IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 33, NO.
2, FEBRUARY 1997
Theory and Simulation on the Threshold
of Water Breakdown Induced by
Focused Ultrashort Laser Pulses
Q. Feng, J. V. Moloney, A. C. Newell, E. M. Wright, K. Cook, P. K. Kennedy,
D. X. Hammer, B. A. Rockwell, and C. R. Thompson
Abstract— A comprehensive model is developed for focused
pulse propagation in water. The model incorporates self-focusing,
group velocity dispersion, and laser-induced breakdown in which
an electron plasma is generated via cascade and multiphoton
ionization processes. The laser-induced breakdown is studied first
without considering self-focusing to give a breakdown threshold
of the light intensity, which compares favorably with existing
experimental results. The simple study also yields the threshold
dependence on pulse duration and input spot size, thus providing
a framework to view the results of numerical simulations of the
full model. The simulations establish the breakdown threshold in
input power and reveal qualitatively different behavior for pico-
and femto-second pulses. For longer pulses, the cascade process
provides the breakdown mechanism, while for shorter pulses
the cooperation between the self-focusing and the multiphoton
plasma generation dominates the breakdown threshold.
I. INTRODUCTION
T HE NONLINEAR optical properties of liquid water are
of considerable interest [1], [2] due to their implications
for pulse propagation in the human eye, the vitreous humor
being predominantly water. Water displays a nonlinear Kerr
effect [2], [3], and self-focusing (SF) can occur at megawatt
peak input powers. In addition to SF, water also displays laser-
induced breakdown (LIB) [4]–[6] in which an electron plasma
is generated due to the high intensity of focused laser pulses.
The electron density grows explosively and serves to absorb
and scatter the remaining pulse energy, thus shielding the
area beyond the focus from further radiation, i.e., the retina
[7]–[10]. Indeed, SF leading to LIB has been suggested as the
explanation for anomolies in the retinal damage data for visible
femtosecond pulses [11]–[13]. Nonlinear pulse propagation in
water is therefore of considerable interest in determining laser
safety standards in the ultrashort pulse regime.
Manuscript received April 8, 1996; revised September 9, 1996. The work
of Q. Feng, J. V. Moloney, E. W. Wright, and A. C. Newell was supported
by the Air Force Office of Scientific Research, Air Force Materiel Command,
USAF, under Grants F49620-94-1-0463 and F49620-94-1-0051. E. M. Wright
was supported in part by the Joint Services Optical Program.
Q. Feng, J. V. Moloney, and A. C. Newell are with the Arizona Center for
Mathematical Sciences, University of Arizona, Tucson, AZ 85721 USA.
E. M. Wright and K. Cook are with the Optical Sciences Center, University
of Arizona, Tucson, AZ 85721 USA.
P. K. Kennedy, D. X. Hammer, and B. A. Rockwell are with the Optical
Radiation Division, Armstrong Laboratory, Brooks AFB, San Antonio, TX
78235 USA.
C. R. Thompson is with Operational Technologies, Inc., Armstrong Labo-
ratory, Brooks AFB, San Antonio, TX 78235 USA.
Publisher Item Identifier S 0018-9197(97)00957-3.
0018–9197/97$10.00  1997 IEEE
127
In this paper, we report numerical results on linear and
nonlinear propagation in liquid water using 200-ps to 100-fs
duration pulses. We consider a geometry in which a Gaussian
input field is focused through a lens into the water sample,
thus simulating to first order the focusing properties of the
eye. The studies incorporate LIB, SF, multiphoton absorption
(MPA), and group velocity dispersion (GVD). The model is an
extended nonlinear Schrödinger equation (NLSE) coupled with
an equation describing the plasma generation. The NLSE is
widely used to describe light propagation and has been studied
extensively [14]. An interesting mechanism modeled by the
NLSE is self-focusing and beam collapse. When a focused or
unfocused propagating laser beam undergoes self-focusing, the
on-axis intensity gets enhanced considerably if the power of
the beam is near but below a critical value. When the power is
above the critical value, the collapse, namely an infinitely high
intensity, occurs at finite propagation distance. The collapse
is clearly unphysical and one would expect other nonlinear
effects, such as material breakdown, to come into play due to
the high intensity produced. Here, we extend the NLSE to in-
clude plasma generation, pulse–plasma interaction, and MPA.
The plasma generation is described by two processes: cascade
(or avalanche) ionization and multiphoton ionization. Included
also is the GVD, which has been shown to lead to pulse-
spliting and to arrest the collapse of subpicosecond pulses
[15]–[21]. Two focusing mechanisms exist in our studies:
linear focusing, due to the lens, and nonlinear self-focusing.
A framework for these studies is provided by first studying
LIB without SF, and this model compares favorably with
existing experimental results. We then establish a threshold
for LIB by numerical simulations. Our results show that SF is
a major player in understanding LIB in water for pulses of the
durations considered in this paper and of spot-sizes that can
be easily determined from a criterion given in this paper.
The remainder of this paper is organized as follows. In
Section II, we describe the model equations, the focusing
geometries and material parameters pertinent to experiments,
and write the equations into dimensionless form with ap-
propriate scalings. Although we will present our results in
subsequent sections in physical units to facilitate comparison
with experiments, a quick inspection of the dimensionless
equations leads to an interesting and important result, i.e.,
the relation of the breakdown threshold intensity to the pulse
duration for the case where multiphoton ionization is the
128
dominant breakdown mechanism. Section III is devoted to a
study of the equation for plasma generation. This equation is
solved using the electric field profile of the linearly focused
laser pulse. This simulates experimental situations where either
the pulses are focused into water samples placed at the focal
point of the lens or the focused pulses propagate within the
water when the input power is low, and thus SF and the plasma
generated do not affect the pulse propagation very much. This
allows us to easily extract useful information such as input spot
size dependence and relative roles played by the multiphoton
initiation and background electron density. More importantly,
solving this simple equation gives us a breakdown threshold
in the electric field intensity which agrees rather well with
the simulation results of the full model system for the whole
range of the pulse durations we studied. Section IV contains
the results of the numerical simulations using the fully coupled
system, and this is the core material of the paper. We give the
breakdown threshold in the peak input power and intensity,
address the problem of plasma shielding, and describe both
long and short pulse propagation and the strong SF behavior.
Some concluding remarks are given in Section V.
II. THEORETICAL MODEL
In this section, we describe our theoretical model for ul-
trashort pulse propagation in water and rescale the model
equations into a dimensionless form to facilitate both analytical
and numerical studies of the problem.
A. Model Equations
Our theoretical model for nonlinear propagation in water is
a generalization of the model previously studied by Feit and
Fleck [22] for LIB in gases. These authors solved a Drude
model for the electron density generated via LIB [5] coupled
to the paraxial wave equation for the electric field propagation.
Here, we generalize their model to include GVD, SF, and
MPA, all of which have been dealt with either individually or
in pairs previously [14]. Then assuming propagation along the
axis, the equation for the electric field envelope in
a reference frame moving at the group velocity is
(1)
where the terms on the right-hand side describe transverse
beam diffraction, GVD, nonlinear SF, absorption and de-
focusing due to the electron plasma, and MPA involving
photons, respectively. Here, is the optical frequency,
the intensity, , the quantity
controls the magnitude and sign of the GVD, with
corresponding to normal dispersion and to
anomolous dispersion, is the nonlinear coefficient such
that the nonlinear change in refractive index is
the cross section for electron–neutral inverse bremsstrahlung,
is the electron collision relaxation time, and is the
nonlinear coefficient for -photon absorption. The order of
the MPA is obtained from , which
IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 33, NO. 2, FEBRUARY 1997
is the minimum number of photons of energy needed to
overcome the ionization energy for liberating an electron
from a water molecule. In general, more than one order of
MPA may contribute, resulting in a sum over relevant ’s in
the last term of (1). For water, we have found that keeping
one order of MPA at a time using the above prescription is
sufficient for our simulations. To monitor whether the next
order is needed, we calculate which
is the intensity at which orders and yield the same
nonlinear absorption. When using order MPA, we ensure
that the intensity remains below .
In (1), we have retained only the instantaneous electronic
contribution to the Kerr nonlinearity. For pure water, there
is also a Raman active vibrational mode due to the O–H
stretching bond with frequency 3650 cm , which
produces a noninstantaneous nonlinear optical response and
stimulated Raman scattering (SRS). However, previous studies
of supercontinuum generation in water indicate that additives
reduce the noninstantaneous component in comparison to the
instantaneous one, with SRS absent in high-concentration
solution [23]. Since our primary motivation for studying
the present model is its application to ocular media, which
contain many additives, here we retain only the instantaneous
contribution to the Kerr nonlinearity, though the model is
easily extended to allow for SRS.
The dynamics of the electron plasma should strictly be
described by a Fokker–Planck equation for the electron energy
distribution function [24], [25]. This approach allows
for a complete description of the energy-dependent scattering
rates and diffusion of the distribution function as the plasma
evolves. However, the detailed scattering rates are not avail-
able for water as they are, e.g., for silicon [26], and we must
content ourselves here with a simpler model for the evolution
of the plasma. In particular, in the limit that the scattering rates
are approximated by the same constant value, and diffusion of
the distribution function is ignored, the remaining drift term
in the Fokker–Planck equation yields the Drude model for the
evolution of the plasma density . Here
we employ the Drude model to describe the evolution of the
electron plasma in water, and note that the model is more
appropriate for sub-picosecond pulses since it ignores diffusion
of the distribution functions in the energy domain. The model
equation which describes the electron density is then [5],
[27]–[29]
(2)
The first term on the right-hand-side of this equation describes
growth of the electron plasma by cascade (avalanche) ioniza-
tion, the second term is the contribution of MPA, which acts
both as a source for the cascade process and as a contributor
to plasma growth, and the third term describes the radiative
electron recombination. In addition, if there is a free-electron
density in the water prior to the pulse, it will also be amplified
by cascade ionization. Thus, we also include a background
initial density as an initial condition for (2).
In obtaining the density (2), we have neglected the effects
of plasma diffusion. This neglect is justified on the basis that
FENG et al.: THEORY AND SIMULATION ON THE THRESHOLD OF WATER BREAKDOWN 129

