Characterization of a cross-reactive electronic nose with vapoluminescent array elements

Article abstract of DOI:10.1021/ac011255d

A three-channel cross-reactive sensor array based on vapoluminescent platinum(II) double salt materials has been characterized. Two arrays were studied, one consisting of [Pt(CN-cyclododecyl)₄][Pt(CN)₄] (1), [(phen)Pt(CN-cyclohexyl)₂][Pt(CN)₄] (2), and [Pt(CN-n-tetradecyl)₄][Pt(CN)₄] (3) materials, where phen = 1,10-phenanthroline, and a second array that has compound 3 replaced by the mixed double salt material [(phen)Pt(CN-cyclododecyl)Cl)]₂[(phen)Pt (CN-cyclododecyl)₂]₂[Pt(CN)₄]₃ (4). Compounds 2, 3 and 4 are characterized here for the first time. Inclusion of solvent vapors into these materials often leads to dramatic shifts in their solid-state absorption and luminescence spectra. In these studies the arrays were exposed to a set of 10 test solvent vapors to determine the ability of each cross-reactive array to give reproducible vapoluminescent spectra characteristic of each solvent vapor. It was discovered that temperature programming between solvent vapor exposures greatly improved the reproducibility of the luminescence spectra obtained. A statistical analysis of three-dimensional resolution factors between pairs of solvent clusters in principal component space supported this assertion. The success of the temperature programming protocol was limited by the thermal stability and the sensitivity to low background water vapor levels of some platinum(II) double salt materials. The ability of the cross-reactive sensor array to differentiate between two different solvent vapors over a large concentration range was also investigated. Acetone and methanol were found to occupy two distinct regions of the three-dimensional principal component space. Detection limits for acetone and methanol were estimated from the principal component analysis as 75 and 6 g/m³, respectively.

Full text of DOI:10.1021/ac011255d

Anal. Chem. 2002, 74, 2547-2555
Characterization of a Cross-Reactive Electronic
Nose with Vapoluminescent Array Elements
Steven M. Drew,† Daron E. Janzen, and Kent R. Mann*
Department of Chemistry, University of Minnesota, Minneapolis, Minnesota 55455
tion system, capable of recognizing simple or complex odors.”2
The development and characterization of new cross-reactive
transducer materials is critical if electronic nose sensors are to
be applied in practical analytical measurements. Many types of
transducer materials for cross-reactive arrays have been investi-
gated including the following: semiconducting metal oxides,1b,3
conducting polymer films,4 acoustic wave devices,5 field-effect
transistors,6 carbon black-loaded polymer film chemoresistors,7
and conductometric sensors based on electrolyte/ polymer com-
posites.8 Optical transducers also show great promise as cross-
reactive sensor arrays. Some examples include the immobilization
of dyes such as Nile Red9 and metalloporphyrins10 on chemically
modified porous silica microspheres or surface-modified photo-
luminescent semiconductor materials.11
Our previous studies12 suggested that pure vapochromic and
vapoluminescent compounds13 could be used in the solid state as
optical transducers. Vapochromism in crystalline [Pt(CNR)4][Pt-
(CN)4] salts arises from highly anisotropic packing forces that
enable solvent vapors to reversibly penetrate the interior of the
material to form a new crystalline phase with precisely determined
solvent-chromophore interactions.12 These include changes in
the dielectric constant near the chromophore, hydrogen bonding
A three-channel cross-reactive sensor array based on
vapoluminescent platinum(II) double salt materials has
been characterized. Two arrays were studied, one consist-
ing of [P t(CN-cyclododecyl)4][P t(CN)4] (1), [(phen)P t(CN-
cyclohexyl)2][P t(CN)4] (2), and [P t(CN-n -tetradecyl)4][P t-
(CN)4] (3 ) materials, where phen ) 1 ,1 0 -phenanthroline,
and a second array that has compound 3 replaced by the
mixed double salt material [(phen)P t(CN-cyclododecyl)-
Cl)]2 [(phen)P t(CN-cyclododecyl)2 ]2 [P t(CN)4 ]3 (4 ). Com-
pounds 2 , 3 and 4 are characterized here for the first
time. Inclusion of solvent vapors into these materials often
leads to dramatic shifts in their solid-state absorption and
luminescence spectra. In these studies the arrays were
exposed to a set of 1 0 test solvent vapors to determine
the ability of each cross-reactive array to give reproducible
vapoluminescent spectra characteristic of each solvent
vapor. It was discovered that temperature programming
between solvent vapor exposures greatly improved the
reproducibility of the luminescence spectra obtained. A
statistical analysis of three-dimensional resolution factors
between pairs of solvent clusters in principal component
space supported this assertion. The success of the tem-
perature programming protocol was limited by the thermal
stability and the sensitivity to low background water vapor
levels of some platinum(II) double salt materials. The
ability of the cross-reactive sensor array to differentiate
between two different solvent vapors over a large concen-
tration range was also investigated. Acetone and methanol
were found to occupy two distinct regions of the three-
dimensional principal component space. Detection limits
for acetone and methanol were estimated from the prin-
cipal component analysis as 7 5 and 6 g/ m3 , respectively.
(2) Gardner, J. W.; Bartlett, P. N. Sens. Actuators, B 1 9 9 4 , 18-19, 211.
