PERSPECTIVES

To create tight binders (dissociation constant < 50 nM) from their initial hits (apparent affinities of >2000 nM and >5000 nM),
Fleishman et al. performed affinity maturation
with yeast display. An examination of the stabilizing mutations suggested ways of improving their computational methods; for instance,
the authors concluded that future modeling
efforts should try to take into account subtle
movements of sequence backbones, attractive forces known as long-range electrostatics, and the energy costs associated with protein interactions (such as the desolvation cost
for burying polar atoms). Optimizing a protein energy function, however, presents a challenge, and the insights gained from this single
study will need to be combined with results
from other design and modeling studies in
order to identify robust improvements.
Although Fleishman et al. have produced
a landmark result, it is evident that computational protein interface design is not a solved
problem. Researchers should not be satisfied
with one or a few successes in solving these
astronomically complex molecular puzzles.
Each new puzzle is different from the last; for
example, the region of hemagglutinin targeted
in this work was hydrophobic and -helical.
Will the computational protocol developed by

Fleishman et al. also be effective for designing binders for polar surface patches, or targeting alternative secondary structures (see
the figure) such as sheets, strands, or loops?
Creating many of these interfaces will require
accurate modeling of protein conformations
and accurate evaluation of desolvation, electrostatics, and hydrogen bonding.
The endeavor to understand protein interactions will undoubtedly continue for decades
to come. And the pursuit should remain persistent, as the impacts of rational design and
manipulation of protein-protein interactions
can hardly be overstated.
References
1. S. J. Fleishman et al., Science 332, 816 (2011).
2. D. J. Mandell, T. Kortemme, Nat. Chem. Biol. 5, 797 (2009).
3. J. Karanicolas, B. Kuhlman, Curr. Opin. Struct. Biol. 19,
458 (2009).
4. A. Leaver-Fay et al., Methods Enzymol. 487, 545 (2011).
5. S. S. Sidhu, S. Koide, Curr. Opin. Struct. Biol. 17, 481
(2007).
6. I. A. Wilson, N. J. Cox, Annu. Rev. Immunol. 8, 737 (1990).
7. D. C. Ekiert et al., Science 324, 246 (2009).
8. J. Sui et al., Nat. Struct. Mol. Biol. 16, 265 (2009).
9. P.-S. Huang et al., Protein Sci. 16, 2770 (2007).
10. R. K. Jha et al., J. Mol. Biol. 400, 257 (2010).
11. J. Karanicolas et al., Mol. Cell 42, 250 (2011).
12. S. A. Gai, K. D. Wittrup, Curr. Opin. Struct. Biol. 17, 467
(2007).
13. S. Liu et al., Proc. Natl. Acad. Sci. U.S.A. 104, 5330 (2007).
10.1126/science.1207082

PHYSICS

Brownian Motion Goes Ballistic

Peter N. Pusey

hen Einstein explained the origin of Brownian motion in 1905,

he described the erratic movement of a microscopic particle driven by the
thermal motion of liquid molecules as a random walk with sharp changes of direction
between each step (1). He realized that this
picturethe one we seem to see if we watch
a particle under the microscopemust break
down if we were to look more closely. A moving object would require an infinite force to
change its speed or direction discontinuously.
The particle actually moves ballistically
along a smooth trajectory (24), as if it were a
microscopic ocean liner on an erratic course
(see the first figure, panels A and B). It has
taken more than a century to observe this ballistic motion. The studies of Li et al. (5) were
conducted on particles in air; its low viscosity
allowed ballistic motion to be followed accuSUPA, School of Physics, University of Edinburgh, Edinburgh EH9 3JZ, UK. E-mail: p.n.pusey@ed.ac.uk

802

rately for extended periods and showed that

a particles instantaneous velocities along its
path obey a statistical distribution consistent
with thermal motion. Huang et al. (6) used
liquid water; in this higher-density medium,
the transition from ballistic motion at short
times to diffusive motion at longer ones could
be studied in detail.
Einstein estimated (24) that a particle of
diameter 1 m in water would move for only
about 0.1 s over a distance of only about 2
before completely changing its speed and
direction. The minute magnitudes of these
estimates led Einstein to conclude that only
the larger-scale diffusive random walk would
be observed in practice. The present experiments used technologies undreamed of in
Einsteins time. Single particles were held
in optical trapsradiation pressure prevents
them from settling under gravity. A fast position-sensitive detector (7) split the interference pattern formed between light scattered
by the particle and the trapping laser beam

