Anal. Chem.

2001, 73, 4486-4490

Opt im ize d Ga t ing Sc he m e for Ra pid Life t im e

De t e rm ina t ions of Single -Ex pone nt ia l
Lum ine sc e nc e Life t im e s
Sing Po Chan, Z. J. Fuller, J. N. Demas,* , and B. A. DeGraff* ,

Department of Chemistry, University of Virginia, Charlottesville, Virginia 22904-4319 and Department of Chemistry,
James Madison University, Harrisonburg, Virginia 22807

The rapid lifetime method (RLD) for determining excitedstate lifetimes uses the ratio of the areas under two regions
of the decay. To get good precis ion with the s tandard
method, prior knowledge of the lifetime is es s ential to
s electing the integration regions . As will be s hown, the
us ual method of s electing integration regions is far from
optimal. An optimal gating s cheme that is more precis e
and much more forgiving in the s election of integration
region than any of the prior methods will be shown. Monte
Carlo s imulations were performed to determine the optimal gating. Experimental data was us ed to confirm the
capabilities of the optimized RLD. The s peed of the
optimal RLD is s imilar to the s tandard RLD but without
the neces s ity of matching the integration interval to the
lifetime for precis e res ults .
Time-resolved luminescence lifetime measurements have been
used to study various photophysical phenomena and for chemical
analysis. Because luminescence lifetimes can be very sensitive
to the immediate environment of the luminophore, they are widely
used to study the environment of molecular probes.1 These probes
can provide information that includes the rotational correlation
times, molecular separations, diffusion, and kinetics of excitedstate reactions (e.g., proton transfer, energy transfer). Lifetime
measurements in which the lifetime of the sensing molecule
depends on analyte concentration also have analytical applications.
While intensity measurements can sometimes serve to measure
these quantities, lifetime measurements are preferred because
they are less sensitive to instrumental changes and sample
properties. For example, the lifetime is largely independent of
excitation intensity or a small amount of sample decomposition
while luminescence intensity is not.
A common method of luminescence lifetime decay evaluation
is to acquire a multipoint decay curve and fit it by least-squares
methods. This approach requires considerable computational time
and is, therefore, unsuitable for real-time systems such as HPLC
and for imaging (e.g., microscopy, flames) where obtaining
multiple time points on the image is difficult.2 The rapid lifetime
determination method (RLD) is a family of evaluation schemes
used to quickly analyze the decay data and allow real-time data
*Corresponding authors. J.N.D.: (phone) 434-924-3343; (fax) 434-924-3710;
(e-mail) Demas@virginia.edu. B.A.D.: (phone) 540-568-6246; (fax) 540-568-7938;
(e-mail) degrafba@jmu.edu.
University of Virginia.
James Madison University.
(1) Lakowicz J. R. Principles of Fluorescence Spectroscopy; Plenum Press: New
York, 1983.
(2) Ni, T. Q.; Melton, L. A. Appl. Spectrosc. 1 9 9 3 , 47, 773-781.

4486 Analytical Chemistry, Vol. 73, No. 18, September 15, 2001

analysis of the lifetime.,3-5 For a single-exponential decay, the

lifetime is evaluated from the ratio of integrated areas under two
different regions of the decay curve. The two integrated areas
can be determined by the accumulated photon intensity or, in the
case of a gated CCD camera, directly from the pixel intensity. If
the two time intervals (gates) are equal, there is a simple closed
form solution for a single-exponential luminescence decay. An
extensive amount of analysis has been done to determine and
improve the precision of the RLD.6-9 While the RLD has been
traditionally used for single-exponential decays, it was recently
extended to the double-exponential case.10
The standard RLD belongs to a broad class of similar data
reduction methods, and we will give the general solution. We will
present an optimized RLD that is far more forgiving in terms of
the experimental conditions that must be used for good accuracy
and precision. Further, this optimized RLD has better precision
under a wide variety of conditions than the standard RLD has
under the best experimental conditions. Both experimental data
and theoretical Monte Carlo simulations will be used to show the
capabilities of the optimized RLD.
THEORY
The gating scheme and nomenclature for the standard RLD
with equal contiguous gating is shown in Figure 1a where the
Ds are areas under the indicated portion of the decay curve. For
a single-exponential decay, the lifetime is calculated from