wavelength of 580 nm, for which the parameters

appropriate to water are 1.33, 4.1 10 cm /W
[3], 4, 1.55 10 m W , 6.5 eV,
10 s [28]. Our choice of this wavelength is motivated
by the fact that there are direct measurements of at this
wavelength, so that comparisons between experiments and
simulations will be more realistic. The order of the MPA
is for wavelengths from 570–760 nm, so this
value is pertinent to much of the visible range. The nonlinear
coefficient for MPA was obtained using the Keldysh formalism
for the multiphoton ionization rate [28], [31]. The cross section
for electron–neutral inverse bremsstrahlung may be calculated
using [22], giving
1.4 10 cm . The GVD for water was obtained using the
accurate Sellmeier formula for the refractive index given
by Maher [32], from which was formed, and
the GVD calculated. At the wavelength of 580 nm, this yields
normal GVD with a value 0.05 ps /m. We remark that
for infrared wavelengths beyond 1 m the GVD is anomolous,
Fig. 1. Focusing geometries considered in the text. (a) The incident laser
but we do not consider that case here. Our calculations show
pulse is focused from air through a lens with f= 1.7 cm directly into a water that the GVD is normal over the whole visible range and we
=
sample, where it comes to a linear focus at d 1.69 cm, and (b) the incident concentrate on that case. The critical power for the collapse,
laser pulse is focused from air through a lens with f = 1.7 cm, back into air
, in the absence of plasma generation is calculated using
and onto a thin water sample at the linear focus.
, which yields 1 MW for the above
parameters [3].
we are considering ultrashort pulses, picoseconds, and below, We have considered a specific focusing geometry with
and plasma diffusion is not operative on these short time 1.7 cm, this being roughly the focal length of the eye. Unless
scales, but rather occurs long after the pulses have left the otherwise stated, the input spot size was taken as 200
interaction region. In this respect, the theory developed here is m, which focuses at 1.69 cm with 11.8 m. The
for single pulses: if instead trains of pulses were considered, Rayleigh range of the focused beam is 0.1 cm.
then plasma diffusion could accumulate between pulses and
become significant. C. Dimensionless Equations
To facilitate analysis, it is often convenient to rescale the
B. Focusing Geometry and Material Parameters
model equations and write them in a dimensionless form. For
Fig. 1(a) shows the focusing geometry to be studied. We our system, we introduce the following scales:
are interested in the solutions of (1) and (2) for an initial
collimated Gaussian beam which enters the water sample
following a lens of focal length :