Measurements of the Brownian motion of

particles in air and in water reveal a smooth
ballistic motion at very short times.

itself and fed the two parts to two photodiode detectors. Lateral motion of a particle in
the trap increases the intensity at one detector
but decreases it at the other. The difference
between the two signals measures one component of the particles position. Coupled to
fast electronics, this system can measure displacements as small as 0.3 in time intervals
as short as 0.01 s.
In a vacuum, a particle in an optical trap
would oscillate indefinitely. Gas molecules
both dampen the oscillations and introduce random impulses that induce Brownian motion. Li et al. determined the time
evolution of the position and velocity of a
3-m-diameter silica sphere trapped in air
at about 1/36 of atmospheric pressure (see
the second figure, panel A). The underlying
oscillatory motion, with a period of about
300 s, is evident, as is a degree of randomness induced by the gas molecules. Einstein (24) estimated the duration of ballistic Brownian motion to be the time p = m/

13 MAY 2011 VOL 332 SCIENCE www.sciencemag.org

Published by AAAS

Downloaded from www.sciencemag.org on January 8, 2013

tion, avoiding gene-cloning steps, and investigators can obtain initial measures of binding affinity while the protein is displayed on
the yeast surface. This obviates the need to
individually purify 73 different proteins. And
after biophysical characterization of the two
binders, the yeast are then used for a selection
process called affinity maturation. The effectiveness of this experimental workflow may
pave the way for future efforts to test computationally designed protein interactions.
Another possible reason for their achievement involves the computational design protocol. Previous modest-affinity binders were
developed by focusing first on the placement of the two proteins, and then on highresolution all-atom side chain design (9, 10).
Fleishman et al. took a different approach.
They first focused on all-atom placement of
disembodied side chains to establish critical
hot spot interactions, and then on docking
the designed protein to its target. This strategy is reminiscent of a previous approach that
involved grafting key residues from a known
interaction onto a new protein scaffold to
generate a new binding pair (13). If there are
no known hot spot binding motifs, however,
these hot spot interactions must be designed
de novo (1).

PERSPECTIVES

www.sciencemag.org SCIENCE VOL 332 13 MAY 2011

Published by AAAS

Downloaded from www.sciencemag.org on January 8, 2013

Mean-square displacement (nm2)

Velocity (mm/s)

Position (nm)