-t
ln(D1/ D0)

(1)

In general, however, there is no reason the integration periods

must be equal. Also, as has been shown, the intervals do not have
to be contiguous.6,7 We show the general case in Figure 1b. The
first integral runs from 0 to t while the second starts at Yt and
(3) Ashworth, H. A.; Sternberg D.; Debiran, A., Seton Hall University, South
Orange, NJ, unpublished work, 1983.
(4) Wang, X. F.; Uchida, T.; Coleman, D. M.; Minami., S. Appl. Spectrosc. 1 9 9 1 ,
45, 360-366.
(5) Woods, R. J.; Scypinski, S.; Cline Love, L. J.; Ashworth, H. A. Anal. Chem.
1 9 8 4 , 56, 1395-1400.
(6) Waters, P. D.; Burns, D. H. Appl. Spectrosc. 1 9 9 2 , 47, 111-115.
(7) Tellinghuisen, J.; Wilkerson, C. W. Anal. Chem. 1 9 9 3 , 65, 1240-1246.
(8) Soper, S. A.; Legendre B. L., Jr. Appl. Spectrosc. 1 9 9 3 , 48, 400-405.
(9) Ballew, R. M.; Demas., J. N. Anal. Chem. 1 9 8 9 , 61, 30-33.
(10) Sharman, K. K.; Periasamy, A.; Ashworth, H. A.; Demas, J. N.; Snow N. H.
Anal. Chem. 1 9 9 9 , 71, 947-952.
10.1021/ac0102361 CCC: $20.00

2001 American Chemical Society

Published on Web 08/15/2001

Figure 1 . Graphs showing the integrated areas used for single-exponential decay with standard RLD using equal contiguous intervals (a) and
the generalized RLD (b).

runs to Yt + Pt. Y is the fractional delay in units of t and can

assume values from 0 to >1. P is the width of the second
integration period in units of t and can assume values greater
than 0. For Y ) P ) 1, we have the standard contiguous RLD
while for Y * 1 and P ) 1, we have the family of noncontiguous
delayed standard RLDs including those with overlapping gates.
The general expression is derived as follows:

D0 )
D0 )

A exp(-t/ ) dt

Yt+Pt

Yt

A exp(-t/ ) dt

(2)
(3)

where the decay is assumed to be given by A exp(-t/ ), A is the

preexponential factor, t is time, is the lifetime, and Y and P are
the adjustable coefficients for the integration intervals. The ratio
of D1 to D0 is then given by

D1 exp(-R(P + Y)) - exp(-YR)

)
D0
exp(-R) - 1

(4)

R ) t/

(5)

simulations. The integrated number of counts under the entire

decay curve was 106.
For our treatment, we assumed that both integrals were taken
from the same decay. This required special treatment for the
overlapping case (Y < 1). It was necessary to use three integrals.

S0 )
S1 )
S2 )

Yt

Yt

A exp(-t/ ) dt

(6)

A exp(-t/ ) dt

(7)

Yt+Pt

A exp(-t/ ) dt

(8)

For each simulation, Poisson noise was added to each integral.

For the overlapping case, D0 ) S 0 + S 1 and D1 ) S 1 + S 2. For
nonoverlapping intervals (Y g 1), we used eqs 2 and 3.
Reagents and Chemicals . The luminophore was aqueous 50
mM Na3[Tb(dpa) 3)] (dpa ) 2,6-pyridinedicarboxylic acid) at pH
7.11 The structure is shown below.

