mean-field approximation. This is due
812: to dissipation caused by ion-ion collisions which are fully taken into account
813: in the H-MD simulation. As shown in \cite{Sac85}, by adding an ion viscosity
814: term to the hydrodynamic equations of motion, dissipation tends to
815: stabilize the ion density and prevents the occurrence of wavebreaking which was
816: found to be responsible for the diverging ion
817: density at the plasma edge in the case of a dissipationless plasma
818: expansion. Furthermore, the initial correlation heating of the ions largely
819: increases the thermal ion velocities leading to a broadening of the peak
820: structure compared to the zero-temperature case.
821:
822: Apart from the deviations at the plasma edge, the ionic density is
823: rather well reproduced
824: by the Gaussian approximation for the spatial distribution. In particular, there
825: is good agreement between the rms-radii obtained from the MD simulation and the
826: kinetic model. On the other hand, the spatial distribution of atoms
827: significantly deviates from that of the ions even
828: at relatively early times due to the nonlinear density dependence of the
829: collision rates in Eq.\ (\ref{rate}). However, as also stated in \cite{Rob03},
830: the total number of atoms is too small to
831: significantly influence the macroscopic expansion of the system.
832:
833: \subsubsection{Spatial dependence of the radial velocities}
834:
835: \begin{figure}[tb]
836: \centerline{\psfig{figure=fig6,width=8cm}}
837: \caption{\label{fig6}
838: Distribution of thermal ionic velocities at $t=0.3\:\mu$s sampled from three
839: different regions of the plasma: $r\leq 1.3\sigma$ (a), $1.3\sigma<r\leq 2
840: \sigma$ (b), and $r> 2\sigma$ (c), and from the total plasma volume (d).
841: The solid lines show a fit to a Maxwell-Boltzmann distribution corresponding to
842: the temperatures specified in the figure. Initial-state parameters are the same
843: as in Fig.\ \ref{fig1}.}
844: \end{figure}
845:
846: Another assumption used in the derivation of the kinetic model is the
847: proportionality of the hydrodynamical expansion velocity to the distance
848: from the center of the plasma cloud, $\mathbf{u} = \gamma \mathbf{r}$, both
849: for the ions and the atoms. In order to check
850: this assumption, we have calculated the radial velocity component
851: $v_r={\bf{v}}{\bf{r}}/r$ of each particle, which is plotted as
852: a function of the radial distance from the plasma center in Fig.\ \ref{fig5}.
853: At an early time the velocity distribution is spread out about its mean value
854: predicted from the kinetic approach due to the finite ionic temperature.
855: Note that the expansion is slower near the plasma edge due to the deviation
856: from quasineutrality as discussed above.
857: Consequently, the inner part of the plasma
858: which expands more quickly will catch up with the outer rim, leading to the
859: formation of the density spike seen in Fig.\ \ref{fig4}a. In the case of the
860: H-MD simulation the velocity spread, caused by the initial ion heating,
861: is of the same order of magnitude as the hydrodynamical expansion velocity
862: itself, leading to a significant broadening of the density spike as discussed
863: above. At later stages of
864: the system evolution, the ions cool adiabatically due to the plasma expansion,
865: and the width of the velocity distribution decreases significantly. Moreover,
866: as discussed in connection with the decay of the ion density peak in
867: Fig.\ \ref{fig4}b, the decrease of the ion velocities near the plasma edge
868: apparent at early times has disappeared.
869:
870: A comparison with the result of the kinetic model equations (\ref{par})
871: shows once more that the H-MD simulation not only reproduces the linear
872: radial dependence of the hydrodynamical velocity, but also yields a
873: quantitative agreement between both methods.
874:
875: \subsubsection{Spatial dependence of the thermal velocities}
876:
877: \begin{figure}[tb]
878: \centerline{\psfig{figure=fig7,width=8cm}}
879: \caption{\label{fig7}
880: Same as Fig.\ \ref{fig6}, but for $t=1.5\:\mu$s.}
881: \end{figure}
882: Due to its marginal influence on the plasma expansion dynamics,
883: the role of the ionic temperature $T_{\rm i}$ for the state of the
884: system has not been addressed before. In the cold fluid model of
885: \cite{Rob02,Rob03}, $T_{\rm i}$ has been set to zero in order to follow the
886: long-time plasma dynamics. However, as stated in
887: the introduction, one of the motivations of the current type of experiments
888: was the creation of a strongly coupled plasma. In this context, knowledge
889: of $T_{\rm i}$ is essential since it directly enters the Coulomb coupling
890: parameter which determines the state of the plasma. Moreover, the ionic
891: temperature gives important insight into the relaxation dynamics of the plasma.