(3) Gardner, J. W.; Pike, A.; de Rooij, N. F.; Koudelka-Hep, M.; Clerc, P. A.;
Hierlemann, A. Sens. Actuators, B 1 9 9 5 , 26, 135-139.
(4) (a) Bartlett, P. N.; Archer, P. B. M.; Ling-Chung, S. K. Sens. Actuators 1 9 8 9 ,
19, 125-140. (b) Pearce, T. C.; Gardner, J. W.; Friel, S.; Bartlett, P. N.;
Blair, N. Analyst 1 9 9 3 , 118, 371-377.
(5) (a) Grate, J. W. Chem. Rev. 2 0 0 0 , 100, 2627-2648. (b) Grate, J. W.; Zellers,
E. T. Anal. Chem. 2 0 0 0 , 72, 2861-2868.
(6) Lundstrom, I.; Hedborg, E.; Spetz, A.; Sundgren, H.; Winquist, F. Electronic
Nose Based on Field Effect Structures. In Sensors and Sensory Systems for
an Electronic Nose; Gardner, J. W., Bartlett, P. N., Eds.; NATO ASI Series
212; Kluwer: Dordrecht, The Netherlands, 1992; pp 303-319.
(7) (a) Lonergan, M. C.; Severin, E. J.; Doleman, B. J.; Beader, S. A.; Grubbs,
R. H.; Lewis, N. S. Chem. Mater. 1 9 9 6 , 8, 2298-2312. (b) Sotzing, G. A.;
Briglin, S. M.; Grubbs, R. H.; Lewis, N. S. Anal. Chem. 2 0 0 0 , 72, 3181-
3190. (c) Doleman, B. J.; Lewis, N. S. Sens. Actuators, B 2 0 0 1 , 72, 41-50.
(8) (a) Miller, L. L.; Boyd, D. C.; Schmidt, A. J.; Nitzkowski, S. C.; Rigaut, S.
Chem. Mater. 2 0 0 1 , 13, 9-11. (b) Miller, L. L.; Bankers, J. S.; Schmidt, A.
J.; Boyd, D. C. J. Phys. Org. Chem. 2 0 0 0 , 13, 808-815.
(9) (a) Dickinson, T. A.; White, J.; Kauer, J. S.; Walt, D. R. Nature 1 9 9 6 , 382,
697-700. (b) Michael, K. L; Taylor, L. C.; Schultz, S. L.; Walt, D. R. Anal.
Chem. 1 9 9 8 , 70, 1242. (c) Dickinson, T. A.; Michael, K. L.; Kauer, J. S.;.;
Walt, D. R. Anal. Chem. 1 9 9 9 , 71, 2192. (d) Albert, K. J.; Walt, D. R.; Gill,
D. S.; Pearce, T. C. Anal. Chem. 2 0 0 1 , 73, 2501-2508.
(10) (a) Rakow, N. A.; Suslick, K. S. Nature 2 0 0 0 , 406, 710-713. (b) Di Natale,
C.; Salimbeni, D.; Paolesse, R.; Macagnano, A.; D’Amico, A. Sens. Actuators,
B 2 0 0 0 , 65, 220-226.
(11) (a) Seker, F.; Meeker, K.; Kuech, T. F.; Ellis, A. B. Chem. Rev. 2 0 0 0 , 100,
2505-2536. (b) Ivanisevic, A.; Ellis, A. B.; Ashkenasy, G.; Shanzer, A.;
Rosenwaks, Y. Langmuir 2 0 0 0 , 16, 7852-7858. (c) Sohn, H.; Letant, S.;
Sailor, M. J.; Trogler, W. C. J. Am. Chem. Soc. 2 0 0 0 , 122, 5399-5400.
The development of cross-reactive arrays for gas sensing
continues to be an active area of research. The goal of many
investigators in this area is the development of a practical
“electronic nose”.1 Gardner and Bartlett define an electronic nose
as “an instrument which comprises an array of electronic chemical
sensors with partial specificity and an appropriate pattern recogni-
† Current address: Department of Chemistry, Carleton College, Northfield,
Minnesota 55057.
(1) (a) Gardner, J. W.; Bartlett, P. N. Electronic Noses: Principles and Applica-
tions, Oxford University Press: New York, 1999. (b) Persaud, K.; Dodd, G.
H. Nature (London) 1 9 8 2 , 299, 352. (c) Albert, K. J.; Lewis, N. S.; Schauer,
C. L.; Sotzing, G. A.; Stitzel, S. E.; Vaid, T. P.; Walt, D. R. Chem. Rev. 2 0 0 0 ,
100, 2595-2626.
10.1021/ac011255d CCC: $22.00 © 2002 American Chemical Society
Published on Web 04/26/2002
Analytical Chemistry, Vol. 74, No. 11, June 1, 2002 2547

Chart 1
between the solvent and coordinated cyanide, and expansion or
contraction of the unit cell that is coupled to the Pt-Pt distance.12
The [Pt(CNR)4][Pt(CN)4] solids consist of infinite stacks of
alternating [Pt(CNR)42+] dications and [Pt(CN)42-] dianions.