over which the particles velocity

A 80
B
Diffusion
would decay through friction with
Trapping
10
x2(t) = 2Dt
40
the gas (or liquid), if the random
impulses from its molecules were
0
1
neglected; here m is the particles
mass and is its Stokes friction
40
coefficient that describes drag (8,
0.1
80
9). For the experiment of Li et
al., p 150 s, so that random
1
changes in the velocity occurred
0.01
on a time scale similar to that
0
Ballistic motion
of the coherent oscillations (the
x2(t) = v2t2
0.001
motion is underdamped).
1
Li et al. constructed histograms
p
of the instantaneous velocities at
0
1
2
0.01
0.1
1
10
100
1000
air pressures of both 1/36 and 1
Time
(ms)
Time (s)
atm, where the particles velocity
changed more rapidly (p 50 s). Resolving ballistic motion. (A) Li et al. measured the position and velocity of a 3-m-diameter silica sphere optically
These were accurately described trapped in air at about 1/36 of atmospheric pressure. Oscillations in the underdamped motion were randomly perturbed
by the Maxwell-Boltzmann dis- by collisions with the gas molecules. (B) Huang et al. measured the mean-square displacement of a 1-m-diameter silica
tribution with a root-mean-square sphere trapped in water, which displayed the slow transition from ballistic to diffusive Brownian motion. Note the remark(rms) velocity of about 0.42 mm/s, able resolution of the measurement: 0.001 nm2 corresponds to a displacement of 0.3 , measured in about 0.02 s.
in good agreement with the theo- [Panel B adapted from (6)]
retical value from equipartition of
energy for conditions of thermal equilibrium a detailed examination of the transition from
What next? Li et al. mention the fascinat[vrms = (kBT/m), where kB is the Boltzmann ballistic to diffusive motion. Huang et al. ing prospect of laser cooling a trapped particonstant and T the temperature]. Li et al. also measured the MSD of a 1-m-diameter sil- cle to a temperature at which quantization of
showed that the particles mean-square dis- ica sphere in water (11) (see the second fig- the energy of this mesoscopic object could be
placement (MSD, a measure of average dis- ure, panel B). At the shortest times (t 0.01 observed (16). Huang et al. suggest extending
tance traveled) agreed well with old calcula- to 0.03 s), the MSD increases as t2 (a line of their measurements to Brownian motion in
tions of underdamped Brownian motion in a slope 2 in the double logarithmic plot), which confined regions and heterogeneous media.
harmonic potential (10).
corresponds to ballistic motion. Between Here, understanding the details of predifIn a liquid, the duration of a Brownian roughly 2 s < t < 20 s, the particle makes fusive motion over subnanometer distances
step, p 0.1 s, is much smaller than the many random steps, and the MSD increases could well be relevant to some biological pronatural oscillation period of the trap, making as t, which is characteristic of diffusion (11). cesses, such as the lock-and-key mechanism
the motion strongly overdamped. Although it The particles free diffusion is finally limited of enzyme action.
is more difficult to access the ballistic regime by the trapping potential, causing the MSD
References and Notes
itself, experimental studies in liquids enable to saturate at t > 100 s.
1. A. Einstein, Ann. Phys. 322, 549 (1905).
2. A. Einstein, Zeit. Elektrochem. 13, 41 (1907).
For many years after Einsteins contri3. Both (1) and (2) are translated and reprinted in (4).
butions, it was expected that the transition
4. A. Einstein, in Investigations on the Theory of Brownian
A
B
from ballistic to diffusive motion would be
Movement, R. Frth, Ed. (Dover, New York, 1956).
5. T. Li et al., Science 328, 1673 (2010).
quite sharp, corresponding to an exponen6. R. Huang et al., Nat. Phys. (2011); 10.1038/
tial decay of the particles memory of its earnphys1953:arxiv:1003.1980.
lier velocity. However, about 50 years ago,
7. I. Chavez et al., Rev. Sci. Instrum. 79, 105104 (2008).
8. The Stokes friction coefficient is given by = 6R, with
~L
hints from computer simulations and theory
R the particles radius and the shear viscosity of the
started to suggest a more complex scenario.
liquid. At low pressure, where the mean-free path of a
In particular, hydrodynamic vortices in the
gas molecule is comparable to R, the friction is reduced
by a substantial Cunningham slip correction (9).
liquid created by the particles motion lead
9. E. Cunningham, Proc. R. Soc. A 83, 357 (1910).
to memory effects, and the particles veloc- 10. G. E. Uhlenbeck, L. S. Ornstein, Phys. Rev. 36, 823 (1930).
ity decays much more slowly than exponen- 11. In an isotropic random process like Brownian motion, the
Timing is everything. (A and B) A particle undergoaverage velocity and average displacement are zero. The
tially, exhibiting a t3/2 long-time-tail (12).
ing Brownian motion is buffeted at random by thersimplest descriptor of average motion is the MSD, the
mally agitated fluid molecules, and it follows a torDetailed analysis by Huang et al. of data
square of x(t), the distance moved in time t, averaged
tuous but locally smooth path (black line). It loses
like that shown in the second figure, panel
over many measurements. For ballistic motion (straight
memory of its speed and direction over the very small
line and constant speed) x2(t) = v2t2; for diffusion,
B, where the ballistic-to-diffusive transition
Einsteins result is x2(t) = 2Dt, with the diffusion condistance ~L. What an observer actually sees depends
spans more than three decades in time, has
stant given by D = kBT/ (1).
on the resolution of the position measurement. In the
now provided a thorough verification of the 12. B. J. Alder, T. E. Wainwright, Phys. Rev. A 1, 18 (1970).
case considered by Einstein (A), the particles speed
full, complicated hydrodynamic theory (13, 13. E. J. Hinch, J. Fluid Mech. 72, 499 (1975).
and direction have changed many times between
14. H. J. H. Clercx, P. P. J. M. Schram, Phys. Rev. A 46, 1942
14). Although several previous experiments
(1992).
measurements (the red dots), and the observer sees
15. B. Lukic et al., Phys. Rev. Lett. 95, 160601 (2005).
a diffusive random walk with uncorrelated steps (the had observed the breakdown of the simple
16. T. Li et al., Nat. Phys. (2011); 10.1038/nphys1952.
blue arrows). Measurements made at shorter time diffusion picture [e.g., (15)], the present stud10.1126/science.1192222
ies extend into the ballistic regime.
intervals (B) resolve the smooth ballistic motion.