892: For comparing the kinetic model with the H-MD calculations,
893: the very definition of $T_{\rm i}$
894: for the MD simulation requires some discussion. As discussed in section II, we
895: assume a Gaussian velocity distribution, i.e.\ a well-defined temperature
896: $T_{\rm i}$, for the plasma ions in our kinetic
897: model. This, of course, is an approximation since the
898: plasma is not created in an equilibrium state.
899: The total kinetic energy of the ions is a sum of the hydrodynamical expansion
900: energy and a contribution due to the thermal motion of the ions. Since the
901: hydrodynamical velocity is directed radially (Eq.\ (\ref{sol})),
902: we determine the thermal energy of the ions from the
903: average of the velocity component perpendicular to the radial direction
904: \begin{equation} \label{temp_md}
905: k_{\rm{B}}T_{\rm{i}}=\frac{m_{\rm{i}}}{2}\left<\left(\frac{{\bf{v}}\times{
906: \bf{r}}}{r}\right)^2\right>=\frac{m_{\rm{i}}}{2}\left<v_{\perp}^2\right>\;.
907: \end{equation}
908: Clearly, such an assignment of a temperature to the average velocity is only
909: well defined if the ion
910: velocities $v_{\perp}$ are distributed according to a Maxwell distribution. In
911: order to check the validity of this requirement, we have sampled the ion
912: velocity distribution $v_{\perp}$ from three different regions in the plasma:
913: $r\leq 1.3\sigma$, $1.3\sigma<r\leq 2\sigma$, and $r> 2\sigma$, which have been
914: chosen so that each region is occupied by approximately the same number of
915: ions. The resulting distributions are plotted at two different times
916: $t=0.3\:\mu$s $=1.3 \omega_{\rm{p,i}}^{-1}$ and
917: $t=1.5\:\mu$s $=7 \omega_{\rm{p,i}}^{-1}$ in Figs.\ \ref{fig6} and
918: \ref{fig7}, respectively. Additionally, we have fitted a Maxwell-Boltzmann
919: distribution to the numerical results, formally defining a temperature in the
920: corresponding plasma region. As can be seen in Fig.\ \ref{fig6}, even at the
921: very early stage of the plasma evolution the numerical
922: data is well fitted by an equilibrium distribution in the inner plasma region,
923: while there are deviations in the outer region of the plasma since the
924: relaxation time is longer due to the lower density far away from the plasma
925: center. However, already after a relatively short time of $t=1.5\:\mu$s the
926: velocity distributions are well fitted by a Maxwell-Boltzmann distribution in
927: all three plasma regions (Fig.\ \ref{fig7}). Hence, the ion thermal energy can
928: be represented by a local temperature $T_{\rm i}(r)$, decreasing with growing
929: distance from the plasma
930: center as can be seen from Figs.~\ref{fig6} and \ref{fig7}. This is due to
931: the fact that the initial heating arises from a
932: compensation of the negative correlation energy, which is larger in the central
933: plasma region where the density is higher. However, as becomes apparent by
934: comparing Figs.\ \ref{fig6} and \ref{fig7}, the thermal energy equilibrates over
935: the whole plasma volume rather quickly as the system evolves. While the
936: temperatures defined in the inner and outermost region deviate by a factor of
937: eight at $t=0.3\:\mu$s, they differ by a factor of two only $1.2\:\mu$s later.
938:
939: \begin{figure}[tb]
940: \centerline{\psfig{figure=fig8,width=8cm}}
941: \caption{\label{fig8}
942: Average thermal ionic energy as a function of the distance from the plasma
943: center at four different times: $t=0.3\:\mu$s (solid), $t=1.5\:\mu$s (dashed),
944: $t=6.0\:\mu$s (dotted) and $t=25.3\:\mu$s (dot-dashed). Initial-state
945: parameters are the same as in Fig.\ \ref{fig1}.}
946: \end{figure}
947: Figure \ref{fig8} gives a more detailed account of this equilibration process.