Interionic metal-metal interactions produce the chromophore.12
The vapor inclusion causes color changes that result from a
combination of chemical interactions with the chromophore
including changes in the dielectric constant near the chro-
mophore, hydrogen bonding between the solvent and coordinated
cyanide, and expansion or contraction of the unit cell that is
coupled to changes in the Pt-Pt distance.
In an earlier communication,14 we reported on initial findings
with an array of three new vapoluminescent salts (see Chart 1):
[Pt(CN-cyclododecyl)4][Pt(CN)4] (1), [(phen)Pt(CN-cyclohexyl)2]-
[Pt(CN)4] (2 ), and [Pt(CN-n-tetradecyl)4][Pt(CN)4] (3 ), where
phen ) 1,10-phenanthroline. Since that communication, we have
also investigated an additional vapoluminescent salt, specifically,
[(phen)Pt(CN-cyclododecyl)Cl)]2[(phen)Pt(CN-cyclododecyl)2]2-
[Pt(CN)4]3 (4 ), (see Chart 1). In this paper, we report additional
data that address the reversibility, stability, and sensing charac-
teristics of these vapoluminescent materials for potential electronic
nose analytical applications.
precursors (formamides) are given in the Supporting Information
or were previously reported.12 The solvents used to generate
solvent vapors (acetone, cyclohexane, benzene, chloroform, 2-pro-
panol, 1-propanol, ethanol, methanol, water, dichloromethane)
were all of spectral grade or better. Nitrogen gas saturated with
solvent vapors was generated by slowly bubbling compressed
nitrogen through a flask, submerged in a room-temperature water
bath, containing the liquid solvent of interest. For most experi-
ments, no attempt was made to “dry” the compressed nitrogen
gas (industrial grade) used to carry solvent vapors into the three-
channel sensor unless otherwise noted. “Dry” nitrogen gas, when
used, was generated from the bleed of a liquid nitrogen dewar.
Solvent vapors were assumed to be saturated in the nitrogen
stream. This corresponds to the following concentrations (in
g/ m3): acetone (620), cyclohexane (380), benzene (300), chlo-
roform (1100), 2-propanol (130), 1-propanol (64), ethanol (130),
methanol (200), water (23), and dichloromethane (2000). Dilution
of the saturated vapor streams of acetone and methanol was
carried out with a partitioned flow system. The partitioned gas
flow system was constructed using mass flow controllers and
solenoid valves under computer control. Specifically, a 0-10 SSCM
mass flow controller (Omega FMA 1400) passed nitrogen gas
through a glass bubbler containing the solvent of interest. The
flow of nitrogen saturated with solvent vapor was teed into the
flow of a diluting nitrogen gas stream regulated by either a 0-10
or a 0-50 SCCM mass flow controller. A PC equipped with
LabVIEW (National Instruments) controlled the mixing of the
saturated vapor stream with the diluting nitrogen gas stream.
Percentages available ranged from 100 (pure saturated vapor, no
diluting nitrogen gas) to 0% (no saturated vapor, pure diluting
nitrogen gas). Percentages between 0 and 100% were corrected
for the expansion that occurs when the nitrogen gas flowing
through the solvent bubbler picks up solvent vapor, thus increas-
ing the flow rate of the solvent-saturated stream above the setting
of the mass flow valve feeding nitrogen into the solvent bubbler.
For clarity, the percentages listed in Figure 7 are uncorrected
values. The details of this correction and a table of corrected values
EXPERIMENTAL SECTION
General Considerations. The details for the synthesis and
characterization of 1 -4 , the isocyanide ligands, and the ligand
(12) (a) Exstrom, C. L.; Sowa Jr., J. R.; Daws, C. A.; Janzen, D.; Moore, G. A.;
Stewart, F. F.; Mann, K. R. Chem. Mater. 1 9 9 5 , 7, 15-17. (b) Daws, C. A.;
Exstrom, C. L.; Sowa Jr., J. R.; Mann, K. R Chem. Mater. 1 9 9 7 , 9, 363-
368. (c) Exstrom, C. L.; Pomije, M. K.; Mann, K. R. Chem. Mater. 1 9 9 8 ,
10, 942-945. (d) Buss, C. E.; Anderson, C. E.; Pomije, M. K.; Lutz, C. M.;
Britton, D.; Mann, K. R. J. Am. Chem. Soc. 1 9 9 8 , 120, 7783-7790. (e) Buss,
C. E.; Mann, K. R. J. Am. Chem. Soc. 2 0 0 2 , 124, ASAP. (f) Grate, J. W.;
Moore, L. K.; Janzen, D. E.; Veltkamp, D. J.; Kaganove, S.; Drew, S. M.;
Mann, K. R. Chem. Mater. 2 0 0 2 , 14, 1058-1066.
(13) (a) Cariati, E.; Bu, X.; Ford, P. C. Chem. Mater. 2 0 0 0 , 12, 3385-3391. (b)
Beauvais, L. G.; Shores, M. P.; Long, J. R. J. Am. Chem. Soc. 2 0 0 0 , 122,
2763-2772. (c) Mansour, M. A.; Connick, W. B.; Lachicotte, R. J.; Gysling,
H. J.; Eisenberg, R. J. Am. Chem. Soc. 1 9 9 8 , 120, 1329-1330.
(14) Drew, S. M.; Janzen, D. E.; Buss, C. E.; MacEwan, D. I.; Dublin, K. M.;
Mann, K. R. J. Am. Chem. Soc. 2 0 0 1 , 123, 8414-8415.