803

REPORTS

30.
31.
32.
33.
34.
35.
36.

Consequently, the heat gradient and thermal stress are

uniform for the tilt angles recorded in this study.
P. Poncharal, Z. L. Wang, D. Ugarte, W. A. de Heer,
Science 283, 1513 (1999).
L. Meirovich, Elements of Vibration Analysis (McGraw-Hill,
New York, ed. 2, 1986).
X.-L. Wei, Y. Liu, Q. Chen, M.-S. Wang, L.-M. Peng,
Adv. Funct. Mater. 18, 1555 (2008).
M. M. J. Treacy, T. W. Ebbesen, J. M. Gibson, Nature 381,
678 (1996).
G. V. Hartland, Annu. Rev. Phys. Chem. 57, 403
(2006).
T. Nakamiya et al., Thin Solid Films 517, 3854
(2009).
C. Ni, P. R. Bandaru, Carbon 47, 2898 (2009).

Measurement of the Instantaneous

Velocity of a Brownian Particle
Tongcang Li, Simon Kheifets, David Medellin, Mark G. Raizen*
Brownian motion of particles affects many branches of science. We report on the Brownian motion
of micrometer-sized beads of glass held in air by an optical tweezer, over a wide range of pressures,
and we measured the instantaneous velocity of a Brownian particle. Our results provide direct
verification of the energy equipartition theorem for a Brownian particle. For short times, the
ballistic regime of Brownian motion was observed, in contrast to the usual diffusive regime. We
discuss the applications of these methods toward cooling the center-of-mass motion of a bead in
vacuum to the quantum ground motional state.

n 1907, Albert Einstein published a paper in

which he considered the instantaneous velocity of a Brownian particle (1, 2). By measuring this quantity, one could prove that the
kinetic energy of the motion of the centre of gravity of a particle is independent of the size and
nature of the particle and independent of the
nature of its environment. This is one of the
basic tenets of statistical mechanics, known as
the equipartition theorem. However, because of
the very rapid randomization of the motion,
Einstein concluded that the instantaneous velocity of a Brownian particle would be impossible to
measure in practice.
We report here on the measurement of the
instantaneous velocity of a Brownian particle in a
system consisting of a single, micrometer-sized
SiO2 bead held in a dual-beam optical tweezer in
air, over a wide range of pressures. The velocity
data were used to verify the Maxwell-Boltzmann
velocity distribution and the equipartition theorem
for a Brownian particle. The ability to measure
instantaneous velocity enables new fundamental
tests of statistical mechanics of Brownian particles and is also a necessary step toward the cooling of a particle to the quantum ground motional
state in vacuum.
The earliest quantitative studies of Brownian
motion were focused on measuring velocities,
and they generated enormous controversy (3, 4).
Center for Nonlinear Dynamics and Department of Physics,
University of Texas at Austin, Austin, TX 78712, USA.

The measured velocities of Brownian particles

(3) were almost 1000-fold smaller than what was
predicted by the energy equipartition theorem.
Recent experiments with fast detectors that studied
Brownian motion in liquid (57) and gaseous
(810) environments observed nondiffusive motion of a Brownian particle.
Einsteins theory predicts that [Dx(t)]2
2Dt, where [Dx(t)]2 is the mean square displacement (MSD) in one dimension of a free Brownian particle during time t, and D is the diffusion
constant (11). The diffusion constant can be calculated by D kB T =g, where kB is Boltzmanns constant, T is the temperature, and g is
the Stokes friction coefficient. The mean velocity
measured over
an p
interval
of time t is v
p
p

[Dx(t)]2 /t 2D/ t . This diverges as t apFig. 1. Simplified schematic

showing the counterpropagating dual-beam optical
tweezers, and a novel detection system that has a 75-MHz
bandwidth and ultralow noise.
The s-polarized beam is reflected by a polarizing beamsplitter cube after it passes
through a trapped bead inside
a vacuum chamber. For detection, it is split by a mirror with
a sharp edge. The p-polarized
beam passes through the cube.

37. S. Jonic, C. Vnien-Bryan, Curr. Opin. Pharmacol. 9, 636

(2009).
38. Supported by NSF (grant DMR-0964886) and Air Force
Office of Scientific Research (grant FA9550-07-1-0484)
in the Physical Biology Center for Ultrafast Science and
Technology supported by Gordon and Betty Moore
Foundation at Caltech. A patent application has been
filed by Caltech based on the methodology presented
herein.