where is the prescribed breakdown electron density. The

(3) use of and as transverse and longitudinal length units,
respectively, is appropriate because the pulse intensity is high
where is the peak input power [ near the focal point ( ) where the plasma is expected
], the spot size, and is to be generated in a volume of . The dimensionless
the full temporal width at the points of the pulse intensity equations read
distribution. In what follows, we shall quote the values of
employed in the numerical simulations since it is the more
generally accessible experimental quantity. Under conditions (4)
of linear focusing, the input Gaussian beam comes to a focus
at a longitudinal position [30], where (5)
is the Rayleigh range (diffraction length) of
the input beam. The size of the laser beam at the focus where
is , with the
Rayleigh range of the focused beam.
Previous experiments involving LIB in water have been
conducted for both visible (514–625-nm range) wavelengths
and near infrared (1064 nm). For the numerical simulations
to be presented here, we consider the characteristic visible
130
The coefficients and represent three different effects
studied in some detail in this paper, namely, SF, cascade,
and multiphoton plasma generation. Insights regarding their
relative roles in the breakdown can be obtained from their
present forms.
The coefficient indicates the significance of SF in the
focusing geometry considered in this paper: SF cannot be ne-
glected unless the input power ( ) is low. The intensity
amplification due to SF has been studied in the framework of
NLSE [14], [33]. We have seen in our numerical simulations
that for an input power , compared to the linear
amplification which is independent of the input power, the total
amplification of the field intensity near the lens focal point is
considerably enhanced. It is well known that for slightly
above the critical value , without the lens, the collapse
occurs at very long propagation distance, while with the lens
the explosive intensity amplification occurs at a point near
the focus [14]. This is confirmed in our simulation when the
plasma generation and multiphoton absorption are turned off.
Since whether or not SF is significant is only determined by
the input power, we will choose to present our results in terms
of breakdown threshold in the input power. Also, when strong
SF is required for breakdown to occur, we would expect a
threshold due to the explosive nature of SF. We
will see this is indeed the case for subpicosecond pulses.
The coefficient to the cascade term can be rewritten as
where is the total energy
within the pulse. Thus, the maximum cascade ionization rate,
, is limited by the geometrical factor, . We will
see in the following sections that for long picosecond pulses,
SF is not significant at the breakdown threshold and that
cascade ionization dominates the process of plasma generation.
For subpicosecond pulses, however, cascade ionization plays
a secondary role. (This does not mean that the effect of
the cascade is negligible.) The multiphoton process, which is
the dominant breakdown mechanism for the shortest pulses,
requires high pulse intensity that is provided by the SF. An
interesting conclusion can be drawn from the coefficient by
inspecting (5). We see that when the multiphoton amplification
dominates the process of plasma generation, a breakdown
density (dimensionless ) is reached if . This
defines a breakdown threshold in the electric field intensity,
(6)
This gives the dependence of on the pulse duration:
. We will see later that the estimate of the intensity
threshold, (6), agrees well with the numerical results for short
pulses.
IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 33, NO. 2, FEBRUARY 1997
III. LIGHT-INDUCED BREAKDOWN WITH LINEAR FOCUSING
To provide a framework in which to view the numerical
simulations with SF and LIB, we first look at LIB with no
SF. This is closely realized experimentally by the focusing
geometry shown in Fig. 1(b) where the input beam is focused
by the lens through air onto a thin water sample placed at
the linear focus. Since the linear refractive indexes of air and
water are different, it is necessary to use an input spot size
for the focusing geometry in Fig. 1(b) to obtain the
same focal spot size 11.8 m, as in Fig. 1(a).
A. Plasma Generation Equation
In the absence of SF, we may simply solve (2) for the
density at 1.69 cm, which is the linear focus of the beam.
Restricting our attention to the region of highest intensity at
the beam center, , the field intensity may be written as
(7)
where is the peak intensity at focus. The corresponding
input power is then calculated using .
For simplicity, we neglect for now the recombination term
( ) in the plasma generation (2), since it plays a minor role
at the breakdown threshold. Then the equation has an exact
analytic solution, which can be written as
(8)
where
is the homogeneous part of the solution which arises
from the background electron density , and is the
inhomogeneous solution which grows from the electron den-
sity initiated by MPA. We remark that both the background
and multiphoton generated components of the total electron
density experience cascade ionization as described by the first
term in (2).
B. Breakdown Threshold
Since the analytic solution (8) is not very informative, we
have numerically solved the density equation (2) with the
intensity profile (7) to obtain the threshold for LIB. Here we
use as our criterion for LIB that 10
cm , where is the final density following the
pulse (here we neglect relaxation processes which reduce the
density well after the pulse has passed). This criterion for LIB
is selected to coincide with the appearence of vapor bubbles in
the water sample, also known as the bubble endpoint. Kennedy
et al. [28], [29] have previously compared a first-order model
for LIB based on this criterion with experimental data of the
intensity for 50% probability of bubble formation at pulse
durations of 2.4 ps, 400 fs, and 100 fs and have found
good agreement.
FENG et al.: THEORY AND SIMULATION ON THE THRESHOLD OF WATER BREAKDOWN 131

TABLE I
BREAKDOWN THRESHOLD INTENSITIES (W/cm2 )

