A Photothermal Interferometer for Gas-Phase Ammonia Detection

Melody Avery Owens,*,† Christopher C. Davis,‡ and Russell R. Dickerson§ ... (4) Lee, D. S.; Halliwell C.; Garland, J. A.; Dollard, G. J.; Kingdon...
0 downloads 0 Views 109KB Size
Anal. Chem. 1999, 71, 1391-1399

A Photothermal Interferometer for Gas-Phase Ammonia Detection Melody Avery Owens,*,† Christopher C. Davis,‡ and Russell R. Dickerson§

Chemical Physics Program, University of Maryland, College Park, Maryland 20742

Detection of gas-phase ammonia is particularly challenging because ambient ammonia concentrations may be less than 1 ppb (molecules of NH3 per 109 molecules of air), ammonia sticks to many materials commonly used to sample air, and particles containing ammonium may interfere with gas-phase measurements. We have built a new and sensitive photothermal interferometer to detect gas-phase ammonia in situ, under typical atmospheric conditions. Ammonia molecules in sampled air absorb infrared radiation from a CO2 laser at 9.22 µm, with consequent collisional heating, expansion, and refractive index change. This change in refractive index is detected as a phase shift in one arm of a homodyne interferometer. Measurements of vibrational and electrical noise in the interferometer correlate to an instrumental lower limit of detection of 6.6 ppt ammonia in 1 s. The CO2 laser output is modulated at 1.2 kHz, and the ac signal from the interferometer is measured with a lock-in amplifier. The detector is zeroed by sampling through a H3PO4-coated denuder tube and is calibrated by dynamic dilution of two permeation tube outputs and by standard addition. Signal gain is insensitive to CO2 or H2O in the sample, and the signal is linear over 5 orders of magnitude. The instrument 2σ precision is 31 ppt when the signal is integrated for 100 s and 250 ppt with a 1-s integration time. The windowless sample cell and inlet is fabricated entirely of glass to minimize sample loss and hysteresis. The instrument response time is demonstrated to be about 1 s. Ammonia in the atmosphere plays an important role in diverse ecological and physical systems. Ammonium is an important plant nutrient and a natural product of organic waste processing by microorganisms in soils. Plants and soils exchange gas-phase ammonia with the atmosphere, and ammonia deposition may be a significant fraction of total nitrogen deposition.1-4 Estimates of † Present address: Center for Atmospheric Sciences, Department of Physics Hampton University, Hampton, VA 23668: (e-mail) [email protected]; (fax) (757) 864-5841. ‡ Also with Electrical Engineering Department, University of Maryland: (email) [email protected]. § Also with: Meteorology Department, University of Maryland: (e-mail) [email protected]. (1) Sprent, J. I. The Ecology of the Nitrogen Cycle; Cambridge University Press: New York, 1987; pp 1-151. (2) Fisher, D. C.; Oppenheimer, M. Ambio 1991, 20, 102-107. (3) Harper, L. A.; Sharpe, R. R. Atmos. Environ. 1998, 32 (3), 273-277. (4) Lee, D. S.; Halliwell C.; Garland, J. A.; Dollard, G. J.; Kingdon, R. D. Atmos. Environ. 1998, 32 (3), 431-439.

10.1021/ac980810h CCC: $18.00 Published on Web 02/20/1999

© 1999 American Chemical Society

the atmospheric-biospheric budget of ammonia and total reactive nitrogen cycling are hampered by a lack of accurate gas-phase ammonia measurements. Anthropogenic loading of ammonia to the atmosphere appears to be increasing in some regions because of population increase and consequent fertilizer use and increased animal husbandry, though the difficulty of obtaining an accurate record of ammonia measurements makes the establishment of trends difficult.5-10 Direct ammonia flux measurements would be particularly valuable because surfaces such as soil, plants, or water may be either sources or sinks for ammonia. Accurate gas-phase ammonia and ammonia flux measurements are needed but lacking because ammonia is particularly challenging to measure. The polar ammonia molecule adheres to many materials commonly used to fabricate sample inlets, causing hysteresis, and measurement inaccuracies and inhibiting true realtime fast-response measurements. Ammonia forms particles that frequently interfere with gas-phase measurements, and sampling efficiency or signal gain is often sensitive to atmospheric water vapor. Several reviews and intercomparisons of ammonia detection techniques have been published.11-13 Sampling methods include collection of ammonia on coated filters, denuder tubes, or with diffusion scrubbers, followed by detection of ammonium ions in solution. Denuder tubes are also used to preconcentrate ammonia for conversion to NO on a hot metal catalyst, followed by chemiluminescent detection of NO. Filter packs, denuder tubes, and scrubbers may have very low background detection limits, but measurement frequency is limited by blank or sample collection time and information about the variability of ambient ammonia during that time is lost. Although results from inter(5) Nihlgard, B. Ambio 1985, 14 (1), 1-8. (6) Galloway, J. N.; Schlesinger, W. H.; Levy, H., II; Micheals, A.; Schnoor, J. L. Global Biogeochem. Cycles 1995, 9 (2), 235-252. (7) Fuhrer, K.; Neftel, A.; Anklin, M.; Staffelbach, T.; Legrand, M. J. Geophys. Res. D 1996, 101, 4147-4164. (8) Doscher, A.; Gaggeler, H. W.; Schotterer, U.; Schwikowski, M. Geophys. Res. Lett. 1996, 23 (20), 2741-2744. (9) Galloway, J. N.; Dianwu, Z.; Thomson, V. E.; Chang, L. H. Atmos. Environ. 1996, 30 (10/11), 1551-1561. (10) Horvath, L.; Sutton, M. A. Atmos. Environ. 1998, 32 (3), 339-344. (11) Williams, E. J.; Sandholm, S. T.; Bradshaw, J. D.; Schendel, J. S.; Langford, A. O.; Quinn, P. K.; LeBel, P. J.; Vay, S. A.; Roberts, P. D.; Nortons, R. B.; Watkins, B. A.; Buhr, M. P.; Parrish, D. D.; Calvert, J. G.; Fehsenfeld, F. C. J. Geophys. Res. 1992, 97, 11591-11611. (12) Harrison, R. M.; Kitto, A.-M. N. Atmos. Environ. 1990, 24A, 2633-2640. (13) Wiebe, H. A.; Anlauf, K. G.; Tuazon, E. C.; Winer, A. M.; Biermann, H. W.; Appel, B. R.; Solomon, P. A.; Cass, G. R.; Ellestad, T. G.; Knapp, K. T.; Peake, E.; Spicer, C. W.; Lawson, D. R. Atmos. Environ. 1990, 24A, 1019-1028.