For contiguous and equal time intervals (P ) 1 and Y ) 1), eq 4

reduces to the standard contiguous RLD in eq 1. If, however, the
time intervals are not equal, P * 1, R cannot be isolated and
cannot be obtained in closed form. Nevertheless, R and then
can be calculated rapidly and efficiently by solving eq 4 using an
iterative Newton-Rhapson method.
EXPERIMENTAL SECTION
Monte Carlo Simulations. Monte Carlo simulations were first
performed to investigate different gating schemes for precision
and accuracy. These simulations were performed using programs
in Mathcad 8 Professional (Mathsoft, Inc., Cambridge, MA 02142).
In these simulations, artificial exponential decays were generated,
Poisson noise was added to the simulated experimental data, and
the lifetime was computed from the selected algorithm. The
process of adding noise and reducing the data was repeated 1000
times and the mean, standard deviation (), and relative standard
deviation (/ ) were computed. Simulations were repeated for
t/ values from 0.25 to 10 in increments of 0.25. The number of
simulations make it possible to produce very smooth, wellcharacterized curves for computed parameters for each set of

This complex was available from earlier work, has a long easily
measured luminescence lifetime (2.0 ms), is photochemically
(11) Metcalf, D. H.; Snyder, S. W.; Demas, J. N.; Richardson, F. S. J. Am. Chem.
Soc. 1 9 9 0 , 112, 5681-5695.

Analytical Chemistry, Vol. 73, No. 18, September 15, 2001

4487

Figure 2 . Percent standard deviation of lifetime at various t/ using Monte Carlo simulations. Six simulations are shown with varying integration
periods involving the optimized new methods and previously used methods. The first integral runs from 0 to t while the second starts at Yt
and runs to Yt + Pt.

and thermally very stable, and has a low temperature coefficient

of the decay time near room temperature (<0.01%C).12
Ins trumentation and Data Evaluation. A laser pulsed
lifetime method was used for acquiring all data. Samples were
excited with a 337-nm VSL337 20-kW peak power, 3-ns duration,
20 pps nitrogen laser (Laser Science, Inc. Franklin, MA). Cobalt
blue filters were placed in the laser excitation path to attenuate
the intensity. The sample emission was filtered with a combination
of yellow and orange filters placed in the emission path to reduce
scattered laser light. The luminescence signal was detected with
an R928 photomultiplier tube (PMT) in a refrigerated housing
with low dark current counts. A Stanford Research System SR430
(Palo Alto, CA) multichannel scaler/ averager was used as a photon
counter to obtain the decay waveforms. The signal was detected
by counting the individual single-photon pulses using the scalers
built-in discriminator and storing the counts in individual time bins
in the scaler. Acquisition was triggered by a PMT monitoring the
laser pulse. Multiple shots were used to build up enough counts
in each bin to give good signal-to-noise ratio. The bin width used
for acquisition was 2.56 s with 16 000 bins (total observation time
40 ms). The bin widths were small enough that the bin counts
can be used as intensities at individual times. Since the data were
obtained in very fine time intervals, the integration could be
carried out over any selected interval for testing the effect of
different t/ s. The integrals were simple sums over the appropriate range of bins. Photon pulse rates were kept low enough
to prevent pulse pile up. Typical peak counts were 11 000-13 000
counts/ bin with total counts under the decay of (8-12) 106.
Typically, 20 000 laser pulses were collected to generate a single
transient.
The interface software for communicating between the computer and the scaler was written in Labview 5.1.1 (National
(12) Glover-Fischer, D. Ph.D. Thesis, University of Virginia, August 1994.