948: Here, the local ionic temperature is plotted as a
949: function of the radial distance from the plasma center at four different
950: times, where $T_{\rm i}(r)$ has been defined from the velocity average of a
951: shell of 2000 ions with a central shell radius $r$.
952: The temperature decrease with increasing distance from the center as discussed
953: above is clearly
954: visible. Nevertheless, the ion temperature is seen to equilibrate rather
955: quickly, so that the approximation of a homogeneous ion temperature, used in
956: the derivation of the kinetic model in section \ref{za}, becomes better and
957: better at later
958: times. Moreover, the numerically calculated distribution of thermal velocities
959: sampled over the whole plasma volume is well represented by a Maxwell-Boltzmann
960: distribution with some average temperature intermediate between the
961: temperatures of the inner and outer region, respectively (Fig.\ \ref{fig7}d).
962: This shows that the Gaussian phase-space distribution assumed for the ions in
963: section \ref{za} agrees very well with the results of the MD simulation
964: averaged over the spatial coordinates, even if the
965: temperature still shows substantial inhomogeneities.
966:
967: \subsection{Spatially averaged ionic observables}
968:
969: As we have demonstrated in section \ref{GlobalResults} the kinetic model
970: describes the global temporal evolution of the plasma including recombination
971: quite accurately. From the detailed analysis of the spatially resolved plasma
972: dynamics in the previous subsection we
973: may expect that the kinetic model describes spatially averaged
974: observables, such as the kinetic energy of the expansion, the thermal energy,
975: and the correlation energy of the plasma quite well. This is indeed the case
976: over almost the entire evolution time as Fig.\ \ref{fig9} demonstrates
977: for the correlation energy and the thermal ion energy.
978: Only at an early stage of the plasma evolution, differences
979: between MD simulation and kinetic model are visible, showing that the
980: correlation-time approximation Eq.\ (\ref{CTA}) does not accurately
981: describe this early phase of equilibration starting from a completely
982: uncorrelated state in all details. Since the initial state is very far from
983: equilibrium, the initial relaxation process is not exponential, as assumed in
984: the correlation-time approximation Eq.\ (\ref{CTA}). Rather,
985: it is connected with transient oscillations of the temperature (inset of
986: Fig.\ \ref{fig9}) which have been found both theoretically
987: \cite{Zwi,Mor03,PPR04b} and experimentally \cite{Kilpri}.
988: However,
989: the timescale of the initial ion heating as well as the maximum temperature
990: are well reproduced by the simple model. After the system has come
991: sufficiently close to local equilibrium, the quality of the correlation-time
992: approximation becomes better and, once
993: again, close agreement between the two approaches is found, supporting our
994: argument put forward in the derivation of the kinetic approach in section
995: II. At later times differences become apparent, which may be attributed to the
996: fact that the ion relaxation is considerably disturbed by recombination and
997: ionization events leading to sudden local changes of the charge density,
998: which is not taken into account by the kinetic model.
999:
1000: Furthermore, according to both approaches, the correlation energy and the
1001: thermal kinetic energy of the ions
1002: are almost identical roughly to the time where both curves reach their
1003: maximum values, showing that the total correlation energy is completely
1004: converted into thermal kinetic energy of the ions, as expressed by Eq.\
1005: (\ref{parc}). At later times, this additional kinetic energy is transferred to
1006: the outward directed motion of the ions, leading to an indirect enhancement of
1007: the plasma expansion by the development of IC and to adiabatic
1008: cooling of the ions. Therefore, the thermal ion kinetic energy starts to
1009: deviate from the correlation energy as the plasma expansion sets in.
1010: \begin{figure}[tb]
1011: \centerline{\psfig{figure=fig9,width=8cm}}
1012: \caption{\label{fig9}
1013: Correlation energy (solid line: H-MD simulation, dot-dashed: kinetic
1014: result) and thermal ion kinetic energy (dashed: H-MD, dotted: kinetic).
1015: Initial-state parameters are the same as in Fig.\ \ref{fig1}. The inset
1016: shows the ion thermal energy in the early stage of the relaxation process
1017: with its characteristic transiently oscillatory behavior.}
1018: \end{figure}
1019: \section{Conclusions}
1020: In summary, we have presented two different theoretical approaches for the
1021: simulation of ultracold neutral plasmas. First, we have introduced a simple
1022: kinetic model along the lines of \cite{Rob02}, and we have shown how to include
1023: a description
1024: of IC into the model in an approximate way. Moreover, we
1025: have developed a hybrid molecular dynamics approach which allows for an
1026: accurate description of the strongly coupled ion motion on microsecond
1027: timescales, by treating the electronic component as a fluid using an adiabatic
1028: approximation while the ions are fully accounted for on an MD level.