1 Experimental thresholds for 50% probability of breakdown.

2 First-order calculation of the intensity breakdown threshold.
3 Intensity breakdown threshold from present work.

is not operative in the focusing geometry under consideration

[see Fig. 1(b)], the critical power is still a useful power unit
(a)
since Fig. 2(b) then gives us an idea of when SF will become
relevant at the breakdown threshold for the focusing geometry
in Fig. 1(a): when , we expect the thresholds
for both geometries to coincide. We shall return to this in the
next section.
Table I shows a comparison of the breakdown intensity
calculated here (fourth column) with the previous experimental
(second column) and first-order theoretical (third column)
results of Kennedy et al. [28] [29], and Hammer et al. [34],
with good overall agreement. The experiment of Hammer et al.
[34] was performed using a focusing geometry as in Fig. 1(b)
with a thin water sample and at a wavelength of 580 nm.
Comparison with the experimental data provides verification
of the first-order model and also of the results from the Drude
model contained in (2). This comparison gives validation of
(b) the Drude model as a reasonable model for ultrashort pulse
LIB and provides confidence that when we extend the model
Fig. 2. The breakdown intensity IBD versus pulse duration, and (b) the
breakdown input power PBD normalized to the critical power for collapse, to include SF using the focusing geometry in Fig. 1(a), which
=
Pcr 1 MW. The three curves correspond to initial background densities of is our primary interest, we are starting from a firm theoretical
b = =
0 (dotted line), b 1010 cm03 (solid line), and b = 1012 cm03 basis.
(dashed line).

C. Input Spot Size Dependence

The threshold intensity for LIB was determined for a
The results shown in Fig. 2(b) are for an input spot size
given pulse duration by varying the peak input intensity
200 m. This value was chosen, since for a focal
until the final density reaches 10 cm , thus
length 1.69 cm, the focused spot size is 23.6 m, which
giving the breakdown intensity . The breakdown power is
simulates the focusing properties of the eye to first-order and is
then calculated using . The breakdown
therefore of particular interest. However, it is also important to
intensity versus pulse duration is shown in Fig. 2(a), and
investigate the variation in breakdown power for more general
the breakdown power in Fig. 2(b), for the three backround
spot sizes. This is accomplished by using the breakdown
densities 0 cm (dotted line), 10 cm (solid line),
intensities from Fig. 2(a), which are independent of focusing
and 10 cm (dashed line). The threshold intensity and
conditions, and calculating the breakdown power using
power are seen to be insensitive to the background density
for pulse durations less than 10 ps, and the variation in the
predicted threshold is only a factor of two even for pulses as (9)
long as 200 ps. This shows that the precise initial density is
not a critical parameter in our simulations, and hereafter we The breakdown power versus input spot size is shown in Fig. 3
adopt the value 10 cm . Both the threshold intensity for pulse durations 200 ps (dash–dot line), 20 ps (dashed
and power increase with decreasing pulse length. This is to be line), 2 ps (dotted line), and 200 fs (solid line). In each case,
expected on physical grounds since for shorter pulses a higher one may compare with . If , where the
input power is required to supply enough energy to create the factor may be taken to be , then SF is relevant and
density required for breakdown in the focal volume. must be taken into account. From the figure, one sees that at
In Fig. 2(b), we have normalized the breakdown power to given , for in the interval , where
the critical power for collapse, 1 MW. Although SF and can be obtained by solving . The
132 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 33, NO. 2, FEBRUARY 1997

, MPA dominates the cascade initiation, and for

, the background density dominates the cascade initiation.
The transition in initiation mechanism is logical since short
pulses require higher intensities to attain breakdown, and
high intensities enhance the generation of electrons via MPA.
Experimentally, the transition to multiphoton initiation causes
the impurity dependence of the cascade breakdown threshold
to vanish at short pulsewidths. Both Docchio et al. [8] and
Kennedy et al. [28], [29] noted this phenomenon for visible
and near-infrared LIB at pulsewidths less than 100 ps.
It is clear that our estimate, 160 ps, depends on the
initial background density used, which was 10 cm .
For ultrapure water, with essentially no background electrons,
MPA is needed to initiate breakdown at all pulse durations. For
very impure water, on the other hand, the background density
Fig. 3. Variation of the breakdown power with input spot size w0 for f =
1.7 cm, b = 10 10 cm03 , and pulse durations p = 200 fs (solid line), 2 may dominate initiation for pulse durations shorter than the
ps (dotted), 20 ps (dashed), and 200 ps (dash–dotted). estimate above. However, even for very high background
densities, up to 10 cm , we find that MPA initiates
breakdown for pulse durations less than 10 ps, so that certainly
maximum breakdown power,
for subpicosecond pulses the background density plays no role.
For subpicosecond pulses, we have also calculated the pulse
(10)
duration for which cascade ionization plays only a minor role.
To do this, we solved the density equation for the breakdown
occurs when the focused spot size assumes its maximum
threshold both with and without the cascade ionization term
value for 48 m. For the spot size
[first term on the right-hand side of (2)]. Cascade ionization
200 m used here and a pulse duration of 200 ps,
was deemed to play only a minor role when the two thresholds
0.05 MW, so SF will not be relevant at threshold; whereas
were within a factor of two, and this occured only for pulse
0.8 MW at 48 m, thus SF will be very
durations less than 40 fs. Here we shall consider only pulse
important even at threshold.
durations greater than 100 fs, since a 40-fs pulse duration is
From the maximum breakdown power, we may infer the
well beyond the validity of the field envelope approximations
minimum pulse duration for which SF is irrelevant at
employed in deriving the field propagation (1). Thus, for all
the threshold for any input spot size. Let us deem that SF is
of our simulations, cascade ionization is a player of varying
irrelevant if or . Then
significance.
the minimum pulse duration can be determined as the
pulse duration for which . For the water
parameters quoted earlier, we have 200 ps. IV. NUMERICAL SIMULATIONS WITH SELF-FOCUSING
In this section, we describe our numerical simulations for
D. Multiphoton Initiation versus Background Electron Density LIB in water including SF. This corresponds to the experimen-
The calculations performed include components of the gen- tal focusing geometry in Fig. 1(a). For these simulations, the
erated electron density which are initiated both from the field envelope (1) was solved simultaneously with the density
background electron density and from MPA. Both of these (2) using the parameters given in Section II-B.
seed electron sources are amplified by cascade ionization [see
(8)]. It is therefore of interest to explore which of these is A. Numerical Method
the dominant seed electron source at intensities corresponding The simulations were performed using the beam propagation
to breakdown thresholds. To do this, we have calculated the method in which the propagation is broken into small linear
ratio of the background density component to the total and nonlinear steps which are performed sequentially [35].
density To propagate a small step in the spatial coordinate from
to , a half step is first taken, advancing the
(11) nonlinear terms in the field equation. A full step is then
taken to advance the linear terms, namely, the transverse
where we evaluate this expression after the pulse has passed Laplacian describing beam diffraction and the second-order
( ). With our definition of breakdown, the denominator time derivative describing GVD. The Laplacian is dealt with
corresponds to a density of 10 cm . By evaluating this ratio using a Crank–Nicholson scheme, and the GVD is handled
as a function of pulse duration, we find that for spectrally. Finally, a second half step is taken, advancing
160 ps, so that MPA and the backround components are the nonlinear terms. This process is continued to propagate
in exact balance. This value is further supported by the first- over any desired distance.
order model [28] which yields a pulse duration of 120 ps During the nonlinear steps, the plasma equation is solved
for the cross-over. For pulse durations less than the cross-over, using the trapezoidal rule. Due to linear and nonlinear self-
FENG et al.: THEORY AND SIMULATION ON THE THRESHOLD OF WATER BREAKDOWN 133