2548 Analytical Chemistry, Vol. 74, No. 11, June 1, 2002

solvent vapor in the raw spectral data matrix. The first three scores
of the PCA usually described greater than 90% of the variance in
the data set. See Supporting Information for loadings plots and
PC3 versus PC1 plots associated with all PCA calculations
reported. Differentiation of the solvent clusters in PC space was
quantified with a form of linear discrimination similar to that used
by Lewis et al.7 In short, the distance in PC space between solvent
clusters was described in a pairwise manner. First, a vector (wb),
which passes through the three-dimensional centroids of two
adjacent clusters (a and b), was defined and the distance between
the two centroids (dwb) was calculated. Next, the PC coordinates
of the two solvent clusters were projected onto the one-
dimensional distance vector and a standard deviation for each
cluster was calculated (σa,wb and σb,wb). A resolution factor (rf) was
then defined and calculated with eq 1.
Figure 1. Schematic of three-channel vapoluminescence sensor:
(A) Band-pass filter, 436 nm; (B) blue diodes; (C) CCD spectropho-
tometer; (D) three-channel sample holder and vapor flow cell; (E)
bifurcated fiber optics; (F) filter holder; (G) solvent vapor flow in; (H)
solvent vapor flow out.
for the methanol and acetone concentrations are included in the
Supporting Information.
Three-Channel Vapoluminescence Sensor. A three-channel
vapoluminescence sensor was constructed using the following
components: a set of blue light-emitting diodes (Radio Shack, λmax
) 468 nm), bifurcated fiber optics (Edmond Scientific), a three-
channel sensor gas flow cell (constructed in-house), and a 256 ×
1024 CCD spectrophotometer (Princeton Instruments). A sche-
matic diagram of the apparatus is depicted in Figure 1. The blue
diodes were mounted on one side of a 436-nm band-pass filter
and monochromatic excitation light was collected by three fiber-
optic cables mounted on the opposite side. The excitation light
entered the three-channel sensor gas flow cell through three
bifurcated fiber-optic cables that also collected the luminescence
from the sensor compounds and passed it on to the CCD camera.
The gas flow cell was constructed by mounting the three
bifurcated fiber-optic cables in a plate that could be positioned
above three wells that comprise the three channels of the sensor.
For experiments that did not require heating of the sensor
compounds, the compounds were deposited from a diethyl ether
suspension into each well. The amount of sensor compound
deposited ranged from 0.5 to <0.1 mg. For experiments that
required a heating cycle, the sensor compounds were deposited
onto a flexible, foil heater (Minco) or a set of thermistors
(DigiKey) that were designed to heat to a specific temperature
(50, 80, 100 °C). The temperature of the heating device was
monitored by mounting a thermocouple in the cell. A space
between the plate with the mounted fiber optics and the plate
holding the sensor compound was maintained with a Teflon
gasket. A nitrogen stream carrying solvent vapors was passed
through this space. The luminescence collected by the three
bifurcated fiber optics was passed though an in-line filter to remove
any reflected excitation light. The ends of the three fibers were
mounted in a vertical line on a disk that was mounted at the
entrance slit to the CCD camera. The luminescence from the three
fibers was used to position the disk so that an image of three
distinct luminescence spectra, one for each channel of the sensor
array, could be acquired simultaneously. The three spectra
obtained were binned, and dark current was subtracted on-line.
P rincipal Component Analysis. The binned luminescence
spectra acquired for each channel of the sensor were normalized
to the maximum intensity found on each channel and mean
centered before assembling the data into a matrix for principal
component analysis (PCA).15 PCA was based on the covariance
matrix of the raw spectral data. Typically, 46 points were used
from each channel, for a total of 138 data points describing each
2
rf ) d /
σ
x
2 + σb,wb
(1)
wb
a,wb
Placing the constraint of a Gaussian distribution on the data
set allows the overlap between the two clusters to be determined.
In general, if two adjacent clusters have similar standard deviations
when projected onto their distance vector, an rf value of 3.75 or
greater implies less than 1% overlap between the two Gaussian
curves.
RESULTS AND DISCUSSION
Transducer Materials. The pure cation platinum(II) double
salt materials synthesized for this study (1 -3 ) were synthesized
and characterized using methods we have previously reported.12
All three materials were synthesized by reacting [(n-C4H9)4N]2-
[Pt(CN)4] with solutions of [Pt(C12H23-NC)4]2+, [(phen)Pt(C6H11-
NC)2]2+, or [Pt(C14H29-NC)4]2+ in acetonitrile to give 1 -3 ,
respectively. Compounds 1 -3 were all highly colored and gave
the expected IR spectra and elemental analyses. Compound 4 is
a mixed cation platinum(II) double salt, the likes of which we
have not previously reported. In this case, the reaction of the
starting material [(phen)Pt(DMSO)Cl)]+ with C12H23-NC at a 1:2
stoichiometry consistently led to the formation of a 1:1 mixture
of platinum(II) cations: [(phen)Pt(C12H23-NC)Cl]+ and [(phen)-
Pt(C12H23-NC)2]2+. Addition of [(n-C4H9)4N]2[Pt(CN)4] to the
solution of cations precipitated a double salt with a reproducible
mixture of +1 and +2 cations, [(phen)Pt(CN-cyclododecyl)Cl)]2-
[(phen)Pt(CN-cyclododecyl)2]2[Pt(CN)4]3, as evidenced by the
elemental analysis and spectroscopic characterizations. Additional
evidence for the formulation of 4 as [(phen)Pt(CN-cyclododecyl)-
Cl)]2[(phen)Pt(CN-cyclododecyl)2]2[Pt(CN)4]3 is provided in the
Supporting Information.