1029:
1030: Supporting our results from \cite{PPR03},
1031: both methods show that the inclusion of IC enhances the
1032: number of recombined Rydberg atoms by a few percent, but only slightly
1033: affects the macroscopic expansion dynamics of the plasma itself. As we have
1034: shown, this is due to the fact that mainly the population of very highly
1035: excited states is increased if IC are taken into account,
1036: which have a small binding energy and therefore hardly influence the electron
1037: temperature.
1038:
1039: By comparison of the two methods,
1040: we could show that the simple kinetic description adequately describes the
1041: evolution of global, i.e.\ spatially averaged, plasma observables.
1042: Thus, the kinetic model, which allows for a much faster
1043: computation, may be used to quickly and efficiently scan the vast space of
1044: initial-state parameters, e.g.\ in order to obtain a ``phase diagram'' for
1045: Rydberg gas / plasma systems \cite{FBS}. Moreover, it permits extending the
1046: description of ultracold plasmas to a parameter range where the plasma is so
1047: large that the number of particles ($N_i \agt 10^6$) prohibits an MD
1048: simulation. Maybe even more importantly,
1049: the simple kinetic equations give additional insight into the
1050: dynamics beyond that possible on the basis of MD simulations by providing
1051: simple evolution equations for the macroscopic parameters describing the
1052: plasma state.
1053:
1054: On the other hand, spatially resolved quantities such as ionic density, ion
1055: velocities or local temperature show deviations from the behavior
1056: predicted by the simple kinetic model. However,
1057: the developed H-MD approach provides a powerful method for the study of these
1058: quantities, and for the detailed description of the relaxation dynamics of the
1059: strongly coupled ions on a microsecond timescale.
1060: Moreover, it permits the study of scenarios where the ions are so strongly
1061: coupled that Coulomb crystallization occurs \cite{PPR04},
1062: which cannot be described by the kinetic model.
1063:
1064: \begin{acknowledgments}
1065: We gratefully acknowledge helpful discussions with T.C.\ Killian and F.\
1066: Robicheaux. This work was supported by the DFG within the Priority
1067: Programme SPP1116 (Grant-No.\ RO1157/4).
1068: \end{acknowledgments}
1069:
1070: \begin{appendix}
1071: \section{Derivation of the correlation force}
1072: In this section, the approximation Eq.~(\ref{Fc}) for $\mathbf{F}_{\rm{ii}}$
1073: is derived. We start from Eq.~(\ref{cforce})
1074: \begin{equation} \label{a1b}
1075: \mathbf{F}_{\rm{ii}} = e^2\int \rho_{\rm{i}}({\mathbf{r}}^{\prime}) \,
1076: g_{\rm{ii}}({\mathbf{r}}, {\mathbf{r}}^{\prime}) \, \frac{\mathbf{r}- {
1077: \mathbf{r}}^{\prime}}{\left| {\mathbf{r}}-{\mathbf{r}}^{\prime}\right|^3} \,
1078: d{\mathbf{r}}^{\prime} \;\; ,
1079: \end{equation}
1080: where the explicit expression
1081: $\varphi_{\rm{ii}}=e^2/|{\mathbf{r}}-{\mathbf{r}}^{\prime}|$ for the
1082: inter-ionic Coulomb potential has been inserted.
1083: In general, the correlation function $g_{\rm{ii}}$ is a function of both
1084: coordinates
1085: $\mathbf{r}$ and ${\mathbf{r}}^{\prime}$. However, in the case of a homogeneous
1086: density, $g_{\rm{ii}}$ depends only on the interparticle distance
1087: $x=|{\mathbf{r}}-{\mathbf{r}}^{\prime}|$.
1088: Since the relevant property which distinguishes the two points
1089: ${\mathbf{r}}$ and ${\mathbf{r}}^{\prime}$ is the corresponding density (from
1090: the way the plasma is created, no other differences, e.g.\ that part of the
1091: plasma would be in a state with equilibrium correlations while a different
1092: part would be totally uncorrelated, are apparent), it
1093: seems a reasonable approximation to assume that the space dependence of the
1094: correlation function enters only via the densities at
1095: the respective coordinates \cite{Eva79}.