focusing, the pulse is compressed tremendously in the trans-

verse plane as it propagates in the medium, and this poses a
difficulty to our fixed grid scheme. Considerable advantage is
gained by making use of the fact that for or ,
in the initial stage of the propagation, when the nonlinear
effect is not important and the plasma has not been generated,
the propagation is described by the linear equation with only
diffraction and dispersion. The solution of this equation can
be written down exactly:

(12)

where

(a)

We use this linear equation to propagate the initial field into the
nonlinear interaction zone, which is then used as the input to
the nonlinear code. This represents a considerable reduction in
computation time, and also allows us to use the computational
grid more effectively, since the bulk of the linear focusing
by the lens is dealt with before passing over to the nonlinear
propagation code.

B. Breakdown Threshold
For the case including SF, we have determined the threshold
for breakdown, in particular the bubble endpoint, in the same
manner as before, namely, we seek the minimum input power
for which the peak final density becomes 10
cm . By the peak density, we specifically mean the maximum
density which is encountered anywhere in space in the
course of a simulation, rather than just at the linear focus. Self-
focusing may move the position of peak plasma generation
well away from the linear focus.
The results of our numerical simulations are summarized
in Fig. 4 where we show (a) the breakdown threshold of
peak input power ( ), normalized to the critical power for
collapse, and (b) the threshold intensity, both as functions of
pulse duration . The solid circles in Fig. 4(a) are our data
points corresponding to a peak plasma density of 10 cm .
The solid curve is the threshold power curve obtained in the
previous section without SF, and also shown in Fig. 2(b). We
see, therefore, that for pulses longer than 10 ps, the
thresholds with and without SF agree very well. This is to
be expected since for these pulse durations the breakdown
power is considerably below the critical power, , symbols for pulses shorter than 1 ps, so that the thresholds do
and SF plays only a minor role. In contrast, for 10 ps,
the breakdown power is seen to tend toward the critical power
. This is to be expected intuitively, since as the input power
approaches the critical power, small changes in the input power
can cause massive changes in the peak intensity at the laser
beam focus [14], which in turn can cause very large changes in
the peak plasma density generated, so that the plasma density
for breakdown is easily reached. This is particularly true of the
(b)
Fig. 4. LIB thresholds with SF: (a) Threshold power normalized to the
critical power for collapse, and (b) the threshold intensity, both as a function
of pulse duration p . In (a), the solid circles give the threshold for a peak
density of 1018 cm03 ; the triangles are the threshold powers obtained using
the criterion of a peak density of 1019 cm03 , the numbers above the triangles
are the % absorption of energy from the pulse using this density, and the solid
curve is the threshold power curve obtained without SF, as in Fig. 2(b). The
dashed line in (b) is the approximate threshold intensity given in (6), and the
solid line is the threshold intensity curve without SF, as in Fig. 2(a).
component of the density initiated by MPA since the source
term varies as [see (2)], with , and MPA is
expected to be the dominant initiation mechanism for plasma
generation for pulse durations less than 160 ps.
Since the criterion for LIB is somewhat subjective, we
have also determined the breakdown thresholds for a peak
density of 10 cm , and these data points are indicated by
the triangles in Fig. 4(a). Fortunately, the two criteria yield
the same threshold power to within the resolution of the
not depend too strongly on the precise definition used there.
Once again, this arises since SF enhancement of the pulse
intensity causes very large changes in the peak density due
to MPA. For pulses longer than 1 ps, the threshold power is
much higher for the larger threshold density [see the data for
20 ps in Fig. 4(a)]. The numbers above the triangles
are the percentage absorption of energy from the pulse using
the criterion of 10 cm peak plasma density, and the
134 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 33, NO. 2, FEBRUARY 1997