4488 Analytical Chemistry, Vol. 73, No. 18, September 15, 2001

Instruments, Austin TX), which provided remote control of the

multichannel scaler and allowed transfer of the transients to the
computer for further analysis. Once the data were transferred, it
was then possible to apply different data reduction algorithms to
the same data set. The majority of the data analysis, done with
programs also written in Labview, was the summation of the data
to obtain the areas and performing the lifetime calculations.
A total of six decays were recorded. The average luminescence
lifetime calculated by the weighted nonlinear least-squares (NLLS)
method with a baseline (weighting factor equal to the reciprocal
of the number of counts in a bin) was taken as the actual lifetime
of [Tb(dpa) 3] 3-. The fitted baseline contribution was always below
3 counts/ bin and usually essentially zero. Decays were fit by NLLS
to an exponential plus a baseline. To avoid problems with zero
bin counts or Poisson rather than normal statistics, only the
portion of the decay with more than 50-60 counts/ bin (12 000
peak) was used. Using the same six data sets, the luminescent
lifetimes were calculated with the standard RLD and several sets
of conditions for the generalized RLD. Each evaluation method,
the mean lifetime, standard deviation, and percent relative
standard deviations were computed. To test the simulation and
its accuracy, the ts were varied from 0.25 to 7.75 with an
increment of 0.25, which was the same for the simulations. The
standard RLD was done as a comparison with the optimized RLD.
RESULTS AND DISCUSSION
We begin with the Monte Carlo calculations, which allowed
rapid assessment of the optimal conditions to use for reduction
of the real data. We then show that the experimental results
conform to the theory.
Monte Carlo Simulations . Some of the Monte Carlo simulation results are shown in Figure 2. Six simulations are shown
involving the optimized new methods and previous methods: (a)
standard contiguous RLD (P ) 1, Y ) 1); (b) standard RLD with

a 50%overlap (P ) 1, Y ) 0.5); (c) nonadjacent, nonoverlapping

RLD with a long second interval (P ) 12, Y ) 1.5); (d) 100%
overlapping with a long second integration period (P ) 12, Y )
0); (e) contiguous intervals with a long second integration period
(P ) 12, Y ) 1); (f) our optimized RLD (P ) 12, Y ) 0.25). For
the range of parameters tested, all methods gave accurate lifetimes
with no systematic bias above the noise level.
Except for being more precise for small t/ values (<2), the
nonoverlapping RLD standard methods (a, c, e) are inferior to all
others in terms of precision and tolerance to variation in t/ . Of
the nonoverlapping standard RLDs, the noncontiguous method
(c) gave the poorest results. The standard contiguous RLD (a)
gave the next poorest results with the best precision being
obtained in a rather narrow window around t/ ) 2 and good
performance only over the range 1.5-3; these results agree with
earlier results.9 Method e was comparable to the standard
contiguous RLD (a) for large t/ values and significantly more
forgiving for small t/ values (down to 0.5). The 100%overlap
(d) was indistinguishable from (e).
The very best performance was obtained with partial overlap.
The 50% overlap standard RLD (b) was very forgiving for large
t/ values but not quite as good as the standard RLD at small
values (<1.75). Far and away the best performance was obtained
with the optimized RLD (f) with P ) 12 and Y ) 0.25. This method
had the best precision of any method, equaled the best performance at small t/ values, and showed almost no degradation
in performance at t/ values up to 10.
Not shown is that the performance of the optimized RLD was
almost independent of P for values of 9 or greater. Values of P as
low as 4 showed only a small degradation in performance. Also,
the selection of Y was fairly critical. Performance degraded
noticeably for Y ) 0.1 or 0.35. Therefore, the optimum RLD should
use P g 9 and Y ) 0.25, at least for data dominated by photon
noise. Good performance should be expected over t/ range of
0.5 to >10, which makes it extremely insensitive to the choice
of t.
Experimental Res ults . Although detailed tests were not
carried out to test all of the predictions of the Monte Carlo
calculations, we did verify that the more important results were
correct.
The true luminescence lifetime for the [Tb(dpa) 3] 3- was
determined with the standard NLLS method. The luminescence
lifetime was 2.0363 ms with the standard deviation of 0.0010 ms
(0.05%relative standard deviation).
Figure 3 shows the relative standard deviation for both the
standard RLD and the optimized RLD (Y ) 0.25, P ) 12). As
predicted, the standard RLD gives its most precise results when
the t/ is at the optimum range of 1-2. An interval smaller or
larger than that range resulted in an increase in error in the
calculated lifetime with the effect being most severe for small
values. This result points out again that, for good precision with
the standard RLD, the lifetime of the luminophore must first be
known and the parameters adjusted to optimize the precision. We
can understand this phenomenon when we look at the integration
intervals. If they are too long, D1 will only include the tail of the
decay, which has a poor signal-to-noise ratio. At short intervals,
of course, all methods degrade since the signal does not decay
appreciably.