1096: Hence, we write the correlation function as
1097: \begin{equation} \label{a2}
1098: g_{\rm{ii}}({\mathbf{r}},{\mathbf{r}}^{\prime}) \approx g_{\rm{ii}}\left[
1099: \rho_{\rm{i}}({\mathbf{r}}),\rho_{\rm{i}}(
1100: {\mathbf{r}}^{\prime}),|{\mathbf{r}}-{\mathbf{r}}^{\prime}|\right] \;\; .
1101: \end{equation}
1102: With the substitution ${\mathbf{r}}^{\prime}={\mathbf{r}}+{\mathbf{x}}$,
1103: Eq.~(\ref{a1b}) becomes
1104: \begin{equation} \label{a3b}
1105: \mathbf{F}_{\rm{ii}} = -e^2\int \rho_{\rm{i}}({\mathbf{r}}+{\mathbf{x}})
1106: g_{\rm{ii}}\left[\rho_{\rm{i}}({\mathbf{r}}), \rho_{\rm{i}}({\mathbf{r}}+
1107: {\mathbf{x}}),x\right] \frac{\mathbf{x}}{x^3} \, d{\mathbf{x}} \;\; .
1108: \end{equation}
1109: Since the correlation function rapidly decreases for distances $x$ larger than
1110: the correlation length $\lambda_{\rm{c}}$, we may restrict the integration in
1111: Eq.~(\ref{a3b}) to a sphere with a radius of approximately
1112: $\lambda_{\rm{c}}$. If the
1113: plasma density does not vary strongly on the scale of the correlation
1114: length, we may use a linear Taylor expansion of the density
1115: \begin{equation} \label{a4}
1116: \rho_{\rm{i}}\left(\mathbf{r}+\mathbf{x}\right)\approx\rho_{\rm{i}}\left(
1117: \mathbf{r}\right)+
1118: \mathbf{x}\cdot\nabla \rho_{\rm{i}} \left(\mathbf{r}\right)
1119: \end{equation}
1120: and the correlation function
1121: \begin{equation} \label{a5}
1122: g_{\rm{ii}}(\rho_{\rm{i}},\rho_{\rm{i}}^{\prime},x)=g_{\rm{ii}}(\rho_{\rm{i}},
1123: \rho_{\rm{i}},x)+ \left. \frac{\partial
1124: g_{\rm{ii}}(\rho_{\rm{i}},\rho_{\rm{i}}^{\prime},x)}{\partial \rho_{
1125: \rm{i}}^{\prime}} \right|_{\rho_{\rm{i}}^\prime=\rho}
1126: \; \left(\mathbf{x}\cdot\nabla\rho_{\rm{i}} \right) \;\; ,
1127: \end{equation}
1128: where $\rho_{\rm{i}}=\rho_{\rm{i}}(\mathbf{r})$ and $\rho_{\rm{i}}^{\prime}=
1129: \rho_{\rm{i}}(\mathbf{r}+\mathbf{x})$.