percentage absorption is consistently larger for the higher peak

density. However, the percentage absorption at LIB threshold,
seen in the experiments of Hammer et al. [34] using the bubble
endpoint, is typically a few percent, as opposed to the 12%
seen in Fig. 4(a) for the 20-ps pulse using the 10 cm
density criterion: We are therefore led to adopt the 10 cm
peak density criterion for the bubble endpoint.
The corresponding breakdown intensities, chosen as the
peak intensities which appeared during the respective numeri-
cal simulations, versus pulse duration, are shown in Fig. 4(b)
(solid circles) along with the result without SF from Fig. 2(a)
(solid line). Here, we see a remarkable agreement, especially (a)
for longer pulses, between the results with and without SF.
This agreement stems from the fact that LIB is triggered by
the local electric field strength, or intensity, so that LIB has an
intensity threshold [5] which is essentially independent of the
focusing conditions. In contrast, when we convert this intensity
threshold to a power threshold, due account must be taken of
the change in the beam profile due to linear focusing and/or
SF. The dashed line in Fig 4(b) is drawn using (6). It agrees
remarkably well with the numerical results for femtosecond
pulses. Therefore, the simple estimate (6) should be very useful
in assessing the breakdown intensity threshold of femtosecond
pulses.
(b)
From Fig. 4(a), we see that the breakdown power with
SF for 200 fs is ten times less than without Fig. 5. Peak plasma density (in units of 1018 cm03 ) and percentage of
energy absorption as functions of input power for p = 200 ps (solid line),
SF (solid line), whereas for 10 ps, the two power 100 ps (dotted), and 20 ps (dashed), with b = 1010 cm03 .
thresholds agree within a factor of two. Thus, we are led to
the conclusion that for pulse durations less than 10 ps and
for the input spot size used in this study, SF is a key player plasma via LIB, thus shielding the region beyond the plasma
in understanding LIB in water. This is the case even though from the full pulse energy. As discussed earlier, at the bubble
the breakdown powers do not exceed the critical power . endpoint threshold, only a few percent or less of the incident
The key point is that SF enhances the focal intensities leading energy is absorbed, so that the plasma shielding effect is weak.
to very large enhancements in the electron density due to In Fig. 5, we show the peak generated plasma density and
MPA. It is therefore plausible that SF is also a key element in corresponding percentage energy absorbed for pulse durations
understanding laser eye damage for femtosecond laser pulses 200 ps (solid line), 100 ps (dotted line), and 20 ps
and may well account for observed anomalies in the retinal (dashed line). In each case, the lowest power indicated is
damage data [11], [12]. However, without further elucidation slightly above the threshold power for LIB, with weak plasma
of the eye damage mechanisms, it is not possible to push this shielding in each case. However, far above the threshold, the
conclusion any further. peak plasma density and the plasma shielding both increase,
with 10%–25% absorption depending on the pulse length.
C. Multiphoton Initiation versus Background Density This clearly demonstrates that plasma shielding becomes more
In Section III-D, we showed that, in the absence of SF and significant above the bubble endpoint threshold, as expected
for 10 cm , the background density dominates the on physical grounds.
plasma initiation for pulses longer than 160 ps, and
MPA initiation dominates for pulses shorter than . With E. Long Pulse Limit
reference to Fig. 4(a), we see that for a 160-ps pulse, the The previous results have shown that SF is not significant at
power for breakdown is much less than the critical power, so threshold for pulses longer than 10 ps, and we discuss that limit
we expect that is the same with SF. We have verified this here. In Fig. 6, we show the maximum intensity,
numerically for a 200-ps pulse, , for which , and the maximum density,
the MPA term may be removed from (2) and the generated , as functions of propagation distance for
plasma density remains the same. In contrast, for a 100-ps 200 ps, and . This result is typical of
pulse, , if the MPA term is removed, the long-pulse behavior at threshold. Here, since both the input
generated plasma density is much lower. power and absorption are low, the focusing is mainly due
to the lens, and the pulse propagation is not affected by the
D. Plasma Shielding generated plasma. The field propagation is then well described
Plasma shielding arises when the incident pulse energy by the linear approximation (12) with only beam diffraction
is absorbed and converted to the generation of the electron and GVD, as we have verified numerically.
FENG et al.: THEORY AND SIMULATION ON THE THRESHOLD OF WATER BREAKDOWN
Fig. 6. On-axis maximum intensity maxt I (r = 0; z; t) 
jE j
maxt (r = 0; z; t) 2 and the maximum density maxt (r = 0; t; z ) as
functions of propagation distance z for p = 200 ps and P = 0:0472Pcr .
F. Short Pulse Limit
Fig. 7 shows the on-axis maximum intensity
and the plasma density as functions
of propagation distance for 200 fs and .
Compared to long pulses, the intensity at the focal point is
much higher due to self-focusing, even though the density
is the same (see Fig. 6). In Fig. 8, we show surface plots
of the electric field magnitude and the density
at a propagation distance 1.719 cm, which
lies beyond the linear focus at 1.69 cm. Here, we see
that past the focus the pulse starts to split temporally. This
pulse splitting of the propagating pulse has been studied and
is attributed to the effect of GVD, which is included in our
simulation [15]–[21]. The splitting is not symmetric in our