Sensor Array Characterization. As we previously described,14
arrays of platinum(II) double salt compounds produce lumines-
cence spectra that depend on the solvent vapor in contact with
the compounds. When these spectra were analyzed by PCA,
distinct clusters representing various solvent vapors were ob-
tained. It was reported that exposure of the array to nitrogen gas
saturated with acetone vapor between analyte solvent vapor
exposures seemed to improve the reversibility and reproducibility
(15) (a) Jurs, P. C.; Bakken, G. A.; McClelland, H. E. Chem. Rev. 2 0 0 0 , 100,
2649-2678. (b) Jackson, J. E. A User’s Guide to Principal Components; Wiley
and Sons: New York, 1991.
Analytical Chemistry, Vol. 74, No. 11, June 1, 2002 2549

Figure 2. PC2 vs PC1 score plot of luminescence spectra acquired with array I with intervening acetone exposures (T ) 22 °C). Individual
percent variances: PC1 48; PC2 24. Solvent vapors: acetone (b), 1-propanol (0), 2-propanol (]), benzene (4), cyclohexane (O), chloroform
(9), dichloromethane ([), water (2), methanol (1), and ethanol (3).
of the array response;14 however, this was not totally effective as
certain solvent vapor exposure sequences still gave irreproducible
spectra.14 Temperature programming between solvent vapor
exposures, i.e., a regulated ramped heating and cooling cycle of
the array elements in the presence of nitrogen gas, was found to
be much more effective. Figures 2 and 3 provide a comparison of
the effects of acetone vapor exposure relative to temperature
programming on the discrimination power of an array consisting
of 1-3, which will be referred to as array I. Tables 1 and 2 provide
the rf values for the 45 solvent cluster pairs in Figures 2 and 3,
respectively. Statistical analysis of the rf tables showed that the
intervening acetone exposure gave a set of rf values with a mean
of 6.3, a minimum of 0.51, and a maximum of 17. In addition, 13
of the rf values out of the 45 solvent vapor pairs were less than
3.75. In general, an rf value of less than 3.75 indicates a greater
than 1% overlap and therefore relatively poor discrimination
between the two solvents. When intervening temperature pro-
gramming was introduced, the mean increased to 13, with a
minimum rf of 1.4 and a maximum rf of 41. The number of rf
values less than 3.75 decreased to 5 out of the 45 solvent vapor
pairs; however, the resolution between several pairs decreased,
especially some involving benzene vapor (vide infra). Obviously,
temperature programming improved the average discrimination
ability of the array and merited further investigation.
much above room temperature, so it was deposited into an
unheated well in the array flow cell. Four temperatures were
chosen for study: 25, 50, 80, and 100 °C. After a round of solvent
exposure and acquisition of the associated luminescence spectra
(2-3 min.), the gas flow was switched to pure compressed
nitrogen. For the room-temperature studies, the nitrogen flow was
maintained for 4-5 min before the next solvent vapor exposure
occurred. The cycle time is more a function of the thermal time
constant and dead volume of the cell rather than the intrinsic
response time of the array, which is on the order of 10 s or less
for most vapors. To achieve higher temperatures between solvent
vapor exposures, a voltage (15 V dc) was placed across the
thermistors, which quickly (<30 s) heated them to their pre-
scribed temperature maximums. The temperature maximum was
held for an additional 30 s, after which the voltage difference was
discontinued and the thermistors were allowed to cool back to
room temperature (25 ( 1 °C). The cooling time was determined
by the maximum temperature and ranged from ∼3 min for a 50
°C temperature maximum to greater than 7 min for a 100 °C
temperature maximum. When room temperature was reestab-
lished in the sensor cell a “background” nitrogen spectrum was
acquired. Next, the nitrogen flow was diverted through a solvent,
creating a nitrogen stream saturated with solvent vapor. A new
luminescence spectrum was established very quickly, usually in
less than 30 s. This luminescence spectrum was also recorded at
25 ( 1 °C. The array of luminescence spectra acquired for each
solvent vapor at each temperature was examined by PCA followed
by an rf calculation involving all the possible solvent pairs in PC
space. The results of these experiments are described in Table 3.
The data in Table 3 reveal a definite improvement in the ability
Several experiments were performed to optimize the effective-
ness of temperature programming. For the first set of experiments,
array I was used in the three-channel vapoluminescent sensor.
For experiments involving temperature programming, 1 and 2
were deposited on the surface of thermistors mounted in the array
flow cell. Compound 3 was found to decompose when heated
2550 Analytical Chemistry, Vol. 74, No. 11, June 1, 2002

Figure 3. PC2 vs PC1 score plot of luminescence spectra acquired with array I with intervening temperature programming under nitrogen.