1130:
1131: Substitution of Eq.\ (\ref{a4}) and Eq.~(\ref{a5})
1132: into Eq.~(\ref{a3b}) and keeping only terms up to linear order in
1133: ${\bf{x}}\cdot\nabla\rho_{\rm{i}}$ yields
1134: \begin{widetext}
1135: \begin{equation} \label{a6}
1136: \mathbf{F}_{\rm{ii}}=-e^2\left(
1137: \int \frac{\mathbf{x}}{x^3} \; \rho_{\rm{i}} \; g_{\rm{ii}}(\rho_{\rm{i}},x)\;
1138: d\mathbf{x} + \int \frac{\mathbf{x}}{x^3} \; g_{\rm{ii}}(\rho_{\rm{i}},x)\;
1139: \left(\mathbf{x}\cdot \nabla\rho_{\rm{i}}\right) \; d\mathbf{x} \right. +
1140: \left. \frac{1}{2} \int \frac{\mathbf{x}}{x^3} \; \rho_{\rm{i}} \;
1141: \frac{\partial g_{\rm{ii}}(\rho_{\rm{i}},x)}{\partial \rho_{\rm{i}}}\;\left(
1142: \mathbf{x}\cdot \nabla\rho_{\rm{i}}\right)\; d\mathbf{x} \right) \; ,
1143: \end{equation}
1144: where $g_{\rm{ii}}(\rho_{\rm{i}},x) \equiv \left. g(\rho_{\rm{i}},
1145: \rho_{\rm{i}}^\prime, x) \right|_{\rho_{\rm{i}}^\prime =
1146: \rho_{\rm{i}}}$ and we have used the relation
1147: \begin{equation} \label{ag}
1148: \frac{\partial}{\partial \rho_{\rm{i}}} \left( \left. g_{\rm{ii}}(\rho_{\rm{i}},
1149: \rho_{\rm{i}}^\prime, x)
1150: \right|_{\rho_{\rm{i}}^\prime = \rho_{\rm{i}}} \right) = \left. \frac{\partial
1151: g(\rho_{\rm{i}},
1152: \rho_{\rm{i}}^\prime, x)}{\partial \rho_{\rm{i}}} \right|_{\rho_{\rm{i}}^\prime
1153: = \rho_{\rm{i}}} +
1154: \left. \frac{\partial g_{\rm{ii}}(\rho_{\rm{i}}, \rho_{\rm{i}}^\prime,
1155: x)}{\partial \rho_{\rm{i}}^\prime}
1156: \right|_{\rho_{\rm{i}}^\prime = \rho_{\rm{i}}} = 2 \left. \frac{\partial
1157: g(\rho_{\rm{i}}, \rho_{\rm{i}}^\prime,
1158: x)}{\partial \rho_{\rm{i}}^\prime} \right|_{\rho_{\rm{i}}^\prime =
1159: \rho_{\rm{i}}} \;\; ,
1160: \end{equation}
1161: \end{widetext}
1162: which follows from the symmetry of $g_{\rm{ii}}$ under particle exchange,
1163: i.e., under exchange
1164: of $\rho_{\rm{i}}$ and $\rho_{\rm{i}}^{\prime}$.
1165: Since the integrand of the first integral in Eq.~(\ref{a6}) is an odd function
1166: in ${\bf{x}}$ the first term vanishes. The second term yields after some
1167: manipulations
1168: \begin{equation}
1169: e^2 \int \frac{\mathbf{x}}{x^3} \; g_{\rm{ii}}(\rho_{\rm{i}},x) \;\left(
1170: \mathbf{x} \cdot \nabla \rho_{\rm{i}}\right)\; d\mathbf{x}
1171: =\frac{e^2}{3}\; \nabla\rho_{\rm{i}}\; \int\frac{g_{\rm{ii}}(\rho_{\rm{i}},
1172: x)}{x} \; d\mathbf{x}\;.
1173: \end{equation}
1174: Analogously, the third term can be written as
1175: \begin{equation}
1176: \frac{e^2}{2} \int \frac{\mathbf{x}}{x^3} \; \rho_{\rm{i}} \; \frac{\partial
1177: g_{\rm{ii}}(\rho_{\rm{i}},x)}{\partial \rho_{\rm{i}}}\;\left(\mathbf{x} \cdot
1178: \nabla\rho_{\rm{i}}\right) \; d\mathbf{x}=\frac{e^2}{6} \rho_{\rm{i}}\; \nabla\rho_{\rm{i}}\frac{\partial}{\partial
1179: \rho_{\rm{i}}} \int\frac{g_{\rm{ii}}(\rho_{\rm{i}},x)}{x} \; d\mathbf{x}\;,
1180: \end{equation}
1181: which together leads to
1182: \begin{eqnarray} \label{a7}
1183: \mathbf{F}_{\rm{ii}}=-\frac{e^2}{6}\;\nabla\rho_{\rm{i}}\;\left[\int\frac{
1184: g_{\rm{ii}}(\rho_{\rm{i}},x)}{x} \; d\mathbf{x}+\; \frac{\partial}{\partial \rho_{\rm{i}}}\left(\rho_{\rm{i}}\int
1185: \frac{g_{\rm{ii}}(\rho_{\rm{i}},x)}{x}\;d\mathbf{x} \right) \right].
1186: \end{eqnarray}
1187: Finally, substitution of the definition of the correlation energy $u$ as given
1188: by Eq.~(\ref{uc}) yields the result Eq.~(\ref{Fc}).