case, however, in comparison to most other treatments (see
also [36]), with the leading portion of the pulse being larger
than the trailing portion. This may be attributed to the plasma
generation which builds up during the leading edge of the pulse
but is larger and more effectively absorbing on the trailing
edge.
G. Strong Self-Focusing Behavior
An interesting aspect of our model is the behavior of nonlin-
ear propagation. Before the high-density plasma is generated,
the propagation is essentially described by the self-focusing
solution (the MPA in (4) is low because the coefficient
is small). However, once a plasma of density
is generated, it acts to absorb energy and defocus
the beam. The process of absorption and defocusing could
result in very rich nonlinear phenomena which have not
been studied in detail. Here we report a case of numerical
simulation that shows qualitatively different behavior from
what we have described in the previous section. We chose
a smaller input spot size, 120 m, so as to reduce
the effect of linear focusing (see Fig. 3). This allows us to
follow the nonlinear self-focusing, which is very explosive,
within the limit of our code. The process of pulse propagation
and plasma generation is presented in Figs. 9 and 10. Fig. 9
shows (top),
135
Fig. 7. On-axis maximum intensity maxt I (r = 0; z; t) and the plasma
density maxt (r = 0; z; t) as functions of propagation distance z for p =
200 fs and P = 0:928Pcr .
jE j
Fig. 8. Surface plots of (x; 0; z; t) (bottom) and (x; 0; z; t) (top), show-
ing pulse splitting and plasma generation for p = 200 fs and P = 0:928Pcr ,
at a propagation distance z = 1.719 cm.
(middle) and energy (bottom) as functions of propagation
distance . It is interesting to note that there are two peaks
in , one in front of and the other behind
the focal point . The spatial extension of the plasma
generated is wider than all the cases described in previous
sections. One can clearly see that the energy absorption occurs
in a region where the plasma is generated, while in other
regions of low plasma density, the energy is almost constant.
Fig. 10 shows (top) and (bottom) as functions of at
three propagation distances: (a) , (b) , and (c)
. The pulse starts to split in front of the focal point.
While the trailing split piece (the shorter piece corresponding
to higher plasma density) is damped once the splitting occurs,
the leading piece continues to focus and generates high plasma
density behind the focal point [see Fig. 10(c)]. Although here
the energy absorption is relatively higher, the propagating
pulse shows a tendency of not being absorbed. This suggests a
possible damaging mechanism: The leading split piece of the
pulse may escape the plasma generated in front of the focal
point and reach the retina, causing additional damage.
V. SUMMARY AND CONCLUSION
We have developed a comprehensive model for nonlinear
pulse propagation, which solves the nonlinear Schrödinger
136
(a)
(b)
(c)
Fig. 9. (a) On-axis maximum field maxt je(r = 0; z; t)j, (b) plasma density
maxt (r = 0; z; t), and (c) total pulse energy (in arbitratry units) as
functions of propagation distance z for p = 200 fs, w0 = 120 m, and
P = 1:08Pcr .
equation coupled to a rate equation describing plasma gen-
eration. The model has been used to perform numerical
simulations of nonlinear propagation in water, using visible
pulses from 200 to 100 fs and a geometry which simulates to
first-order the focal properties of the eye. Our primary goal in
this study was to examine how LIB and self-focusing interact
during ultrashort pulse propagation in the eye. To this end,
the simulations incorporated the effects of LIB, plasma–pulse
interaction, self-focusing, multiphoton absorption, and group
velocity dispersion.
A framework in which to view the numerical simulations
with self-focusing was provided by first studying LIB without
self-focusing. Our Drude model for LIB was shown to compare
favorably with existing experimental results under such linear
focusing conditions. Estimates of the breakdown threshold
intensity and the breakdown threshold power as a function of
pulse duration were next obtained for the case of LIB with self-
focusing, using numerical solution of the coupled equations for
pulse propagation and plasma generation. Comparison of LIB
thresholds to the critical power for self-focusing indicates that
self-focusing has little effect on LIB thresholds for long ps
pulses. The numerical simulations show, however, that self-
focusing is critical to LIB in water for pulsewidths less than
10 ps. These results indicate that self-focusing may play a role
IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 33, NO. 2, FEBRUARY 1997
(a)
(b)
(c)
Fig. 10. Surface plots of jE (x; 0; z; t)j (bottom) and (x; 0; z; t) (top) at
three propagation distances: (a) z = 1.687 cm, (b) z = 1.7 cm, and (c) z =
1.736 cm. The pulse parameters are the same as in Fig. 9.
in femtosecond–pulse ocular damage, by influencing whether
LIB and plasma–pulse interaction occur at the retina, in the
vitreous, or both.
Several other issues were explored during the study, such as
the dependence of LIB threshold on the background electron
density and on the input spot size. Finally, our numerical
simulations reveal that for femtosecond–pulse propagation,
where GVD may induce pulse splitting, additional interesting
nonlinear behavior results from the interplay of SF and LIB
with GVD. A solid physical groundwork is established for
future studies on nonlinear propagation in water. Such stud-
ies, in conjunction with histopathology of actual biological
exposures, may help determine which linear and nonlinear
mechanisms are responsible for ultrashort pulse damage to
the eye.
FENG et al.: THEORY AND SIMULATION ON THE THRESHOLD OF WATER BREAKDOWN 137