Supporting Online Material

www.sciencemag.org/cgi/content/full/328/5986/1668/DC1
Movies S1 to S3
5 April 2010; accepted 19 May 2010
10.1126/science.1190470

proaches 0 and therefore does not represent the

real velocity of the particle (1, 2).
The equation [Dx(t)]2 2Dt, however, is
valid only when t >> tp ; that is, in the diffusive
regime. Here, tp m=g is the momentum relaxation time of a particle with mass m. At very short
time scales (t << tp ), the dynamics of a particle are
dominated by its inertia, and the motion is ballistic.
The dynamics of a Brownian particle over all time
scales can be described by a Langevin equation
(12). The MSD of a Brownian particle at very
short time scales is predicted to be [Dx(t)]2
(kB T /m)t2 , and its instantaneous velocity can be
measured as v Dxt=t, when t << tp (13).
For a 1-mm-diameter silica (SiO2) sphere in
water, tp is about 0.1 ms and the root mean square
(rms) velocity is about 2 mm/s in one dimension.
To measure the instantaneous velocity with 10%
uncertainty, one would require 2-pm spatial resolution in 10 ns, far beyond what is experimentally achievable today (7). Because of the lower
viscosity of gas, compared with liquid, the tp of a
particle in air is much larger. This lowers the
technical demand for both temporal and spatial
resolution. The main difficulty of performing highprecision measurements of a Brownian particle in
air, however, is that the particle will fall under the
influence of gravity. We overcome this problem
by using optical tweezers to simultaneously trap
and monitor a silica bead in air and vacuum, allowing long-duration, ultrahigh-resolution measurements of its motion.

Downloaded from www.sciencemag.org on January 8, 2013

27. F. Banhart, J. Mater. Sci. 41, 4505 (2006).

28. V. H. Crespi, N. G. Chopra, M. L. Cohen, A. Zettl,
S. G. Louie, Phys. Rev. B 54, 5927 (1996).
29. The extent of displacement may vary depending on the tilt
angle, because the local thickness of the specimen along
the axis of the excitation may change. However, the skin
depth of the MWNT ring specimen for the 532-nm light is
deduced to be 2 mm [absorption coefficient a = 1.0 104
cm1 (35)], which exceeds the largest local thickness along
the ring specimen at a tilt angle of 35. In addition, the
absorption cross section of MWNTs is reported to be weakly
dependent on the polarization of the incident beam for
thick tubes (36). To further suppress any polarization
dependence, we set the polarization of the optical excitation
beam so that it was not along the long axis of the tube.

Vacuum
Chamber

s-polarized

p-polarized

s-polarized
Detector

*To whom correspondence should be addressed. E-mail:

raizen@physics.utexas.edu

www.sciencemag.org

SCIENCE

VOL 328

25 JUNE 2010

1673

For small displacements, the effect of optical

tweezers on the beads motion can be approximated by a harmonic potential. The MSD of a
Brownian particle in an underdamped harmonic trap in air can be obtained by solving the
Langevin equation (14)
[Dx(t)]2
$
"
#%
2kB T
sin w1 t
t=2tp
1

e
cos
w
t

1
2w1 tp
mw20
1
where w0 is the resonant frequency of the trap
q
and w1 w20 1/(2tp )2 . The normalized velocity autocorrelation function (VACF) of the particle is (14)

"
#
sin w1 t
y(t) et=2tp cos w1 t
2w1 tp

In the simplified scheme of our optical trap

and vacuum chamber (Fig. 1), the trap is formed
inside a vacuum chamber by two counterpropagating laser beams focused to the same point by
two identical aspheric lenses with focal lengths
of 3.1 mm and numerical apertures of 0.68 (15).
The two 1064-nm-wavelength laser beams are
orthogonally polarized, and their frequencies differ by 160 MHz to avoid interference. The scattering forces exerted on the bead by the two
beams cancel, and the gradient forces near the
center of the focus create a three-dimensional
harmonic potential for the bead. When the bead
deviates from the center of the trap, it deflects
both trapping beams. The position of the bead is

Fig. 2. One-dimensional trajectories of a 3-mm-diameter silica bead trapped in air at 99.8 kPa (A) and
2.75 kPa (B). The instantaneous velocities of the bead corresponding to these trajectories are shown in (C)
and (D).