Analytical Chemistry, Vol. 71, No. 7, April 1, 1999 1391

comparisons between these instruments are inconclusive, particle interference and sampling or conversion efficiency may depend on the temperature or relative humidity of the ambient air.12-16 Direct ammonia measurements by optical techniques based on spectroscopic properties of the ammonia molecule do not require the two-stage process of sample collection and analysis. Optical ammonia detection methods allow rapid real-time sampling, although instrument response faster than 1 Hz is elusive due to sampling limitations common to all in situ ammonia detection methods. An elegant spectroscopic ammonia measurement technique based on ultraviolet photofragmentation followed by laser-induced fluorescence (PF/LIF) is selective and has the lowest reported ammonia detection limit, 5-10 ppt in 5 min.11,17 However, size limitations reduce the possibility that the PF/LIF detector will be developed to make flux measurements, and there is a need for alternative optical ammonia detection methods that are simpler to deploy in the field. We have designed, built, and tested a promising new instrument to detect gas-phase ammonia in ambient air by photothermal interferometry (PTI). Ammonia molecules in sampled air are excited to upper vibrational rotational states of the ν2 band of ammonia by absorption of modulated 9.22-µm infrared light from a carbon dioxide laser. Rapid collisional relaxation of the excited ammonia molecules causes a small temperature rise, expansion, and refractive index change in the sampled air. We have built a highly sensitive homodyne interferometer to detect this linear refractive index change. This technique can be adapted to detect other molecular species with appropriate matching of pump laser/ molecular absorption frequencies, although this paper will focus on ammonia detection. We will describe our photothermal interferometric technique, including calibration and zeroing methodology. The linear dynamic range and detection limit of our instrument has been measured and will be described, as well as the sensitivity of the signal to CO2 and water vapor. We have also tested the temporal response of the detector to a large step change in ammonia concentration. We will show that our technique exhibits a unique combination of high instrument sensitivity and short response time.

ideal optical pump. Collisional relaxation of the excited ammonia molecules causes a temperature rise, ∆T, in the sampled air:

∆T ) PCO2σNH3NNH3/2fFCpπa2

where PCO2 is the CO2 laser power output, f is the laser modulation frequency, F and Cp are the density and heat capacity of the sampled air, respectively, NNH3 is the molecular number density (molecules per unit volume) of ammonia, and a is the laser beam radius. The density of air, F, is equal to NM, where N is the molecular number density of air and M is the average molecular weight of air. The mixing ratio, ξ, can be defined as NNH3/N. With this substitution, we show that ∆T is proportional to the mixing ratio of ammonia:

∆T ) PCO2σNH3ξNH3/2fMCpπa2

(14) Appel, B. R.; Tokiwa, Y.; Kothny, E. L.; Wu, R.; Povard, V. Atmos. Environ. 1988, 22 (8), 1565-1573. (15) Sickles, J. E., II; Hodson, L. L.; McClenny, W. A.; Paur, R. J.; Ellestad, T. G.; Mulik, J. D.; Anlauf, K. G.; Wiebe, H. A.; Mackay, G. I.; Schiff, H. I.; Bubacz, D. K. Atmos. Environ. 1990, 24A, 155-165. (16) Sorenson, L. L.; Granby, K.; Nielsen, H.; Asman, W. A. H. NERI Technical Report 99. The Danish Ministry of the Environment, 1994. (17) Schendel, J. S.; Stickel, R. E.; van Dijk, C. A.; Sandholm, S. T.; Davis, D. D.; Bradshaw, J. D. Appl. Opt. 1990, 29, 4924-4937. (18) Brewer, R. J.; Bruce, C. W. Appl. Opt. 1978, 17, 3746-3749.

1392 Analytical Chemistry, Vol. 71, No. 7, April 1, 1999

(2)

The rate constant for relaxation of ammonia in the ν2 state at room temperature and pressure is 9.6 × 106 Hz.19 Conduction of heated air in our sample is slow relative to the rapid relaxation of excited ammonia molecules, and displacement due to convection of the heated air is small compared to the radius of the laser probe beam.20 It is thus reasonable to regard heating of the sampled air as both instantaneous and localized within the radius of the laser beam. The PTI detector can be modulated at any frequency free of acoustic and vibrational noise, provided that it is much slower than this relaxation rate. We choose to modulate the CO2 laser and detect the signal at 1.2 × 103 Hz. In the absence of conduction, convection, and acoustical effects, the local instantaneous heating causes a modulated refractive index change, ∆n, following the Gladstone-Dale equation:21

∆n ) ((n - 1)/T0)∆T

(3)

The modulated refractive index change creates a modulated phase shift, φm, in He-Ne light traveling collinear to the CO2 beam along one arm of a sensitive homodyne interferometer:

φm ) 2πl∆n/λ EXPERIMENTAL SECTION Theory. The physical principles behind photothermal interferometric ammonia detection are described in this section. Ammonia has a strong absorption feature at 9.22 µm due to six nearly coincident rotational states and a transition between the ground state and the ν2 vibrational mode. The absorption cross section, σNH3, at this wavelength is quite large, 56 atm-1 cm-1 at 273 K.18 Nearly coincident with this large ammonia absorption is the 9R(30) CO2 rotation transition, so a tunable CO2 laser is an

(1)

(4)

In eq 4, l is the laser beam interaction path length and λ is the He-Ne laser wavelength. An instrumentation amplifier subtracts the photodiode signals, and the output is detected with a phasesensitive lock-in amplifier. The interferometer signal, S, is proportional to ∆φm, the modulated phase difference between the signal and reference beams:

S ) A sin(∆φm) ) A(∆φm), true for small ∆φm

(5)

The interferometer amplification factor, A, is a measure of the total signal gain in the photodiodes and amplifier. A is measured as 10 V per radian of ∆φm by tuning the interferometer through a signal maximum where sin(∆φm) ) 1. (19) Hovis, F. E.; Moore, C. B. J. Chem. Phys. 1978, 69 (11), 4947-4950. (20) Davis, C. C.; Petuchowski, S. J. Appl. Opt. 1981, 20 (14), 2539. (21) Gladstone, J. H. Proc. R. Soc. 1868, 16, 439-444.