1189: An analogous calculation for the expression of the correlation energy
1190: Eq.~(\ref{Uc2}) leads to the familiar LDA result \cite{Eva79}
1191: \begin{eqnarray} \label{a8}
1192: U_{\rm{ii}}&=&\frac{e^2}{2N_{\rm{i}}}\int \rho_{\rm{i}}({\mathbf{r}})\rho_{
1193: \rm{i}}({\mathbf{r}}^{\prime}) \, g_{\rm{ii}}({\mathbf{r}},
1194: {\mathbf{r}}^{\prime}) \, \frac{1}{\left| {\mathbf{r}}-{\mathbf{r}}^{\prime}
1195: \right|} \, d{\mathbf{r}}^{\prime}d{\mathbf{r}}\nonumber \\
1196: &=&N_{\rm{i}}^{-1}
1197: \int\rho_{\rm{i}}u_{\rm{ii}}\; d\mathbf{r}\;.
1198: \end{eqnarray}
1199: \end{appendix}
1200:
1201: \begin{thebibliography}{37}
1202: \bibitem{Kil99} T.C.\ Killian, S.\ Kulin, S.D.\ Bergeson, L.A.\ Orozco,
1203: C.\ Orzel and S.L.\ Rolston, Phys.\ Rev.\ Lett.\ {\bf 83}, 4776 (1999).
1204:
1205: \bibitem{Kul00} S.\ Kulin, T.C.\ Killian, S.D.\ Bergeson and S.L.\ Rolston,
1206: Phys.\ Rev.\ Lett.\ {\bf 85}, 318 (2000).
1207:
1208: \bibitem{Kil01} T.C.\ Killian, M.J.\ Lim, S.\ Kulin, R.\ Dumke, S.D.\ Bergeson
1209: and S.L.\ Rolston, Phys.\ Rev.\ Lett.\ {\bf 86}, 3759 (2001).
1210:
1211: \bibitem{Rob00} M.P.\ Robinson, B.L.\ Tolra, M.W.\ Noel, T.F.\ Gallagher and
1212: P.\ Pillet, Phys.\ Rev.\ Lett.\ {\bf 85}, 4466 (2000).
1213:
1214: \bibitem{Eyl00} E.\ Eyler, A.\ Estrin, J.R.\ Ensher, C.H.\ Cheng, C.\ Sanborn
1215: P.L.\ Gould, Bull.\ Am.\ Phys.\ Soc.\ {\bf 45}, 56 (2000).
1216:
1217: \bibitem{Gou03} P.L.\ Gould, S.M.\ Farooqi, S.\ Krishnan, J.\ Stanojevic,
1218: D.\ Tong, Y.P.\ Zhang, J.R.\ Ensher, A.\ Estrin, C.-H.\ Cheng and E.E.\ Eyler,
1219: in {\em Interactions in Ultracold Gases: From Atoms to Molecules}, eds.\
1220: M.\ Weidem\"uller and C.\ Zimmermann (Wiley-VCH, 2003), p.\ 270.
1221:
1222: \bibitem{Kuz02} S.G.\ Kuzmin and T.M.\ O'Neil, Phys.\ Rev.\ Lett.\ {\bf 88},
1223: 065003 (2002).
1224:
1225: \bibitem{Kuz02b} S.G.\ Kuzmin and T.M.\ O'Neil, Phys.\ Plasmas {\bf 9}, 3743
1226: (2002).
1227:
1228: \bibitem{Maz02} S.\ Mazevet, L.A.\ Collins and J.D.\ Kress, Phys.\ Rev.\
1229: Lett.\ {\bf 88}, 055001 (2002).
1230:
1231: \bibitem{Rob02} F.\ Robicheaux and J.D.\ Hanson, Phys.\ Rev.\ Lett.\ {\bf 88},
1232: 055002 (2002).
1233:
1234: \bibitem{Rob03} F.\ Robicheaux and J.D.\ Hanson, Phys.\ Plasmas {\bf 10},
1235: 2217 (2003).
1236:
1237: \bibitem{Tka01} A.N.\ Tkachev and S.I.\ Yakovlenko, Quantum Electronics
1238: {\bf 31}, 1084 (2001).