monitored by measuring the deflection of one of

the beams, which is split by a mirror with a sharp
edge. The difference between the two halves is
measured by a fast balanced detector (7, 16).
The lifetime of a bead in our trap in air is much
longer than our measurement times over a wide
range of pressures and trap strengths. We have
tested it by trapping a 4.7-mm bead in air continuously for 46 hours, during which the power
of both laser beams was repeatedly changed from
5 mW to 2.0 W. The trap becomes less stable in
vacuum. The lowest pressure at which we have
trapped a bead without extra stabilization is about
0.1 Pa.
For studying the Brownian motion of a trapped
bead, unless otherwise stated, the powers of the
two laser beams were 10.7 and 14.1 mW (15), the
diameter of the bead was 3 mm, the temperature
of the system was 297 K, and the air pressure was
99.8 or 2.75 kPa. The trapping was stable and the
heating due to laser absorption was negligible under these conditions. In typical samples of position
and velocity traces of a trapped bead (Fig. 2), the
position traces of the bead at these two pressures
appear to be very similar. On the other hand, the
velocity traces are clearly different. The instantaneous velocity of the bead at 99.8 kPa changes
more frequently than that at 2.75 kPa, because
the momentum relaxation time is shorter at higher
pressure.
Figure 3 shows the MSDs of a 3-mm silica
bead as a function of time. The measured MSDs
fit with Eq. 1 over three decades of time for both
pressures. The calibration factor a = position/
voltage of the detection system is the only fitting parameter of Eq. 1 for each pressure. tp and
w0 are obtained from the measured normalized
VACF. The two values of a obtained for these
two pressures differ by 10.8%. This is because
the vacuum chamber is distorted slightly when
the pressure is decreased from 99.8 to 2.75 kPa.
The measured MSDs are completely different
from those predicted by Einsteins theory of
Brownian motion in a diffusive regime. The

Fig. 3. (A) The MSDs of a

A
B
3-mm silica bead trapped
in air at 99.8 kPa (red
square) and 2.75 kPa
(black circle). They are
calculated from 40 million position measurements for each pressure.
The noise signal (blue
triangle) is recorded when
there is no particle in the
optical trap. The solid lines
are theoretical predictions
of Eq. 1. The prediction of
Einsteins theory of free
Brownian motion in the
diffusive regime is shown
in dashed lines for comparison. (B) MSDs at short
time scales are shown in detail. The dash-dotted line indicates ballistic Brownian motion of a free particle.

1674

25 JUNE 2010

VOL 328

SCIENCE

www.sciencemag.org

Downloaded from www.sciencemag.org on January 8, 2013

REPORTS

REPORTS

slopes of measured MSD curves at short time

scales are double those of the MSD curves of
diffusive Brownian motion in the log-log plot
(Fig. 3A). This is because the MSD is proportional to t 2 for ballistic Brownian motion, whereas
it is proportional to t for diffusive Brownian motion. In addition, the MSD curves are independent of air pressure at short time scales, which
is predicted by Dxt&2 kB T =mt2 for ballistic Brownian motion, whereas the MSD in the
diffusive regime does depend on the air pressure.
At long time scales, the MSD saturates at a constant value because of the optical trap. Figure 3B
displays more detail of the Brownian motion at
short time scales. It clearly demonstrates that we
have observed ballistic Brownian motion.
The distributions of the measured instantaneous velocities (Fig. 4A) agree very well with
the Maxwell-Boltzmann distribution. The measured rms velocities are vrms = 0.422 mm/s at
99.8 kPa and vrms = 0.425 mm/s at 2.75 kPa.
These values are very close to the prediction
of the
p
energy equipartition theorem, vrms kB T /m,
which is 0.429 mm/s. As expected, the velocity
distribution is independent of pressure. The rms
value of the noise signal is 0.021 mm/s, which
means we have 1.0 spatial resolution in 5 ms.
This measurement noise is about 4.8% of the rms
velocity. Figure 4A represents direct verification
of the Maxwell-Boltzmann distribution of velocities and the equipartition theorem of energy for
Brownian motion. For a Brownian particle in
liquid, the inertial effects of the liquid become important. The measured
rms velocity of the particle
p
will be vrms kB T/m* in the ballistic regime,
where the effective mass m* is the sum of the mass
of the particle and half of the mass of the displaced
fluid (17). This is different from the equipartition theorem. To measure the true instantaneous
velocity in liquid as predicted by the equipartition theorem, the temporal resolution must be
much shorter than the time scale of acoustic
damping, which is about 1 ns for a 1-mm particle
in liquid (17).
Figure 4B shows the normalized VACF of
the bead at two different pressures. At 2.75 kPa,