REFERENCES [28] P. K. Kennedy, “A first-order model for computation of laser-induced

breakdown thresholds in ocular and aqueous media: Part I—Theory,”
[1] A. Penzkofer, A. Laubereau, and W. Kaiser, “Stimulated short-wave IEEE J. Quantum Electron., vol. 31, pp. 2241-2249, 1995.
radiation due to single-frequency resonances of (3) ,” Phys. Rev. Lett., [29] P. K. Kennedy, S. A. Boppart, D. X. Hammer, B. A. Rockwell, G.
vol. 31, pp. 863–866, 1973. D. Noojin, and W. P. Roach, “A first-order model for computation of
[2] W. L. Smith, P. Lui, and N. Bloembergen, “Superbroadening in H2 O laser-induced breakdown thresholds in ocular and aqueous media: Part
and D2 O by self-focused picosecond pulses from a YAG:Nd laser,” II—Comparison to experiment,” IEEE J. Quantum Electron., vol. 31,
Phys. Rev. A, vol. 15, pp. 2396–2403, 1977. pp. 2250-2257, 1995.
[3] B. A. Rockwell, W. P. Roach, M. E. Rogers, M. W. Mayo, C. A. Toth, [30] P. W. Milonni and J. H. Eberly, Lasers. New York: Wiley, 1988, pp.
C. P. Cain, and G. D. Noojin, “Nonlinear refraction in vitreous humor,” 480–495.
Opt. Lett., vol. 18, pp. 1792–1794, 1993. [31] L. V. Keldysh, “Ionization in the field of a strong electromagnetic wave,”
[4] P. A. Barnes and K. E. Rieckhoff, “Laser-induced underwater sparks,” J. Exptl. Theoret. Phys., vol. 47, pp. 1945-1957, 1964; see also Sov. Phys.
Appl. Phys. Lett., vol. 13, pp. 282–284, 1968. JETP, vol. 20, pp. 1307-1314, 1965.
[5] E. Yablonovitch and N. Bloembergen, “Avalanche ionization and the [32] E. F. Maher, “Transmission and absorption coefficients for ocular
limiting diameter of filaments induced by light pulses in transparent media of the rhesus monkey,” Report SAM-TR-78-32, USAF School
media,” Phys. Rev. Lett., vol. 29, pp. 907–910, 1972. of Aerospace Medicine, Brooks AFB, TX, 1978.
[6] C. A. Sacchi, “Laser-induced breakdown in water,” J. Opt. Soc. Amer. [33] C. A. Klein, “Corrections for self-focusing in laser-damage experi-
B, vol. 8, pp. 337–345, 1991. ments,” J. Opt. Soc. Amer. B, vol. 10, pp. 2416–2419, 1993.
[7] C. A. Puliafito and R. F. Steinert, “Short-pulsed Nd:Yag laser mi- [34] D. X. Hammer, R. J. Thomas, G. D. Noojin, B. A. Rockwell, P. K.
crosurgery of the eye: Biophysical considerations,” IEEE J. Quantum Kennedy, and W. P. Roach, “Experimental investigation of ultrashort
Electron., vol. QE-20, pp. 1442–1448, 1984. pulse laser-induced breakdown thresholds in aqueous media,” IEEE J.
[8] F. Docchio, C. A. Sacchi, and J. Marshall, “Lifetimes of plasma induced Quantum Electron., vol. 32, pp. 670–678, 1996.
in liquids and ocular media by single Nd:YAG laser pulses of different [35] J. A. Fleck, J. R. Morris, and M. D. Feit, “Time-dependent propagation
duration,” Lasers in Opthalmology, vol. 1, pp. 83–93, 1986. of high-energy laser beams through the atmosphere,” Appl. Phys., vol.
[9] B. Zysset, J. G. Fujimoto, and T. F. Deutsch, “Time-resolved measure- 10, pp. 129–160, 1976.
ments of picosecond optical breakdown,” Appl. Phys. B, vol. 48, pp. [36] J. E. Rothenberg, “Space-time focusing: Breakdown of the slowly-
139–147, 1989. varying envelope approximation in the self-focusing of femtosecond
[10] S. A. Boppart, C. A. Toth, W. P. Roach, and B. A. Rockwell, “Shielding pulses,” Opt. Lett., vol. 17, pp. 1340–1342, 1992.
effectiveness of femtosecond laser-induced plasmas in ultrapure water,”
in Laser-Tissue Interaction IV, S. L. Jaques, Ed. Proc. SPIE, vol. 1882,
pp. 347–354, 1993.
[11] R. Birngruber, C. A. Puliafito, A. Gawande, W. Lin, R. W. Schoenlein,
and J. G. Fujimoto, “Femtosecond laser-tissue interactions: Retinal Q. Feng, photograph and biography not available at the time of publication.
injury studies,” IEEE J. Quantum Electron., vol. QE-23, pp. 1836–1844,
1987.
[12] C. P. Cain, C. D. DiCarlo, B. A. Rockwell, P. K. Kennedy, W.
P. Roach, G. D. Noojin, D. J. Stolarski, D. X. Hammer, and C.
D. Toth, “Retinal damage and laser-induced breakdown produced by J. V. Moloney, photograph and biography not available at the time of
ultrashort-pulse lasers,” Graefe Archives for Experimental and Clinical publication.
Ophthamology, to be published, 1996.
[13] J. A. Powell, J. V. Moloney, A. C. Newell, and R. A. Albanese, “Beam
collapse as an explanation for anomalous ocular damage,” J. Opt. Soc.
Amer. B, vol. 10, pp. 1230–1241, 1993.
[14] J. H. Marburger, “Self-focusing: theory,” Prog. Quantum Electron., vol. A. C. Newell, photograph and biography not available at the time of
4, pp. 35–110, 1975. publication.
[15] N. A. Zharova, A. G. Litvak, T. A. Petrova, A. M. Sergeev, and A. D.
Yunakovskii, “Multiple fractionation of wave structures in a nonlinear
medium,” Pis’ma Zh. Eksp. Teor. Fiz., vol. 44, pp. 12-15, 1986; see also
JETP Lett., vol. 44, pp. 13-17, 1986.
[16] D. Strickland and P. B. Corkum, “Short pulse self-focusing,” in Proc. E. M. Wright, photograph and biography not available at the time of
Soc. Photo-Opt. Instrum. Eng., 1991, vol. 1413, pp. 54–58. publication.
[17] P. Chernev and V. Petrov, “Self-focusing of light pulses in the presence
of normal group velocity dispersion,” Opt. Lett., vol. 17, pp. 172–174,
1992.
[18] J. E. Rothenberg, “Pulse splitting during self-focusing in normally
dispersive media,” Opt. Lett., vol. 17, pp. 583–585, 1992. K. Cook, photograph and biography not available at the time of publication.
[19] G. G. Luther, J. V. Moloney, A. C. Newell, and E. M. Wright, “Self-
focusing threshold in normally dispersive media,” Opt. Lett., vol. 19,
pp. 862-864, 1994.
[20] G. G. Luther, A. C. Newell, and J. V. Moloney, “The effect of normal
dispersion on collapse events,” Physica D, vol. 74, pp. 59–73, 1994.
[21] G. Fibich, “Self-focusing in the nonlinear Schrödinger equation for P. K. Kennedy, photograph and biography not available at the time of
ultrashort laser-tissue interactions,” Ph.D dissertation, Courant Institute, publication.
New York University, New York, 1994.
[22] M. D. Feit and J. A. Fleck, Jr., “Effect of refraction on spot-size
dependence of laser-induced breakdown,” Appl. Phys. Lett., vol. 24, pp.
169–172, 1974.
[23] T. Jimbo, V. L. Caplan, Q. X. Li, Q. Z. Wang, P. P. Ho, and R. R. D. X. Hammer, photograph and biography not available at the time of
Alfano, “Enhancement of ultrafast supercontinuum generation in water publication.
+
by addition of Zn2 and K+ cations,” Opt. Lett., vol. 12, pp. 477–479,
1987.
[24] Y. B. Zel’Dovich and Yu. P. Raizer, “Cascade ionization of a gas by a
light pulse,” J. Exptl. Theoret. Phys., vol. 47, pp. 1150–1161, 1964; see
also Sov. Phys. JETP, vol. 20, pp. 772-780, 1965. B. A. Rockwell, photograph and biography not available at the time of
[25] L. H. Holway, “Temporal behavior of electron distributions in electric publication.
fields,” Phys. Rev. Lett., vol. 28, pp. 280–283, 1972.
[26] B. C. Stuart, M. D. Feit, A. M. Rubenchik, B. W. Shore, and M. D. Perry,
“Laser-induced damage in dielectrics with nanosecond to picosecond
pulses,” Phys. Rev. Lett., vol. 74, pp. 2248-2251, 1995.
[27] Y. R. Shen, The Principles of Nonlinear Optics. New York: Wiley, C. R. Thompson, photograph and biography not available at the time of
1984, pp. 528–540. publication.