Only 1 and 2 are directly heated by ramping up to 80 °C and then cooling to 25 ( 1 °C before acquiring spectra. Individual percent variances:
PC1 70; PC2 15. Solvent vapors: acetone (b), 1-propanol (0), 2-propanol (]), benzene (4), cyclohexane (O), chloroform (9), dichloromethane
([), water (2), methanol (1), ethanol (3), and nitrogen (s).
Table 1. Rf Values for Array I with Intervening Acetone Exposuresa,b
methanol
dichloromethane
acetone
benzene
ethanol
chloroform
1-propanol
2-propanol
cyclohexane
dichloromethane
acetone
benzene
4.97
16.6
8.71
4.40
7.02
4.46
8.19
16.4
4.03
8.85
8.54
6.32
8.32
5.21
6.84
9.14
7.35
8.81
10.8
2.28
8.46
5.92
3.14
11.5
ethanol
1.39
0.723
5.20
2.44
5.89
5.45
chloroform
1-propanol
2-propanol
cyclohexane
water
1.21
0.510
4.50
12.3
1.43
2.87
1.62
1.92
3.52
5.72
13.6
2.45
6.85
4.12
12.1
a
Array maintained at a constant temperature of 22 °C. b These data previously reported in graphical format in ref 15.
of the sensor array to differentiate between solvent vapor pairs
as the temperature maximum of the temperature program is
increased. The statistics show a meaningful decrease in the
number of unresolved pairs along with an improvement in the
other statistical descriptors. This effect appears to maximize with
the 80 °C temperature program. When a 100 °C temperature
program is applied, the resolution between many of the adjacent
solvent vapor pairs is degraded to nearly the level of the room-
temperature data. This effect is caused by a gradual decrease in
the emission intensity of 3 due to the unintended heat transfer
from the thermistors heating 1 and 2 when the 100 °C temper-
ature program is run many successive times. The nitrogen
“background” spectrum of 3 decreases ∼15% over the span of 80
heating/ solvent exposure cycles and is more erratic when the
temperature program reaches 100 °C relative to the 50 °C or the
80 °C data. Additionally, if the data from 3 are dropped from the
array and PCA is performed on the remaining data generated by
1 and 2 at 80 and 100 °C, the statistical results improve markedly
at 100 °C while they worsen at 80 °C (see Table 3). These data
support the premise that, at 100 °C, it is the gradual decomposition
of 3 that decreases the discriminating power of the array, while
at 80 °C, a more stable temperature for 3 , its presence enhances
the effectiveness of the array.
On the basis of these results, we decided to replace 3 with 4
to form a new array designated as array II. Although 4 has a
response less orthogonal to 1 and 2 , in comparison to 3 , it is
much more thermally stable. This allows the entire array to be
heated during the temperature programming step, a simplification
that facilitated some additional interesting experiments. In this
case, the three compounds were deposited onto a flexible foil
heater and placed in the three-channel array sensor flow cell.
When 25 V ac was applied across the leads of the flexible foil
heater, the temperature at its surface quickly (<30 s) increased
to 75-80 °C, at which time the voltage was turned off and the
Analytical Chemistry, Vol. 74, No. 11, June 1, 2002 2551

Table 2. Rf Values for Array I with an Intervening Temperature Programming Cyclea
methanol
dichloromethane
acetone
benzene
ethanol
chloroform
1-propanol
2-propanol
cyclohexane
dichloromethane
acetone
benzene
22.3
41.0
11.3
16.3
28.7
6.43
11.9
13.7
7.33
30.0
16.9
35.0
24.7
33.4
19.3
12.4
20.3
8.51
22.7
18.7
17.7
19.3
1.83
22.7
ethanol
4.51
1.39
5.40
2.42
4.99
5.00
chloroform
1-propanol
2-propanol
cyclohexane
water
6.51
6.12
6.27
7.23
6.05
9.43
3.68
6.59
9.28
4.79
8.60
1.74
6.25
6.29
8.66
a
Only 1 and 3 are heated, ramped up to 80 °C, and then cooled to 25 ( 1 °C before acquiring spectra.
Table 3. Statistical Results with Array I and Different
Types of Temperature Programming
temperature (°C)
25a
50b
11
0.47
41
80b
100b
80b,c
100b,c
mean rf
min rf
max rf
no. of rf <3.75
no. of rf >3.75
5.5
0.56
13
14
31
13
1.4
41
5
8.1
0.47
21
14
31
13
1.4
4.6
9
8.9
1.6
22
8
5
40
40
36
37
a
Constant temperature. b Ramp up to indicated temperature and
then cool to 25 ( 1 °C before acquiring spectra. c Eliminate 3 data
from PCA.
Table 4. Statistical Results with Array II and Different
Types of Temperature Programming
temperature (°C)
22a
50a
80b
80b,c
mean rf
min rf
max rf
no. of rf <3.75
no. of rf >3.75
9.9
0.63
59
13
32
6.7
1.1
22
13
32
11
0.53
54
7
38
13
2.2
46
6
39
a
Constant temperature. b Ramp up to indicated temperature and
then cool to 25° ( 1 °C before acquiring spectra. c N2 stream dried
with liquid N2.
heater was allowed to cool back to room temperature. The results
of these experiments are described in Table 4.