one can see the oscillations due to the optical

trap. Equation 2 is independent of the calibration
factor a of the detection system. The only independent variable is time t, which we can measure with high precision. Thus the normalized
VACF provides an accurate method to measure
tp and w0. By fitting the normalized VACF with
Eq. 2, we obtained tp = 48.5 T 0.1 ms, w0 = 2p
(3064 T 4) Hz at 99.8 kPa and tp = 147.3 T
0.1 ms, w0 = 2p (3168 T 0.5) Hz at 2.75 kPa.
The trapping frequency changed by 3% because
of the distortion of the vacuum chamber at different pressures. For a particle at a certain pressure and temperature, tp should be independent
of the trapping frequency. We verified this by
changing the total power of the two laser beams
from 25 to 220 mW. The measured tp changed
less than 1.3% for both pressures, thus proving
that the fitting method is accurate, and the heating due to the laser beams (which would change
the viscosity and affect tp ) is negligible. We can
also calculate the diameter of the silica bead
from the tp value at 99.8 kPa (18). The obtained
diameter is 2.79 mm. This is within the uncertainty range given by the supplier of the 3.0-mm
silica beads. We used this value in the calculation of MSD and normalized VACF.
The ability to measure the instantaneous velocity of a Brownian particle will be invaluable
in studying nonequilibrium statistical mechanics
(19, 20) and can be used to cool Brownian motion by applying a feedback force with a direction
opposite to the velocity (21, 22). In a vacuum,
our optically trapped particle should be an ideal
system for investigating quantum effects in a
mechanical system (16, 2325) because of its
near-perfect isolation from the thermal environment. Combining feedback cooling and cavity
cooling, we expect to cool the Brownian motion
of a bead starting from room temperature to the
quantum regime, as predicted by recent theoretical calculations (24, 25). We have directly verified the energy equipartition theorem of Brownian
motion. However, we also expect to observe deviation from this theorem when the bead is cooled
to the quantum regime. The kinetic energy of the

www.sciencemag.org

SCIENCE

VOL 328

bead will not approach zero even at 0 K because

of its zero-point energy. The rotational energy of
the bead should also become quantized.
References and Notes

1. A. Einstein, Zeit. f. Elektrochemie 13, 41 (1907).

2. A. Einstein, Investigations on the Theory of the Brownian
Movement, R. Frth, Ed., A. D. Cowper, Transl. (Methuen,
London, 1926), pp. 6367.
3. F. M. Exner, Ann. Phys. 2, 843 (1900).
4. M. Kerker, J. Chem. Educ. 51, 764 (1974).
5. B. Luki et al., Phys. Rev. Lett. 95, 160601 (2005).
6. Y. Han et al., Science 314, 626 (2006).
7. I. Chavez, R. Huang, K. Henderson, E.-L. Florin,
M. G. Raizen, Rev. Sci. Instrum. 79, 105104 (2008).
8. P. D. Fedele, Y. W. Kim, Phys. Rev. Lett. 44, 691 (1980).
9. J. Blum et al., Phys. Rev. Lett. 97, 230601 (2006).
10. D. R. Burnham, P. J. Reece, D. McGloin, Brownian dynamics
of optically trapped liquid aerosols. In press; preprint
available at http://arxiv.org/abs/0907.4582.
11. A. Einstein, Ann. Phys. 17, 549 (1905).
12. P. Langevin, C. R. Acad. Sci. (Paris) 146, 530 (1908).
13. G. E. Uhlenbeck, L. S. Ornstein, Phys. Rev. 36, 823 (1930).
14. M. C. Wang, G. E. Uhlenbeck, Rev. Mod. Phys. 17, 323 (1945).
15. Materials and methods are available as supporting
material on Science online.
16. K. G. Libbrecht, E. D. Black, Phys. Lett. A 321, 99 (2004).
17. R. Zwanzig, M. Bixon, J. Fluid Mech. 69, 21 (1975).
18. A. Moshfegh, M. Shams, G. Ahmadi, R. Ebrahimi,
Colloids Surf. A Physicochem. Eng. Asp. 345, 112 (2009).
19. R. Kubo, Science 233, 330 (1986).
20. G. M. Wang, E. M. Sevick, E. Mittag, D. J. Searles,
D. J. Evans, Phys. Rev. Lett. 89, 050601 (2002).
21. A. Hopkins, K. Jacobs, S. Habib, K. Schwab, Phys. Rev. B
68, 235328 (2003).
22. D. Kleckner, D. Bouwmeester, Nature 444, 75 (2006).
23. A. Ashkin, J. M. Dziedzic, Appl. Phys. Lett. 28, 333 (1976).
24. D. E. Chang et al., Proc. Natl. Acad. Sci. U.S.A. 107,
1005 (2010).
25. O. Romero-Isart, M. L. Juan, R. Quidant, J. Ignacio Cirac,
N. J. Phys. 12, 033015 (2010).
26. M.G.R. acknowledges support from the Sid W. Richardson
Foundation and the R. A. Welch Foundation grant
number F-1258. D.M. acknowledges support from
El Consejo Nacional de Ciencia y Tecnologa (CONACYT)
for his graduate fellowship (206429). The authors would
also like to thank E.-L. Florin and Z. Yin for helpful
discussions and I. Popov for his help with the experiment.