Figure 1. Schematic diagram of the ammonia detector. A single-mode He-Ne laser beam of 1-mW intensity is split into two beams by an etalon at the front of the detector. The modulated CO2 laser beam is introduced almost collinear to one of the two He-Ne beams. Laser beam and gas interaction path length is 38 cm. The two He-Ne beams are recombined by a second etalon at the rear of the detector and are reflected into two photodiodes. Air may enter the detector through a straight glass inlet or an acid-coated inlet that removes gaseous NH3. Pumping speed is 2 L/min, and residence time inside the sample cell and inlet tubing is 1 s.

The (2σ detection limit of our interferometer at this frequency is 0.12 µrad, as measured by applying a known phase retardation using an electrooptic crystal. This is only a factor of 4 above the quantum noise limit and corresponds to a theoretical lower limit of detection for ammonia of 6.6 ppt in 1 s as determined by vibrational and electrical noise. Instrument Design. Our ammonia detector is illustrated in Figure 1. The interferometer can be described as a modified Jamin design22,23 and is contained in an airtight Plexiglas box. Sampled air is pumped through a straight glass measuring inlet, into an open-ended glass sample tube and through the box at about 2 L/min. The length of the sample tube is 38 cm, and the interior diameter is 0.9 cm, giving a sample volume of 24 cm3. The sample residence time at 2 L/min sample flow is 0.7 s. Measurements of noise due to turbulence as a function of flow indicate that the sample flow may be increased by a factor of 2. A 1-mW single-mode He-Ne laser serves as the probe laser for the interferometer. The interferometer is defined by two special etalons, oriented at 45° to the incoming probe laser beam. Each etalon has a totally reflective rear surface and is faced with a coating that is split vertically into antireflective and 50% reflective sides. The He-Ne beam is introduced into the Plexiglas box through a focusing lens and is split by the front etalon into two beams of approximately equal power. These signal and reference beams are recombined after 50 cm at the rear etalon, and the interferometer output is detected by two photodiodes. The simple geometry of the unfolded Jamin interferometer allows for maximum heating along the He-Ne beam path, and

the etalons are mounted on stainless steel blocks connected by Invar rods for maximum stability. To correct for unmodulated temperature drift, the static phase difference between the signal and the reference beam is maintained at zero by an analog integrator24 that sends a correcting voltage to a piezoelectric transducer mounted on the rear etalon. The integrator time constant of 0.5 s is fast enough to maintain the zero but does not cause oscillation of the signal. Once the interferometer is aligned and balanced, it will stay this way for months. The tuned CO2 laser beam is reflected into the interferometer by two gold-coated mirrors, one of which is parabolic and increases the power density by focusing the CO2 beam to 1 mm in the middle of the interferometer arm. The typical power output of the CO2 laser tuned to 9.22 µm is 7 W. The CO2 laser is modulated at 1.2 kHz, corresponding to a quiet spot in acoustical noise observed at this frequency during tests using an electrooptic crystal. The CO2 beam travels through a BaF2 window into the box and is reflected between two internal gold-coated steel mirrors with small holes to pass the He-Ne beam. The CO2 beam is aligned so it is almost collinear with the signal He-Ne beam and coaxial with the sample cell. Although it may appear that the use of metal mirrors with holes is an awkward way of coaligning the CO2 and He-Ne laser beams, gold is highly reflective and metal has efficient thermal conduction properties. Metal mirrors with holes used in this way produce less background signal than optics fabricated from Pyrex, germanium, or zinc selenide used in a reflective/transmissive mode for excitation and probe laser beams.25 The beam interaction path length, l in eq 4, depends on

(22) Lee, W.-K.; Gungor, A.; Ho, P.-T.; Davis, C. C. Appl. Phys. Lett. 1985, 47, 916-918. (23) Mazzoni, D. L.; Davis, C. C. Appl. Opt. 1981, 30, 756-764.

(24) Horowitz, P.; Hill, W. The Art of Electronics; Cambridge University Press: New York, 1989; p 222. (25) Davis, C. C.; Petuchowski, S. J. Appl. Opt. 1981, 20, 2539-2554.