Comparing the results at 80 °C for array I (Table 3) to array
II (Table 4) brings up several points that deserve comment. First,
the statistical descriptors are fairly similar between the two arrays,
with I producing a little more discriminating power for this solvent
set compared to array II. Even though the individual compounds
on II are not as orthogonal in their response to this solvent vapor
set as I, II still does moderately well by compensating with a more
reproducible spectral pattern across all three channels. Figure 4
demonstrates the more orthogonal response of array I (1 -3 )
relative to II (1 , 2 , 4 ) for methanol vapor exposures relative to
“background” nitrogen exposures. Figure 4 also highlights the
greater reproducibility achieved with II relative to I for methanol
vapor exposures. The reproducibility of I appears to suffer mostly
due to the slow thermal decomposition of 3 , even using an 80 °C
temperature program.
Figure 4. Luminescence spectra of 1-4 under methanol vapor (-)
and nitrogen ()) acquired between a series of solvent exposures
at 25 ( 1 °C after a temperature programming step to 80 °C
during the nitrogen cycle. The nitrogen spectrum is an average of
the four spectra acquired before the displayed methanol vapor
spectra.
was achieved by placing 15 V dc across the leads of the heater.
The thermocouple in the cell indicated that the temperature
equilibrated at ∼48 ( 2 °C during the course of the experiment.
Under these conditions, the spectral changes that occurred in the
presence of the various solvent vapors were much less pronounced
than at constant room temperature (22 °C); however, the repro-
ducibility of the spectra was much better at the constant elevated
The response of II at a constant elevated temperature of ∼50
°C on a flexible foil heater was also studied. This temperature
2552 Analytical Chemistry, Vol. 74, No. 11, June 1, 2002

Figure 6. Luminescence spectra of 1 under varying concentrations
of methanol vapor. Spectra acquired at 25 ( 1 °C after a temperature
programming step to 80 °C during the nitrogen cycle. Each methanol
spectrum is an average of three or four cycles while the nitrogen
spectrum (a) is the average of all nitrogen backgrounds obtained (N
) 27). Concentrations (percent saturation): a, 0; b, 1; c, 6; d, 11; e,
22; f, 54; g, 78; h, 100.
program was run followed by cooling to 25 ( 1 °C, diversion of
the nitrogen through benzene solvent, and finally acquisition of
an equilibrated luminescence spectrum. As seen in Figure 5, a
series of consecutive exposures of benzene vapor to II under
various levels of background humidity led to mostly reproducible
responses from 2 and 4 but an obviously changing response from
1. At the end of this experiment, the array was exposed to nitrogen
saturated with water vapor. The spectra acquired under water
vapor are also included in Figure 5. Note that 2 and 4 showed
markedly different responses to saturated water vapor relative to
their benzene responses; however, 1 only changed slightly. In
addition, to recover the luminescence spectra obtained under
benzene vapor in dry nitrogen after a water vapor exposure, a
heating cycle was required followed by a ∼2-h exposure to dry
nitrogen. This indicates that even this heating cycle may not
remove all the residual water in the sensor materials. On the basis
of these results, 1 appears to be the most sensitive to low
background humidity levels. We also conclude that the equilib-
rium constant for the inclusion of water into sensor 1 is likely
much larger than the corresponding equilibrium constant for
benzene and probably other hydrocarbons as well. This is in line
with expected trends in the strength of hydrogen bonding com-
pared with dispersion forces between the chromophore and the
solvent guest. The sensitivity of some of these sensor materials
to background humidity levels raises an issue that must be suc-
cessfully addressed if these compounds are to be used as a prac-
tical sensor array. Finally, even though these studies suggest dif-
ferent humidity levels are a major source of noise in these systems,
we have not eliminated the possibility that other high-boiling con-
taminants present in the various solvents could also be problem-
atic. Temperature programming between solvent vapor exposures
is expected to partially mitigate this potential problem as well.
The ability of array I to differentiate and detect analyte gases
at concentrations less than saturation was also investigated.
Figure 5. Luminescence spectra of array II (1, 2, 4) under saturated
benzene vapor with varying degrees of humidity present. Humidity
increases in the following order: ‚‚‚ < - - - < - - - < - - - ; the solid
line is the spectrum at 100% humidity with no benzene present.
Spectra acquired at 25 ( 1 °C after a temperature programming step
to 80 °C during the nitrogen cycle. Each spectrum is an average of
four cycles under each of the humidity levels studied.
temperature of 50 °C, resulting in clusters that were tighter at 50
°C relative to 22 °C but also much closer together. In the end,
the degree of differentiation appears to be nearly the same due
to the opposite effect of these two factors.
The reasons for the success of the temperature programming
protocol were investigated. Early on we observed that careful
removal of background water vapor levels in the compressed
nitrogen gas stream used between exposures seemed to slightly
improve the reproducibility, and thus the discriminating power,
of the array. As seen in Table 4, a slight improvement in the
statistical descriptors for the 80 °C temperature programming
protocol results if dry nitrogen gas was used as a carrier instead
of compressed nitrogen. Therefore, it seemed reasonable that
variations in water vapor contamination, present to some degree
in all solvents, were at least partially mitigated by temperature
programming. To further investigate this premise, a series of
experiments were performed with benzene vapor, an analyte that
generally gives the most scattered clusters in PC space. Array II
was exposed repeatedly to benzene vapor after the array had been
extensively dried by exposure to dry nitrogen for 2 h. A cycle
consisted of exposure to nitrogen while the 80 °C temperature
Analytical Chemistry, Vol. 74, No. 11, June 1, 2002 2553

Figure 7. PC2 vs PC1 score plot from PCA of luminescence spectra of array I with varying acetone (red) and methanol (blue) concentrations
in nitrogen. Filled points were used to plot the ascending concentration data, and open points were used for the descending concentration data.