Downloaded from www.sciencemag.org on January 8, 2013

Fig. 4. (A) The distribu- A

B
tion of the measured instantaneous velocities of
a 3-mm silica bead. The
statistics at each pressure
is calculated from 4 million instantaneous velocities. The solid lines are
Maxwell-Boltzmann distributions. We obtained
vrms = 0.422 mm/s at
99.8 kPa (red square)
and vrms = 0.425 mm/s
at 2.75 kPa (black circle)
from the measurements.
The rms value of the noise
(blue triangle) is 0.021
mm/s. (B) The normalized
velocity autocorrelation functions of the 3-mm bead at two different pressures. The solid lines are fittings with Eq. 2.

Supporting Online Material

www.sciencemag.org/cgi/content/full/science.1189403/DC1
Materials and Methods
10 March 2010; accepted 10 May 2010
Published online 20 May 2010;
10.1126/science.1189403
Include this information when citing this paper.

25 JUNE 2010

1675

www.sciencemag.org/cgi/content/full/science.1189403/DC1

Supporting Online Material for

Measurement of the Instantaneous Velocity of a Brownian Particle
Tongcang Li, Simon Kheifets, David Medellin, Mark G. Raizen*
*To whom correspondence should be addressed. E-mail: raizen@physics.utexas.edu
Published 20 May 2010 on Science Express
DOI: 10.1126/science.1189403
This PDF file includes:
Materials and Methods

Measurement of the Instantaneous Velocity of a Brownian Particle

Tongcang Li, Simon Kheifets, David Medellin, and Mark G. Raizen
Supporting Online Material
Methods:
The two laser beams are aligned with the help of a pinhole aperture whose diameter is
1.00.5 m. We first align the pinhole to the focus of the s-polarized beam. We then align the
focus of the p-polarized beam to the pinhole. Because the pinhole has a finite thickness (13 m),
it is difficult to align the foci of the two beams to the same point in the axial direction. We
intentionally make the waist of one beam larger than the other to make this alignment less critical.
The measured waists of the two beams are 2.2 m and 3.0 m in the horizontal direction. The
real waists should be smaller than these values due to the finite size of the pinhole. Also note that
the waists of the two beams are measured at different axial positions separated by the thickness
of the pinhole. Once a bead is trapped, we keep the power of one beam constant, and tune the
power of the other beam to maximize the trapping frequency.
The silica beads are initially held above the trap on the surface of a glass slide. We launch
single beads into air using pulsed ultrasonic vibrations at 340 kHz. The launched beads fall under
gravity, and eventually a bead approaches the laser focus and is trapped.
The bandwidth of our balanced detector is 75MHz. The position signal of a trapped bead is
recorded at a sampling rate of 2 MHz. Because of the limited spatial resolution, we are not able
to obtain accurate instantaneous velocities of a bead at this rate. To reduce the noise, we average
every 10 successive position measurements, and use these averages to calculate instantaneous
velocities with time resolution of 5 s. Although this method reduces the temporal resolution by
a factor of 10, it increases the signal-to-noise ratio if both the trapping period ( 2 /

) and

momentum relaxation time are much larger than 5 s. These conditions are satisfied here since
the trapping period is about 320 s,

= 48 s at 99.8 kPa, and

= 147 s at 2.75 kPa.