Figure 3 . Percent relative standard deviation of experimental

lifetimes at various t/ for the standard RLD (diamonds, solid line)
and the optimized RLD (Y ) 0.25, P ) 12) (squares, dashed line).
NLLS yielded a relative standard deviation of 0.05% for all six data
sets, and this value is shown as a horizontal dashed line.

Figure 3 clearly shows the advantages of the optimized RLD.

This method is superior to the standard RLD since the standard
deviation is much smaller even at higher t/ s. The calculated
lifetimes are within one standard deviation of the values by the
NLLS method, and the precision at the optimum is similar to that
of NLLS for t/ of 1-3. This shows that the optimized RLD can
be used to accurately calculate the single-exponential lifetimes of
systems even if little is known about the lifetime in advance, and
thus, the optimum t is unknown. The major advantages of the
optimized RLD method over the standard RLD are precision and
insensitivity to measurement conditions. It is much faster computationally than NLLS, and the instrumentation can be very
simple and inexpensive. For an RLD measurement, the PMT
signal can be split and sent to two gated integrators, which
accumulate the photon counts or current; these accumulated
intensities are then the integrals needed for eqs 4.
The Monte Carlo simulations with counts comparable to the
experimental data gave 0.05%relative standard deviation. This is
reasonable agreement with the data.
Another unexpected advantage of the optimized RLD over the
standard RLD is that the optimized RLD is largely insensitive to
PMT dark current. At larger values of t/ , the standard RLD
gave increasingly large systematic errors in if the dark current
was not corrected. For t/ ) 10, the error was 15%for a dark
current of 0.3-2 counts/ bin versus a peak signal of 13 000. For
the optimized RLD, there was no discernible difference between
the mean lifetime calculated with or without dark current corrections (<0.1%). The reason for this is that the overlapping RLD
has large counts in the second integral that are less influenced
by the small dark current, while for the standard RLD with large
t/ values, there is little but dark current that contributes to
the second integral.
We compare the accuracy of NLLS to the optimized RLD.
Except for the very lowest t/ values (<0.5), the optimized RLD
lifetimes are all within one standard deviation (the larger of
deviations for the RLD or of the NLLS) to the average for NLLS.
At its optimum region the optimized RLD actually has a somewhat
smaller deviation than NLLS. The optimized RLD appears on the
average to be 0.03%higher than NLLS. However, these discrepancies are inconsequential for most practical purposes. Further,
Analytical Chemistry, Vol. 73, No. 18, September 15, 2001

4489

the discrepancies are at such a low level that we cannot even be

sure which method provides the correct value. Very small
instrumental artifacts or multiple components with similar lifetimes
could systematically bias the different methods differently. What
is clear is that the new method is extremely precise, simple, and
accurate enough for most applications.
CONCLUSIONS
We have developed an optimized RLD for single-exponential
decays that utilizes an improved overlapping gating scheme. For
the optimized RLD, the conditions are 75%overlapping of the gates
and D1 having a width of 4-12 t/ .
The precision of the optimized RLD is superior to the standard
RLD for all integration periods. The precision of the optimized
RLD is fairly constant as t/ increases, making it more forgiving
if (1) the lifetimes are unknown, (2) multiple widely varying
lifetimes must be determined, and (3) the experiment is not readily
repeatable to optimize the conditions. Fluorescence lifetime

4490

Analytical Chemistry, Vol. 73, No. 18, September 15, 2001

detection in HPLC or CE would be examples for which the

optimized RLD would be clearly superior.
As with any two-area RLD, our optimized RLD gives no warning
of more complex decays. However, for many analytical applications, this is not a problem. One only needs a parameter that varies
monotonically with analyte concentration. For example, the
multiple decays in polymer-bound pH sensors can be reduced to
a single monotonically varying apparent lifetime.
ACKNOWLEDGMENT
The authors gratefully acknowledge the University of Virginia
and The National Science Foundation (grants CHE 0094777 and
9726999). We thank Dr. David Metcalf for loan of the Tb complex
and helpful discussions.
Received for review February 27, 2001. Accepted June 29,
2001.
AC0102361