The red and blue lines are visual guides for the methanol and acetone data, respectively. Only 1 and 2 are directly heated by ramping up to 80
°C and then cooling to 25 ( 1 °C before acquiring spectra. Individual percent variances: PC1 84; PC2 8.2. Approximate concentrations (percent
saturation): 0 (-), 1 (×), 5 (3), 10 (O), 20 (0), 50 (]), 75 (4), and 100 (!); please see Supporting Information for exact concentrations.
Saturated methanol and acetone vapors were diluted with nitrogen
gas and passed over the array elements in separate experiments.
These particular solvent vapors were chosen as they represent a
vapor that causes major spectral changes at saturation, i.e.,
methanol, and a vapor that induces only mild spectral changes at
saturation, i.e., acetone. Figure 6 shows an example of the type
of data acquired in these experiments for methanol exposure to
1 . Note that not only does the intensity vary but the spectrum
also blue shifts as the concentration of methanol is increased in
a process that is at least biphasic. Because an isosbestic point
was not obtained, the intensity at one wavelength as a function of
concentration does not extract all the information in the data set.
Therefore, the luminescence spectra collected in the two experi-
ments were pooled and analyzed by PCA. Figure 7 shows the first
two principal components of the analysis. As expected, the meth-
anol vapor data covered a much larger portion of PC space than
acetone due to the more extensive spectral changes methanol
evokes. Data points collected ascending and descending the con-
centration scale are nearly identical, so little or no hysteresis is
associated with the concentration response of the sensor array
with respect to methanol or acetone exposure. The nitrogen back-
ground for the acetone data set was slightly different from the
nitrogen background for the methanol data set (rf ) 5.9), probably
due to the difference in water contamination in acetone relative
to methanol. Apparently, the temperature programming does not
remove all the residual water from the sensor materials; therefore,
a different steady-state background level of water is achieved in
acetone relative to methanol. The rf analysis of the data also
indicates that the entire acetone concentration range was resolved
from every point in the methanol concentration range using the
criterion that neighboring clusters having rf values greater than
3.75 can be differentiated (see Supporting Information).
A detection limit was estimated using the standard deviation
of the nitrogen PC coordinates (blank) along vectors to neighbor-
ing concentration clusters (signal) in a technique similar to the
rf calculation. Invoking a typical “standard deviation of the blank
times three” criterion for the detection limit, a minimal detectable
distance from the centroid of the nitrogen cluster was determined
for acetone and methanol. Based on this approach, the detection
limit for acetone was at 12% saturation (75 g/ m3) using array I
with the specified temperature programming protocol. Methanol
gave a lower detection limit at 3% (6 g/ m3).
CONCLUSIONS
We have further demonstrated the potential analytical applica-
tions of platinum(II) double salt materials in cross-reactive sensor
arrays. Applying a temperature cycle between solvent exposures
greatly improved the ability of the arrays studied to differentiate
between a set of 10 solvent vapors. Care must be taken in the
choice of the maximum temperature in the heating cycle as some
of the platinum(II) double salt materials studied here tend to
decompose at elevated temperatures. In addition, some of these
materials were also sensitive to background water vapor levels
that reduced the reproducibility of the luminescence spectra
obtained for a given solvent vapor. A practical cross-reactive sensor
array of platinum(II) double salt materials may be possible if they
2554 Analytical Chemistry, Vol. 74, No. 11, June 1, 2002

are designed at the molecular level to be thermally stable and
insensitive to low levels of water vapor contamination. The
nonlinear response observed for some of the vapors studied is
also a complicating factor that will need to be addressed. Synthetic
experiments designed to produce new compounds that meet these
additional requirements are underway in our laboratory. Finally,
concentration studies comparing acetone and methanol have
shown for the first time that arrays of platinum(II) double salt
materials can be used to differentiate two solvent vapors over their
entire detectable concentration range.
SUPPORTING INFORMATION AVAILABLE
Synthesis and characterization details for 2 -4 , isocyanide
ligands, and ligand precursors. Additional evidence for the form-
ulation of 4 . Method used to correct percent saturation values
for the uptake of solvent vapor with the partitioning flow apparatus.
PC3 vs PC1 plots for the data presented in Figures 2, 3, and 7.
Loadings plots for the data presented in Figures 2, 3, and 7. Rf
table for the methanol and acetone concentration data. This
material is available free of charge via the Internet at http:/ /
pubs.acs.org.
ACKNOWLEDGMENT
This work is supported by the University of Washington Center
Received for review December 11, 2001. Accepted
February 15, 2002.
for Process Analytical Chemistry (CPAC) and the National Science
Foundation, most recently under Grant CHE-9307837. S.M.D.
acknowledges sabbatical support from the Howard Hughes
Medical Institute and Carleton College.
AC011255D
Analytical Chemistry, Vol. 74, No. 11, June 1, 2002 2555