Analytical Chemistry, Vol. 71, No. 7, April 1, 1999

1393

the relative alignment between the He-Ne and CO2 laser beams; maximum coalignment optimizes the signal. The maximum value of l is 38 cm, corresponding to the full length of the sample tube. This parameter is the most likely to cause variation in the detector sensitivity because it depends on skill at aligning the CO2 beam through the detector. The CO2 beam exits the box through a second window, where the power throughput and frequency tuning can be monitored efficiently by dumping the beam into a small photoacoustic cell filled with 2 Torr of NH3 buffered with 150 Torr of N2. There are several potential particle interferences to any gasphase ammonia measurement. Spurious signal from particle scattering or particle decomposition by absorption of the pump laser energy might cause optical signal interference. Pinnick et al.26 have measured the breakdown threshold for micrometer-sized particles containing water vapor as 3 J/cm2 with a CO2 laser. We do not anticipate any significant interference from either of these optical effects. The energy density of our laser is 0.07 J/cm2, well below the particle breakdown threshold, and any signal from particle scattering is not likely to be detected by the photodiodes as a modulated phase difference at the appropriate phase shift for ammonia detection. A particular concern for any gas-phase ammonia measurement is the decomposition of labile ammonium nitrate particles in sampling inlets and cells27-29 and NH3 adsorption or desorption from the walls of the inlet and sample cell. We have minimized this possibility by thoughtful detector design. Our sample inlet is straight glass, residence time in the inlet and sample cell of the photothermal ammonia detector is brief, (about 1 s), and sampling is done at ambient temperature and pressure. Direct tests for ammonium nitrate decomposition are beyond the scope of this initial work, but we are currently planning a series of tests for particle interference, and we hope to publish these results in a future work. The sample airstream is switched between measure and zero modes through a glass three-way valve. Air is scrubbed of ammonia by passing through a glass tube filled with H3PO4-coated Teflon rings. These rings increase the scrubbing surface without creating a large pressure drop. Electrodes for Stark shifting the ammonia absorption have been added to the sides of a sample cell for alternative NH3 signal zeroing. Calibration Methods. The ammonia detector is calibrated by dynamic dilution of two ammonia permeation tube outputs and by standard addition of high-concentration ammonia gas. Gasphase concentrations are expressed as volume or molar mixing ratios. Mixing ratios will be written as ppm, molecules of a particular gas per 106 molecules of air; ppb, molecules per 109 molecules of air; or ppt, molecules per 1012 molecules of air. All pieces of plumbing used for the dilution and delivery of ammonia to the photothermal ammonia detector are made of passivated Pyrex tubing. After fabrication, the glass is carefully cleaned and rinsed out with a solution of weak NaOH. We minimize NH3 sample loss by wall absorption by using the smooth, (26) Pinnick, R. G.; Biswas, A.; Armstrong, R. L.; Jennings, S. G.; Pendleton, J. D.; Fernandez, G. Appl. Opt. 1990, 29, 910-925. (27) Huebert, B. J.; Luke, W. T.; Delany, A. C.; Brost, R. A. J. Geophys. Res. D 1988, 93, 7127-7135. (28) Hildemann, L. M.; Russell, A. G.; Cass, G. R. Atmos. Environ. 1984, 18 (9), 1737-1739. (29) Makar, P. A.; Wiebe, H. A.; Staebler, R. M.; Li, S. M.; Anlauf, K. J. Geophys. Res. D 1998, 103, 13,095-13,110.

1394 Analytical Chemistry, Vol. 71, No. 7, April 1, 1999

Figure 2. The output from one of three ammonia standards described in the text combined with scrubbed diluent air in a glass mixing volume. A glass frit is added to the end of the diluent line so that turbulence in the diluent air is buffered from the photothermal detector sample cell. A vent at the far end of the mixing volume maintains the sample pressure at one atmosphere. A glass threeway valve is used to switch the sample between zero and measure. The signal rise time between zero and measure is much faster with two separate inlet paths for the scrubbed and unscrubbed air because hysteresis from the inlet tubing is avoided.

clean, nonporous surface that glass provides, and we avoid any contact between sample air containing ammonia and materials such as Teflon and stainless steel that have been shown to remove ammonia from the airstream. Compressed breathing air is the diluent used for the majority of the experiments. The diluent is scrubbed of NH3 and dried using a scrubber fabricated by layering molecular sieve 4A and loosely packed H3PO4-coated glass wool inside a drying tube. All plumbing between the air cylinder and the mixing chamber is clean Teflon tubing. The permeation tube output rates are 22 and 1.0 ng/min, and the tubes are maintained at 35.0 ( 0.1 °C with a small amount of diluent air flowing continuously over them. The outputs from the permeation tubes flow through permanently attached 0.32-cm PFA Teflon tubing that is connected directly to the mixing volume during calibration. Permeation rates are verified by NH3 conversion to nitric oxide on hot platinum foil followed by chemiluminescent NO detection. Permeation rates are also monitored by careful weighing and calculation of the mass of ammonia lost by the permeation tubes as a function of time. Alternatively, the output from a high concentration, 11.5 ppm, ammonia gas cylinder is attached to a mixing chamber. A mass flow controller and Teflon tubing is used to deliver gas from the cylinder to the mixing chamber, but ammonia loss in this plumbing is negligible because the ammonia concentration is so large. With the appropriate gas standard attached, changing the diluent airflow using the mass flow controllers varies dilution. The calibration system is illustrated in Figure 2. The calibration system includes an optional flow stream through a distilled, deionized water bubbler for detector sensitivity tests in humidified air. The flows of dry and wet diluent air are measured with separate mass flow controllers, and the flow

Figure 3. Logarithmic plot of the ammonia calibration curve. The ammonia detector was calibrated with ammonia concentrations between 11.5 ppm and 216 ppt. The dotted line represents the leastsquares best fit to the data; each data point weighted by the inverse square of the standard deviation in the measurement. The slope is 10.9 µV/ppb with a standard deviation of 0.1 µV/ppb, an error of about 1%.

through the bubbler is shut off with a separate valve for dry measurements. The dewpoint of sampled humidified air is verified with a thermoelectric dewpoint hygrometer. RESULTS AND DISCUSSION This section describes extensive testing of the ammonia detector in the laboratory. The ammonia detector is routinely calibrated with low ammonia concentrations (250 ppt-15 ppb), produced by diluting output from the permeation tubes with scrubbed compressed air. The instrument signal is linear with CO2 laser power output and ammonia concentration, as expected from theory. The slope of a typical calibration curve is 16.1 ( 0.6 µV of signal per ppb NH3 with an intercept of 4.2 ( 4.1 µV. The instrument precision, defined here as (2σ of the ammonia measurement, was determined by measuring the standard deviation of the signal from 700 ppt ammonia, generated as described in the calibration section. Instrument precision measured with a 1-s lock-in amplifier time constant and no additional signal averaging was 250 ppt. With a lock-in time constant of 100 s, the instrument precision was measured as 31 ppt. The ammonia concentration inside the detector was varied between 11.5 ppm and 216 ppt by diluting three ammonia standards to experimentally verify the linear dynamic range. High concentrations were generated from a cylinder of 11.5 ppm ammonia gas, with an uncertainty of 10%. The ammonia permeation tube outputs have been verified as described in the calibration section and have an uncertainty of less than 5%. The results from this test are shown as a logarithmic plot in Figure 3 so that the smaller concentrations can be resolved. The instrument was linear over more than 5 orders of magnitude without adjustment of the sampling time or alteration of the detection electronics and did not show a significant memory effect due to desorption of ammonia from the glass even though the concentration was rapidly varied from ppm to ppt. Because the system is linear over such a large range, the detector can be calibrated by standard addition using relatively large concentrations of NH3 that are easier to handle. The detector signal is linear