1239:
1240: \bibitem{PPR03} T.\ Pohl, T.\ Pattard and J.M.\ Rost, Phys.\ Rev.\ A {\bf 68},
1241: 010703(R) (2003).
1242:
1243: \bibitem{Dub99} D.H.E.\ Dubin and T.M.\ O'Neil, Rev.\ Mod.\ Phys.\ {\bf 71},
1244: 87 (1999).
1245:
1246: \bibitem{Ich82} S.\ Ichimaru, Rev.\ Mod.\ Phys.\ {\bf 54}, 1017 (1982).
1247:
1248: \bibitem{PPR04} T.\ Pohl, T.\ Pattard and J.M.\ Rost, Phys.\ Rev.\ Lett.\
1249: {\bf 92}, 155003 (2004).
1250:
1251: \bibitem{Mur01} M.S.\ Murillo, Phys.\ Rev.\ Lett.\ {\bf 87}, 115003 (2001).
1252:
1253: \bibitem{PPR04b} T.\ Pohl, T.\ Pattard and J.M.\ Rost, J.\ Phys.\ B {\bf 37},
1254: L183 (2004).
1255:
1256: \bibitem{Dor98} D.S.\ Dorozhkina and V.E.\ Semenov, Phys.\ Rev.\ Lett.\
1257: {\bf 81}, 2691 (1998).
1258:
1259: \bibitem{Sem99} D.\ Semkat, D.\ Kremp and M.\ Bonitz, Phys.\ Rev.\ E
1260: {\bf{59}}, 1557 (1999).
1261:
1262: \bibitem{Bon96a} M.\ Bonitz, Phys.\ Lett.\ A {\bf{221}}, 85 (1996).
1263:
1264: \bibitem{Bon96b} M.\ Bonitz and D.\ Kremp, Phys.\ Lett.\ A {\bf{212}}, 83
1265: (1996).
1266:
1267: \bibitem{Mor98} K.\ Morawetz, V.\ Spicka, P.\ Lipavsk\'y, Phys.\ Lett.\ A
1268: {\bf{246}}, 311 (1998).
1269:
1270: \bibitem{Cha98} G.\ Chabrier and A.Y.\ Potekhin, Phys.\ Rev.\ E {\bf{58}},
1271: 4941 (1998).
1272:
1273: \bibitem{Man69} P.\ Mansbach and J.\ Keck, Phys.\ Rev.\ {\bf 181}, 275 (1969).
1274:
1275: \bibitem{Cha43} S.\ Chandrasekhar, Astrophys. J. {\bf{98}}, 54 (1943).
1276:
1277: \bibitem{Kin66} I.R.\ King, Astron. J. {\bf{71}}, 64 (1966).
1278:
1279: \bibitem{Bar86} J.E. Barnes and P.\ Hut, Nature {\bf{324}}, 446 (1986).
1280:
1281: \bibitem{Bar90} J.E.\ Barnes, J. Comp. Phys. {\bf{87}}, 161 (1990).
1282:
1283: \bibitem{Bir95} C.K.\ Birdsall and A.B.\ Langdon, {\em
1284: Plasma physics via computer simulation}
1285: (Bristol: Inst. of Physics Publ., 1995).
1286:
1287: \bibitem{Gur66} A.V.\ Gurevich, L.V.\ Pariiskaya and L.P.\ Pitaevskii,
1288: Soviet Phys. JETP {\bf{22}}, 449 (1966)
1289:
1290: \bibitem{Sac85} C.\ Sack and H.\ Schamel, Plasma.\ Phys.\ Contr.\ F. {\bf 27},
1291: 717 (1985).
1292:
1293: \bibitem{Zwi} G.\ Zwicknagel, Contrib.\ Plasma Phys.\ {\bf 39}, 155 (1999).
1294:
1295: \bibitem{Mor03} I.V.\ Morozov, G.E.\ Norman, J.\ Phys.\ A {\bf 36},
1296: 6005 (2003).
1297:
1298: \bibitem{Kilpri} T.C.\ Killian, private communication (2004).
1299:
1300: \bibitem{FBS} T.\ Pattard, T.\ Pohl and J.M.\ Rost, Few-Body Systems,
1301: in press (2004).
1302:
1303: \bibitem{Eva79} R.\ Evans, Adv.\ Phys. {\bf 28}, 143 (1979).
1304:
1305: \end{thebibliography}
1306:
1307: \end{document}
1308: