Anda di halaman 1dari 6

Abstract

Liquid helium is the prototypical example of a superfluid - a liquid that flows without viscosity
and transfers heat without a temperature gradient. These properties are intimately related to
the Bose condensation that occurs in this strongly interacting liquid. Bose condensation is most
directly observed in the single particle atomic momentum distribution, where the Bose
condensate appears as a delta function singularity. In this article, we discuss the experimental
techniques used to observe the condensate and the current status of measurments of the Bose
condensate in liquid helium.

1 Introduction

Liquid helium (4He) has fascinated physicists ever since Kammerlingh-Onnes liquified the last of
the so called permanent gases in 1908. However, evidence of what is without doubt the most
fascinating property of this unique liquid, superfluidity, was not reported until almost 25 years
later. The superfluid phase, where heat is transferred without a thermal gradient and mass flows
without a driving pressure, is a macroscopic manifestation of microscopic quantum effects
governing the behavior of atoms. Bose-Einstein condensation was first proposed by London as
the microscopic explanation for these fascinating phenomena.

Helium is unique among condensed atomic systems since the bulk properties of the liquid are
dominated by quantum effects. The most important of these are the statistical effects that arise
for identical particles. For bosons, such as 4He atoms, there is no limitation on the number of
particles that can occupy a single quantum state. Thus, in the absence of interactions, all the
atoms in the liquid would occupy a single momentum state at zero temperature.

Liquid helium is also unique among the Bose systems considered in this volume. Other systems,
such as spin polarized hydrogen and excitons, are expected to exhibit Bose condensation. (See
the review articles by Greytak and Silvera, Castin et al, Mysyrowicz and Wolfe et al. in this
volume.) At present, however, liquid helium is the only system where the existence of an
experimentally attainable Bose condensed phase is almost universally accepted.

Unlike many of the other systems considered in this volume, liquid helium is a strongly
interacting system. The interaction between the atoms, which is dominated by the hard core
repulsion, is strong enough that the properties of the liquid cannot be treated as a simple
perturbationof the Ideal Bose Gas, as is the case for more weakly interacting systems. As we will
see, these strong interactions have greatly complicated both the theoretical and experimental
studies of the Bose-condensed phase of superfluid 4He.

We begin this review with a discussion of the Ideal Bose Gas, a model system consisting of non-
interacting particles. This model, while not applicable to the strongly interacting liquid phase of
helium, is useful in that it provides a system where the effects of Bose condensation are clearl
delineated. The microscopic properties of liquid helium, which we discuss next, are quite
different from the ideal gas due to the strong interactions between atoms. Despite these
differences, similarities between the ideal gas and the liquid systems remain.

In particular, a Bose-condensed phase still appears in liquid 4He at low temperatures, although
the fraction of particles in the condensate is considerably reduced. Therefore, in this review, the
analogies between an ideal Bose gas and superfluid 4He are stressed, even though the former
does not exhibit the long-range phase coherence associated with Bose-broken symmetry. (For a
discussion of this, see especially the articles in this book by Nozieres, Huang, Stringari and Stoof.)

The single-particle momentum distribution is one of the few quantities which directly reflects
the appearance of a Bose condensate. Therefore, we next turn our attention to the experimental
technique that provides information on the momentum distribution: Deep Inelastic Neutron
Scattering (DINS). This technique, due to the information on n(p) that it provides, provides one of
the few opportunities to determine no directly.

Early attempts to measure the momentum distribution, and to obtain direct information on the
condensate using DINS, were limited by the fluxes and energies of neutrons available. Recent
advances in neutron sources, which now have both higher fluxes and higher energies, have led
to a new generation of experiments. These new experiments are reviewed, and the information
that can be extracted, both directly and with the help of models, are described.

Finally, we turn our attention to the direct evidence for the condensate in liquid helium. The
experimentally observed scattering shows distinct changes consistent with the appearance of a
condensate. However, the sought after direct evidence, an observation of the (5-function
singularity associated with the condensate, is not observed. Nevertheless, comparison to current
theoretical predictions and to empirical expressions based on realistic models give a strong case
for a finite value of no and reasonable results for its temperature and density dependence.

5 Experimental Results

The suggestion that neutron scattering measurements at large momentum transfers could
directly observe the condensate set in motion a number of studies by a variety of investigators.
Unfortunately, none of these studies has succeeded in reaching the original goal: a direct and
unambiguous observation of the condensate. In fact, with our current understanding of these
measurements and FSE which complicate them, we now realize that a direct observation of a 3
-function component in n(p) is unlikely in the foreseeable future. However, these studies of
liquid helium have provided a wealth of detailed information on the momentum distribution in
this strongly interacting quantum liquid.

With the use of some well-founded theoretical results, information on the magnitude of the
condensate fraction can be extracted, including its temperature and density dependence. The
earliest attempts to measure n(p) and no were carried out at reactor based sources. These
measurements provided general information on n(p) and pointed the way toward techniques to
extract the condensate.

However, they were limited by the large FSE at the relatively low Q attainable at reactor based
sources. Some measurements were carried out at larger Q, but only at the expense of resolution
making it difficult to obtain accurate information on n(p). More recently, measurements have
been carried out using spallation neutron sources, such as the Intense Pulsed Neutron Source
(IPNS) at Argonne National Laboratory. These sources have a high flux of epithermal (high
energy) neutrons allowing high resolution measurements to be carried out at much larger Q
than are obtainable with reactors. The higher Q have the advantage that the scattering is
consistent with the predictions of the IA, in terms of the location and symmetry of the observed
scattering. In addition, FSE are more amenable to theoretical treatment [30] at these higher Q.
Therefore, information on the underlying n(p) can be more readily obtained.

To illustrate the results obtainable at spallation sources, we will discuss recent measurements
using the high resolution PHOENIX spectrometer at the IPNS at Argonne. Fig. 6 shows the
measured scattering at a Q of 23 A"1, converted to J(Y), at 4.2 K in the normal liquid. The
scattering shown in Fig. 6 is in qualitative agreement with the predictions of the IA. It is centered
at, and symmetric about, Y — 0. In addition, measurements at a variety of Q have shown that
the shape of J(Y) is independent of Q above approximately 15 A"1. Thus, at least at this
qualitative level, the scattering is well described by the IA.

The observed scattering in the normal liquid is broad and featureless. J(Y) is nearly Gaussian, as
in classical liquids. Upon cooling into the superfluid phase the scattering, shown in Fig. 7,
becomes visibly more peaked near 7 = 0 . However, no distinct condensate peak is observed.

The increase in scattering at small Y is consistent with the appearance of a condensate peak
broadened by the finite instrumental resolution and FSE. However, as we discuss below, it is also
consistent with many other forms for n(p) as well, some of which do not have a finite no.

The temperature dependence of the neutron scattering is displayed in Fig. 8. The scattering is
nearly Gaussian and temperature independent above the superfluid transition. Below the
transition, the scattering becomes non-Gaussian with an increase in intensity around 7 = 0 ,
consistent with the appearance of a condensate. There is a rapid change in the scattering from
just below the transition to 1.5 K, after which the scattering changes little. This is consistent with
Path Integral Monte Carlo predictions that the condensate fraction is largely temperature
independent below 1.5 K but decreases rapidly above that temperature.

The ultimate goal of these studies is to obtain information on n(p) in general and the condensate
in particular. Towards that end, we would like to remove the effects of instrumental resolution
and FSE and to invert the transformation between n(p) and J(Y). However, as discussed
previously, these are ill-defined procedures that are strongly affected by statistical noise present
in the experimental data. Therefore, rather than attempting to deconvolute the instrumental
resolution, it is more appropriate to fit an expression for J(Y), broadened by instrumental
resolution and FSE, to the observed scattering data. Theoretical predictions for J(Y) are discussed
in the following section. Here we discuss a simple expressionfor J(Y) which has sufficient
flexibility to reflect the behavior of the true scattering accurately, yet which is also sufficiently
constrained so that unphysical behavior is not introduced due to the finite statistical ccuracy of
the data.

The scattering function that we have found most convenient for describing the observed
scattering is a sum of two Gaussians whose amplitudes, widths and common center may be
varied. This form is not unique and many other forms could be used to fit the data.

Nevertheless it has sufficient flexibility that it can represent the scattering in both the normal
liquid, where the scattering is nearly Gaussian, and the superfluid. Fig. 9 shows the resultant
model J(Y), which represents the scattering after the removal of instrumental resolution and FSE,
over a broad range of temperatures. As can be seen, the corrected scattering is nearly Gaussian
above Tx. Below Tx the scattering becomes non-Gaussian, with a significant increase in the
strength at small Y. However, in no cases do the fits indicate that we need a 6 -function
component to describe the scattering adequately.

The above results indicate that a <5-function component is not necessary to explain the
scattering in the superfluid. This does not, however, mean that the scattering is not consistent
with a momentum distribution containing a sharp peak at p = 0. As an illustration of the range of
model scattering functions which can fit the data equally well, consider the three different two-
Gaussian fits to the scattering at T = 0.35 K shown in Fig. 10. The dashed and dotted lines show
the underlying distributions which best fit the data if they are constrained to possess a narrow
component by fixing the width of one of the Gaussians to be very narrow {a = 0.03 in this case).
The dashed line has a narrow component with 10% of the total area, and the dotted line has a
narrow component with 4% of the total area. Finally, the solid line is a two- Gaussian fit in which
both the amplitudes and widths are allowed to vary.

The x2 values of these fits are very close to each other. If we think of the narrow component of
the fits as representing the contribution from the condensate and the broad component as
representing the momentum distribution of the uncondensed atoms, then this example shows
that, with an appropriate choice for the uncondensed component, the scattering data is
consistent with a condensate fraction ranging from zero (no condensate) to values above 10 %.

6 Comparison to Theory

A direct model-independent determination of no is beyond current experimental capabilities for


the reasons discussed in Section 5. However, we may still obtain information on the condensate
by comparing theoretical calculations of n(p) with the experimental data. Such comparisons
provide a direct test of the theoretical predictions and, indirectly, give information about the
magnitude of the condensate.

The dashed line in Fig. 11 (a) shows the theoretical prediction for JIA(Y) in the normal liquid
using the PIMC calculations of n(p). The theoretical n(p) has been converted to J(Y) using the IA
and broadened by the instrumental resolution. The agreement between the theoretical
predication and the experiment is excellent. In this case the IA, calculated using the theoretical
n(p)9 provides an excellent description of the scattering in the normal liquid.

FSE have little effect on the observed scattering in the normal liquid at these Q values. However,
they may be included by convoluting the theoretical predictions with the broadening function
shown in Fig. 6. The solid lines in Fig. 11 (a) and (b) are obtained when FSE are included.

Within the statistical accuracy of the measurements, there is no observable change when FSE are
included. Since, based on the co2 sum rule, FSE do not change the second moment of the
scattering, they have little effect on the broad, nearly Gaussian J(Y) of the normal liquid.

The dashed line in Fig. ll(b) shows a similar comparison of the GFMC calculations to the
scattering in the superfluid. The agreement between the theoretical and experimental results is
quite poor, particularly in the region of the peak center, where the condensate would have the
largest contribution. Based on this comparison alone, which has neglected FSE, we would
conclude that no, if not identically zero, would have a much smaller value than the theoretical
predictions.
The inclusion of FSE in the comparison between the experimental and theoretical results is
essential in the superfluid phase. This is due to the appearance of a sharp feature in n(p) for the
superfluid: the condensate. While FSE have little effect on the broad component of the
scattering, as observed in the normal liquid, they significantly broaden the contribution from the
condensate. Taking FSE into account, the agreement between theory and experiment is now
excellent! For the first time, ab initio numerical calculations of n(p) in the superfluid are in good
agreement with experiment [40].

An important point regarding the final-state corrections is in order here. The final-state
broadening prediction of Silver, as shown in Fig. 6, has a narrow central peak and negative tails at
high Y. The negative tails are essential if the broadening function is to satisfy the second-moment
sum rule. Thus, final-state effects not only broaden the condensate peak, they also shift intensity
throughout the entire spectrum. For the particular form of the FSE used here, the negative tails
will cause a depletion of the scattering at intermediate Y when a condensate is present.

In view of the discussion in Section 5 regarding the relationship between J(Y) and n(p\ it is
appropriate to examine how sensitive the observed scattering is to the theoretical n(p).
Inevitably there is a finite statistical accuracy attached to the experimental results, and a whole
range of different n(p)s may give equally good agreement with the data. If the statistical accuracy
of the results is high then only a limited range of n(p)s, all with very similar shapes, will be
consistent with the data. Alternatively, if the statistical accuracy is poor then the experimental
results will only place very weak constraints on the underlying shape of n(p).

Unfortunately, it is not possible to vary the condensate fraction in numerical calculations to see
what effect it has on the comparison with the experimental data. The numerical calculations use
as input only the known interatomic potential. Based on this potential, the full many body state
of the liquid is calculated. There are no parameters that one can adjust to make no larger or
smaller. The result of the simulation, just as in the real system, is determined solely by the
interaction between the atoms. Short of adjusting the interaction, which would undoubtedly
change the entire n(p), we cannot adjust no. We can, however, attempt to make a limited test of
the sensitivity of the experimental results to the value of the condensate fraction. The
momentum distribution can be decomposed into a condensate contribution and a non-
condensed contribution. We may adjust the relative weights of the condensate component and
the non-condensed component such that they still satisfy the normalization of n(p). However,
once we choose a value for the condensate different from the theoretical value the resultant
momentum distribution no longer corresponds to any ab initio calculation for helium. Thus,
while we are testing the sensitivity of the scattering to the value of no, it is only in the limited
range where n(p) has a very special shape for the non-condensed component.

With the above caveats in mind, we may replace the condensate dfunction with a Gaussian of
variable width and amplitude, keeping the shape of the non-condensate distribution the same as
the theoretical prediction. The best agreement is obtained when the width of the Gaussian is
less than 0.05 A"1 and no is 10%, in agreement with the best numerical predictions for the
condensate. Significant deviations are observed when the width is greater than 0.2 A"1 and no is
less than 8% or greater than 12%. For this particular model for the non-condensate part of n(p)
provided by the GFMC calculations, we find that there is indeed a condensate with no — 10 ±
2%. However, we note that changing the shape of the uncondensed n(p) would also have an
effect on the value for the condensate fraction.

In a similar fashion, the sensitivity to the expected singular behavior in n(p) at small p can be
examined. The GFMC results, which give excellent agreement with the observed scattering, do
not contain the expected singular contribution. (Because of finite-size effects, these results are
only valid for p > TC/L, where L is the system size.) The variational n(p), discussed previously,
explicitly includes this behavior. Both results are in excellant agreement with experiment. This is
not very surprising, since the weak singular behavior at small p is suppressed when n(p) is
transformed to J(Y), as discussed earlier. Thus, the predicted small p singular behavior makes
little contribution to the observed scattering and, with the experimental techniques now
available, will be difficult, if not impossible, to observe. Thus, the experimental results in the
superfluid provide a clear indication of a narrow component in n(p) containing approximately 9-
10% of the intensity, which is precisely that expected for the condensate. Unfortunately, due to
the finite statistical error inherent in any experiment, they cannot definitely prove the existence
of a condensate in the form of a <5-function. Some other singular behavior not associated with a
condensate could be responsible for the increase in the scattering at small p observed in the
superfluid. As seen in the comparison with the variational n(p), however, this would have to be
very singular

behavior, much more than the 1/p singularity, to agree with the experimental results. Thus,
while the experimental results cannot rule out a ground state n(p) which does not contain a S-
function condensate, they do provide strong evidence for a very narrow feature containing
10+2% of the total area. The excellent agreement with the numerical results suggests that this
very narrow feature is indeed due to the Bose condensate [3].

Similar comparisons have been carried out at a variety of temperatures and several different
densities in both the normal and superfluid phases. GFMC and variational calculations are used
for comparison with low temperature measurements and PIMC results are used for comparison
to measurements above 1 K. The agreement is excellent over the entire temperature range!
Theory and experiment appear to have converged for n(p) in liquid 4He at low densities (SVP).

Anda mungkin juga menyukai