for concentrations of ammonia throughout the range of concentrations likely to be observed in the natural atmosphere, bounded on the lower end by the detection limit. Ammonia Signal Interference. We have investigated sources of modulated temperature changes that may produce ammonia signal interference. One of these is a background signal from CO2 laser heating of internal mirrors in the interferometer. The size of this background varies with CO2 beam alignment through the interferometer but is small when the beam is properly aligned and the mirror coating is not scratched. Because there is a phase delay between the mirror and ammonia signals, the mirror background can be minimized by tuning the phase offset on the lock-in amplifier. Two major components in air, CO2 and H2O, were tested to determine the amount of signal interference they produce by absorbing light at 9.22 µm. The atmospheric trace gases CO, CH4, N2O, NO, NO2, HNO3, SO2, and O3 are unlikely to interfere with ammonia detection at 9.22 µm because none of these molecules has a significant absorption at this wavelength. Other atmospheric trace compounds that may interfere with ammonia measurement at 9.22 µm include alcohols, amines, and nitriles. In the ammonia detector, CO2 laser energy causes heating through ammonia absorption in the overlapping spectral region between CO2 and ammonia transitions. The CO2 laser light will also excite CO2 in sampled air. This absorption is the inverse of the laser transition, and only the fraction of CO2 in the lower laser energy level, 02°0, will absorb the 9.22-µm light. The relative population distribution between the ground state and 02°0 in the sampled air is described by a Boltzmann distribution. At T ) 298 K, the fraction of CO2 that is in the 02°0 vibrational state and can therefore absorb the laser energy is 0.2%. Absorption of the laser energy excites this fraction of CO2 in the sampled air to the CO2 (00°1) vibrational state. The energy difference between the CO2 (00°1) and N2 (ν ) 1) vibrational states is only 18 cm-1 and can be easily transferred by collisions. This near-resonance, exploited to make CO2 lasers, causes both N2 and CO2 in air to absorb energy from a CO2 laser beam. The average residence time of excited N2 is longer than the residence time of excited ammonia, causing CO2 relaxation to be out of phase with the more rapid ammonia signal.30 The response of our ammonia detector to varying concentrations of CO2 was measured by diluting a cylinder of CO2 laser gas containing 12.6% CO2 in N2 and He to CO2 concentrations between 0 and 670 ppm with compressed air passed through a NaOH scrubber to remove CO2. The results are shown in Figure 4. The measured phase lag between peak ammonia and CO2 signals was 70°. Our results are consistent with Olafsson et al.,31 who noted a phase delay between the CO2 and ammonia signal while measuring ammonia in power plant emissions with a large amount of CO2. In ambient air, the amount of CO2 is much smaller, typically about 360 ppm. The addition of 6 ppb ammonia to the sample with mixing ratios of 0, 325, and 581 ppm of CO2 produced an average signal change of -78.7 µV (2%. Ammonia produces a negative offset to the signal at this phase adjustment because it was optimized for CO2. This represents the maximum possible (30) Wood, A. D.; Camac, M.; Gerry, E. T. Appl. Opt. 1971, 10, 1877-1884. (31) Olafsson, A.; Hammerich, M.; Bulow, J.; Henningsen, J. Appl. Phys. B 1989, 49, 91-97.

Analytical Chemistry, Vol. 71, No. 7, April 1, 1999

1395

Figure 4. Signal variation with CO2 concentration. CO2 laser gas is diluted with NaOH-scrubbed compressed air. The signal is measured at the phase offset with maximum CO2 signal. This phase is independent of varied CO2 concentration. The signal is linear with CO2 concentration, with a slope of 0.615 µV/ppm of CO2. The dotted line in the figure is the linear best fit to the signal. The filled circles represent the signal with the addition of 6 ppb NH3. The difference between the NH3 + CO2 and CO2 signals is negative at this phase offset because the signal has been optimized for CO2 detection.

Figure 5. Signal from 7.2 ppb ammonia measured while varying the relative humidity of diluent air. Triangles are the signal from ammonia at each value of relative humidity. Stars are the signal through the ammonia zero, due to water vapor. The solid line is the least-squares best fit to the data with a slope of 4.9 µV per % relative humidity, and an intercept of 6.1 µV. Error bars represent (2σ of the measurement at each data point. The signal from 7.2 ppb ammonia is 92.2 ( 7.6 µV, plotted as a dotted line in the figure.

When the ammonia detector was calibrated with dry air and with

humidified air, the difference in slopes was 2.5%, within the uncertainty of the measurement. Water vapor in ambient air is problematic for many methods of ammonia detection, as described in the introduction. Even though our rejection ratio for water is almost 6 orders of magnitude, the signal from water vapor is still a concern because of high concentrations and variability in the atmosphere. Because sensitivity of the ammonia detector does not change with relative humidity, the problem is limited to a background correction. This creates a potential problem for making fast-response flux measurements because the instrument will record water vapor as well as ammonia fluctuations unless a continuous zeroing method is used. This issue will be discussed further in the following section on zeroing methodology. The large ammonia absorption that coincides with the 9.2-µm CO2 laser transition is in the R-branch of the ν2 vibrational transition of ammonia. Methanol and methylamine have welldefined vibration rotation transitions with R-branch maximums at 9.5 µm. The absorptions at 9.2 µm of both these molecules are in the broadened wings of these branches and are quite small. Acetonitrile has a vibration rotation transition centered at 9.7 µm that may include a rotation line at 9.2 µm and, therefore, may be a larger interference. Although future work will include the careful measurement of absorptions by these molecules at 9.22 µm, none is likely to be present in the atmosphere in large quantities. The degree of interference from methylamine and acetonitrile, although tested by spiking during an extensive intercomparison of ammonia measurement techniques,11 is not determined conclusively for the PF/LIF instrument, tungsten oxide, molybdenum oxide, or citric acid denuders, or the oxalic acid filter pack. The atmospheric concentration and distribution of methylamine has not been established, but several authors suggest that methylamine concentrations are probably significantly smaller than ammonia concentrations.17,33 Acetonitrile is produced during biomass burning and may represent as much as 1% of volatilized

(32) Hinderling, J.; Sigrist, M. W.; Kneubuhl, F. K. Int. J. Infrared Millimeter Waves 1986, 7 (4), 683-713.

(33) Langford, A. O.; Goldan, P. D.; Fehsenfeld, F. C. J. Atmos. Chem. 1989, 8, 359-376.

CO2 signal interference. The minimum measured rejection ratio between CO2 and ammonia is

RCO2 )

13.1 µV/ppb NH3 6.15 × 10-4 µV/ppb CO2

) 2.1 × 104

(6)

The gain from ammonia in the signal is independent of CO2 concentration at a fixed phase offset, so that variability in CO2 does not affect the net signal from ammonia. Tuning the phase delay on the lock-in amplifier to minimize CO2 and mirror signals optimizes the ammonia signal and doubles the rejection ratio. CO2 laser light is absorbed by water vapor in the broad continuum formed by collisional broadening in the far wings of several strong rotational transitions in the infrared and from the rotational transitions of water dimers.32 Bubbling through deionized, distilled water humidified the diluent air for tests to establish whether water vapor concentration effects the ammonia signal gain of our PTI ammonia detector. The results from varying the relative humidity of the diluent gas with 7.2 ppb ammonia are shown in Figure 5. The signal was linear with increasing relative humidity, with no phase delay between the ammonia and water vapor signals. The sensitivity of the instrument to water vapor, measured through the ammonia sample zero inlet, was 4.9 µV per percent relative humidity. At the experimental temperature and pressure, this sensitivity can be expressed as 158 µV per percent water vapor. The rejection ratio for water vapor is quite high, measured here as

RH2O )

12.8 µV/ppb NH3 1.58 × 10-5 µV/ppb H2O

) 8.1 × 105

1396 Analytical Chemistry, Vol. 71, No. 7, April 1, 1999

(7)

biomass nitrogen in fires34 while ammonia is estimated as 4%. Interference from acetonitrile must therefore be quantified before our ammonia detector can be used to measure ammonia in fires. Methanol and other alcohols are not likely to be present in quantities above a few ppb, even in polluted air. Preliminary tests by spiking the ammonia detector with methanol and 2-propanol indicate that alcohols pass through the ammonia scrubber and therefore the signal from alcohols will be subtracted during the ammonia zero. Although further testing is warranted, the relatively modest signal from substantial concentrations of alcohols shows that these alcohols are unlikely to be a significant interference with the detector as configured. Signal Zeroing Methodologies. In our detector, tuning the lock-in detector phase lag to maximize the ammonia signal can minimize background signal from the internal mirrors and CO2. Water vapor does create a background signal, despite a rejection ratio for water of almost 106 as compared to ammonia. This background may be substantial when the air is moist and ambient ammonia concentrations are very small. The background signal can be subtracted in several ways. Water vapor absorption is mainly a continuum in the 9-10-µm wavelength region, with small absorption lines that are far removed from the large ammonia absorption at 9.2 µm. Absorption by CO2 is correlated with the CO2 laser power output, and the linear CO2 molecule does not exhibit a shift in absorption frequency with an applied electric field. The ammonia molecule does have a second-order Stark shift. To take advantage of these effects, the CO2 laser can be tuned away from the ammonia absorption line or an electric field can be applied to induce an ammonia absorption frequency shift. We have been using a chemical zero, which removes ammonia from the airstream but passes H2O, CO2, and alcohols. The advantage of using a chemical zero is that the method is very simple. However, a practical limit for switching between detector modes using the chemical zero as configured is about 30 s. This limits the true response time because water vapor as well as ammonia may fluctuate during the measuring part of the cycle. The advantage of a spectroscopic zero over a chemical zero is that the sample flowstream does not have to be switched between two inlets. A disadvantage of frequency shifting the laser is that the laser power must stabilize at each frequency, and the signal must be normalized to the laser power output, which will not be constant. The off-line frequency must be chosen so that the CO2 absorption strength is equal to CO2 absorption at the ammonia detection frequency, or CO2 will not be subtracted accurately. There is a practical limit to the speed at which the laser frequency can be switched. de Vries et al.35,36 reported this limit as 15 s. This does not solve the problem of water vapor fluctuations during ammonia measurement and is not significantly faster than the chemical zero. We have tested Stark-shifting the ammonia signal using an applied electric field as an alternative to the chemical zero. Electrodes were added to the outside of a 38-cm-long, 1-cm-wide (34) Lobert, J. M.; Scharffe, D. H.; Hao, W. M.; Crutzen, P. J. Nature 1990, 346, 552-554. (35) de Vries, H. S. M.; Dam, N.; van Lieshout, M. R.; Sikkens, C.; Harren, F. J. M.; Reuss, J. Rev. Sci. Instrum. 1995, 66 (9), 4655-4664. (36) de Vries, H. S. M.; van Lieshout, M. R.; Harren, F. J. M.; Reuss, J. Infrared Phys. Technol. 1995, 36 (1), 483-488.

square glass sample cell. The electrodes were placed on the outside of the glass cell because of the affinity of ammonia for metals. Separation of the electrodes was 1 cm, and applied voltage was limited by severe arcing observed at field densities over 950 V/mm at atmospheric pressure. A Stark signal shift of 7.9 mV was measured for 507 ppb ammonia with 925 V/mm of applied field density during our preliminary tests. This correlates to 70% of the total ammonia signal, substantial for a Stark shift at atmospheric pressure. There was only a small amount of electrical or vibrational noise added to the interferometer signal by the applied electric field, and there did not appear to be any appreciable interference from ozone. Sauren and Bicanic37 suggested that the magnitude of the Stark signal varies as a function of water vapor. They attributed this effect to NH4OH formation or rotational line broadening. We did not observe any variation of the Stark signal with water vapor, but we note that they have placed metal electrodes inside their sample cell and an alternative explanation is a variation of ammonia loss on the metal with varying water content of the air. Modulating the ammonia signal by ac Stark shift warrants further investigation because it may be a powerful tool for continuous selectivity of ammonia measurements. However, it has not been demonstrated that the technique is suitable for measuring very small amounts of ammonia in air because of the potential for molecular ion formation and variability in the applied field strength. We note that even 3% variability is intolerable for precise measurements of small amounts of ammonia, a stringent requirement for fields this large. A small modulation on top of a bias voltage in the region where the signal is changing rapidly with the applied field may be most feasible for Stark modulation of the signal. Instrument Response Time. Development of a detector with rapid response time is a major goal of this research. As with all in situ spectroscopic techniques, detector response to a change in ammonia concentration is limited by signal integration time, residence time in the sampling volume, and adsorption or desorption of ammonia on sample cell walls. Although the frequency with which ambient ammonia fluctuations can be resolved is also limited by H2O background subtraction, it is useful to test whether hysteresis is likely to be a limiting problem. Instrument response to a large step change of ammonia was measured by standard addition of 6.7 ppm ammonia in compressed air to a 2 L/min flow of room air containing less than 20 ppb ammonia. The ammonia source was attached by a very short piece of flexible Tygon tubing to a glass T fastened directly to the measuring inlet. One arm of the glass T was open to admit room air or to serve as a vent. A clamp attached to the Tygon tubing was rapidly closed to stop the ammonia flow, simulating a step change in ammonia concentration. The time constant on the lockin amplifier was set to 0.1 s, and the data were recorded at 1 kHz. A representative step change is shown in Figure 6. The signal decay was fit to a function of the form

S(t) ) A0 exp(-t/A1) + A2

(8)

A0 is the signal from the initial ammonia concentration and A2 is (37) Sauren, H.; Bicanic, D. Anal. Instrum. 1992, 20 (1), 63-78.

Analytical Chemistry, Vol. 71, No. 7, April 1, 1999

1397

Figure 6. Signal vs time for a step change of 6.7 ppm. The ammonia source was clamped and the signal decay observed with signal sampling frequency of 1 kHz and a 0.1-s time constant on the lock-in amplifier. The time for the signal to decay by a factor of e-1 was 1.3 s, including the signal spike caused by shutting off the ammonia source.

the value of the signal as t f ∞. The time constant for the system is equal to A1, calculated as 0.97 s from the fitted equation. The fit did not include four data points during the spike, which is an artifact of the brief pressure change caused by clamping the tubing. Although we plan to conduct future tests at lower ammonia concentrations, our initial results are encouraging, and we note that during linear dynamic range testing, the ammonia concentration was rapidly decreased by 5 orders of magnitude without visible hysteresis effects. The fastest possible instrument response based on sample residence time is 5 Hz. Further reduction of the sample cell residence time is limited by turbulence at higher pumping speeds. The factor limiting the measurement of rapid ammonia fluctuations in the atmosphere, for this as well as other photothermal techniques, is a continuous, rapid subtraction of the water vapor background. The true ammonia response time can only be as fast as this background subtraction. Further investigation of the ac ammonia Stark shift, and Zeeman shifting and laser frequency modulation, are the focus of current research. Comparison with Related Techniques. Several groups have developed photothermal detection techniques that we will compare to PTI. Rooth et al.,38 Sauren et al.,39,40 Moeckli et al.,41 and Thony and Sigrist42 detected pressure waves (sound) from modulated infrared absorption by ammonia using microphones in photothermal acoustic cells (PTA). de Vries et al.35,36,43 and Zimerman and Boccara44,45 detected the photothermal deflection (PTD) of a He(38) Rooth, R. A.; Verhage, A. J. L.; Wouters, L. W. Appl. Opt. 1990, 29, 36433653. (39) Sauren, H.; Bicanic, D.; Hillen, W.; Jalink, H.; van Asselt, K.; Quist, J.; Reuss, J. Appl. Opt. 1990, 29, 2679-2681. (40) Sauren, H.; Gerkema, E.; Bicanic, D.; Jalink, H. Atmos. Environ. 1993, 27A (1), 109-112. (41) Moeckli, M. A.; Fierz, M.; Sigrist, M. W. Environ. Sci. Technol. 1996, 30, 2864-2867. (42) Thony, A.; Sigrist, M. W. Infrared Phys. Technol. 1995, 36, 585-615. (43) de Vries, H. S. M.; Harren, F. J. M.; Wyers, G. P.; Otjes, R. P.; Slanina, J.; Reuss, J. Atmos. Environ. 1995, 29 (10), 1069-1074. (44) Zimering, B.; Boccara, A. C. Rev. Sci. Instrum. 1996, 67 (5), 1891-1895. (45) Zimering, B.; Boccara, A. C. Appl. Opt. 1997, 36, 3188-3194.

1398 Analytical Chemistry, Vol. 71, No. 7, April 1, 1999

Ne probe beam by a modulated CO2 laser beam. An advantage of all three infrared photothermal techniques, PTI, PTA, and PTD, is that carbon dioxide lasers available to generate infrared light are relatively inexpensive and maintenance-free and have low power requirements. Common to all of the photothermal techniques is the necessity of removing a background signal caused by H2O and CO2 absorption of the CO2 laser radiation. Without a continuous zero, subtraction of the background signal limits the speed of true response measurements of ammonia to about 15 s,38,39 the time it takes to tune the CO2 laser between two lines. The PTD groups listed above have measured the sensitivity to ethylene at high concentrations and have extrapolated these measurements to estimate a detection limit for ammonia of 500 ppt in 1 s.34-36,44,45 Ammonia measurements by PTD are shown to compare well with a continuous-flow denuder technique above 10 ppb.36,43 Open-path detection using PTD has the advantage of avoiding sample line hysteresis effects or surface-based particle decomposition, both tricky problems for all ammonia detection techniques that require sample flow. However, measured sensitivity to sub-ppb levels of ammonia has not been reported for PTD, and the open-path configuration makes calibration at subambient NH3 concentrations difficult. Sensitivity extrapolated from detector noise is 3.7 × 10-10 cm-1 for PTI, compared to 6.0 × 10-9 cm-1 reported for the PTD technique.35 Both PTI and PTD techniques compensate for air turbulence using a double beam, but the field sensitivity of both instruments may be somewhat by turbulence noise. PTI is less sensitive to low-frequency mechanical vibrations and pointing noise in the laser because PTI sensitivity is sufficient to allow CO2 laser modulation at higher frequency (1.2 kHz for PTI compared with 20-50 Hz for PTD). PTA detectors built by Sauren et al.,37,39,40 Rooth et al.,38 and Moeckli et al.41,42 do not as yet have detection limits below 1 ppb and require signal integration or zeroing times between 40 s and several minutes. A common problem with resonant PTA is resonant frequency noise and drift with water vapor concentration and temperature, which is avoided by PTI detection at a quiet, nonresonant frequency. Electric field modulation has been used with PTA by Sauren et al.,39,40 with detection at the combined laser and Stark modulation frequency. A disadvantage of measuring these sidebands of the mixed signal is that the signal voltages are very small for concentrations of ammonia in the low-ppb range, less than 1 µV.39 A diode laser (λ ) 1.55 µm) is used as the ammonia excitation laser in the resonant photoacoustic scheme of Miklos and Feher.46 They reported sensitivity limited to the ppb range (8 ppb) but used the promising technique of diode laser frequency modulation (FM) to dither the laser rapidly on and off the ammonia absorption line. We intend to try a similar FM scheme with our CO2 laser and PTI detection scheme to separate the discrete ammonia absorption line from the water vapor continuum. CONCLUSIONS We have demonstrated careful calibration and measured instrument precision for photothermal interferometric ammonia detection below 1 ppb. The 2σ detection limit of our interferometric ammonia detector is 31 ppt in 100 s or 250 ppt in 1 s. The wide linear dynamic range allows sampling over a large range of (46) Miklos, A.; Feher, M. Infrared Phys. Technol. 1996, 37, 21-27.

ambient ammonia concentrations without adjustment to the electronics or to the sampling time. Signal sensitivity is sufficient to allow modulation and signal detection at about 1 kHz, adjustable to match local environmental noise minimums. Our detection limit is comparable to, or lower than, other existing ammonia measurement techniques, except for PF/LIF. Our photothermal interferometer is relatively small and simple to operate, and because we require only low-power visible and infrared radiation, it is less expensive and less dangerous and requires less power than the PF/LIF technique. Engineering improvements and future experiments planned or in progress to the PTI ammonia detector will likely improve the sensitivity and response time. We plan to focus our development on true fast-response ammonia measurements by developing alternative continuous signal zeroing techniques and to improve the sensitivity by upgrading some of our equipment. Explicit testing of detector response to ammonia with ammonium nitrate particles and very high concentrations of water vapor in the sample is also planned. Members of the atmospheric chemistry community generally agree there is a need for simple, but fast and sensitive methods for measuring gas-phase ammonia. Our work indicates the interferometric ammonia detector can make useful, accurate ammonia measurements, with a competitive combination of size, speed, and sensitivity. ACKNOWLEDGMENT We are grateful for partial funding for this project from the NSF-sponsored Center for Clouds, Chemistry and Climate, and from NOAA/ARL’s Cooperative Institute for Climate Studies. The authors thank members of the Laser Sensor Lab and the Air Chemistry Lab at The University of Maryland for help and advice and the University of Maryland Electrical Engineering Department Shop for technical support. We also thank two anonymous reviewers for comments that improved this presentation of our results.

10-W grating-tuned CO2 laser

Synrad model 48G-2

laser chiller

Neslab model RTE 110D

laser power supply

Lambda model LRS-56-28

laser spectrum analyzer

Optical Engineering, Inc.

chemiluminescent NO detector

TeCo model 14B

mass flow controllers

Datametrics, Tylan, various models

temperature controller

Omega model 4201A-PC2

lock-in amplifier

Stanford Research Systems model SR5110

PZT feedback amplifier

Burleigh model PZ-300M

frequency spectrum analyzer

Hewlett-Packard model HP-35665A

computer

286 motherboard, components by various manufacturers

etalons

Virgo Optics

digital oscilloscope

Hewlett-Packard model 54501A

Chemical Reagents compressed air

Air Products medical grade breathing air

anhydrous ammonia

Air Products ultrahigh purity

ammonia permeation tubes

Vici Metronics low-output LW series

Drierite

8-mesh 97% CaSO4, 3% CoCl2, Hammond Drierite Co.

phosphoric acid

86.5%, Baker Analyzed

molecular sieve

type 4A, 8-12-mesh beads, Davison

methanol

absolute, 100.0% assay, Baker Analyzed

2-propanol

99.9% assay, Fisher Scientific

cleaning solution

Micronox

APPENDIX A: EQUIPMENT AND CHEMICAL REAGENTS

sealing compound

Apiezon

Equipment List

Received for review July 24, 1998. Accepted November 22, 1998.

1-mW single-mode HeNe laser

Spectra-Physics model 4201A-PC2

AC980810H

Analytical Chemistry, Vol. 71, No. 7, April 1, 1999

1399