Absolute Steady-State Thermal Conductivity Measurements by Use of a Transient Hot-Wire System

A transient hot-wire apparatus was used to measure the thermal conductivity of argon with both steady-state and transient methods. The effects of wire diameter, eccentricity of the wire in the cavity, axial conduction, and natural convection were accounted for in the analysis of the steady-state measurements. Based on measurements on argon, the relative uncertainty at the 95 % level of confidence of the new steady-state measurements is 2 % at low densities. Using the same hot wires, the relative uncertainty of the transient measurements is 1 % at the 95 % level of confidence. This is the first report of thermal conductivity measurements made by two different methods in the same apparatus. The steady-state method is shown to complement normal transient measurements at low densities, particularly for fluids where the thermophysical properties at low densities are not known with high accuracy.


Introduction
The transient hot-wire system has been accepted widely as the most accurate technique for measuring the thermal conductivity of fluids over a wide range of physical states removed from the critical region. However, one of the drawbacks of this system is the need for increasingly larger corrections for the finite wire diameter and the outer boundary in the limit of zero fluid density. This effectively establishes a lower limit in pressure, at approximately 1 MPa for argon (corresponding to 28 kg m Ϫ3 in density), where the uncertainty in thermal conductivity increases dramatically. If transient measurements are made on gases below 1 MPa, the thermal conductivity is generally higher than the best theoretical estimates. Below 1 MPa, the linear region in the temperature rise vs the logarithm of time is greatly reduced or no longer exists for transient hot-wire measurements of gases. This curvature in the transient temperature rise is due to extremely large effects of the correction for finite physical properties of the wire at short times, and to the penetration of the transient thermal wave to the outer boundary at longer times, as the thermal diffusivity increases significantly in the limit of zero density. At low densities, the magnitudes of these corrections (comparable to the measured temperature rise itself) make it almost impossible to obtain an accurate mathematical description of the observed transient heat transfer in the hot-wire cells.
To overcome these difficulties, researchers often extrapolate measurements along an isotherm from higher densities to obtain the thermal conductivity of the dilute gas. The dilute-gas thermal-conductivity data obtained by such an extrapolation procedure have significantly more uncertainty than the data used in the extrapolation. Near the critical temperature the critical enhancement contributes significantly to the total thermal conductivity even at relatively low densities. This introduces curvature in thermal conductivity isotherms and makes the extrapolation to the dilute gas limit even more uncertain for isotherms near the critical temperature. Furthermore, the 1 MPa restriction makes the transient hot-wire instruments inappropriate for measuring thermal conductivity in the vapor phase at temperatures where the vapor pressure is below 1 MPa. Inconsistencies in dilutegas thermal conductivities obtained by various researchers using transient hot wires have seriously weakened the credibility of the technique.
These problems increase the relative uncertainty at the level of 95 % confidence of the thermal conductivity obtained with the transient hot-wire technique from 0.3 % for measurements at higher densities to about 2 % at the lower densities, which are the focus of the present work. This 2 % relative uncertainty is comparable to the relative uncertainty of measurements obtained with accurate steady-state instruments. The largest uncertainty in steady-state measurements is due to fluid convection and this is known to decrease dramatically in the limit of zero density. Corrections to steady-state measurements actually decrease and become negligible in this dilutegas region where transient measurements encounter their most serious difficulties.
At low densities the transient mode of heat transfer occurs at extremely short real times, where the wire heat-capacity correction is still quite large. This is because the thermal diffusivity of the gas is very large at low densities. This fast approach to steady state at low gas densities is an advantage for steady-state measurements using the same wire geometry. This paper examines the possibility of using the steady-state mode of operation to obtain the thermal conductivity of the dilute gas which is consistent with the higher density data obtained with the transient mode. This would allow any hot-wire instrument to operate in a transient mode at higher densities and in a steady-state mode down to the dilute-gas limit.
Measurements at low density have been made on argon gas to test this concept. Argon was selected because the dilute-gas value can be evaluated from the secondorder Chapman-Enskog kinetic theory using the known pair-interaction potential. In addition, there are many accurate measurements of the thermal conductivity of argon available in this region using both steady-state and transient techniques. Each series of measurements was made over a wide range of applied powers and included a transient and a steady-state measurement at each power level. Both the transient and the steady-state measurements are compared with the best available predictions from kinetic theory and the other data from the literature. Agreement between measurements using both modes of operation demonstrates the validity of both techniques. Researchers with transient hot wire instruments can potentially select the optimum technique for a given fluid state.
The transient hot-wire systems at NIST are completely described in previous papers [1][2][3][4]. The apparatus for high-temperature measurements [4] was used in the present work. The transient measurements were of one-second duration, as is typical in our previous measurements. The major change was modification of the data-acquisition system to operate in a steady-state mode, at times up to 40 s. A composite picture of the voltage rises across the Wheatstone bridge, obtained in five transient runs made at the same power level but with different experimental times, is shown in Fig. 1. Time is shown on a logarithmic scale, with the experimental times ranging from 1 s to 40 s. Two linear segments of the voltage rises are clearly visible. The transient thermal conductivity is obtained from the linear portion of finite slope which is proportional to the logarithm of time, while the steady-state thermal conductivity is calculated from the horizontal portion.
In Fig. 2, a typical voltage rise for a measurement in argon gas is shown, at a pressure below 1 kPa (mild vacuum), with an experimental duration of 1 s. This run would normally be evaluated as a transient experiment, the thermal conductivity being obtained from the apparent linear portion between 0.05 s and 0.15 s. There are two reasons to make measurements under these conditions. First, one can show that there is a sufficient section of a horizontal straight line, even at times below 1 s, to obtain a reliable result for steady-state conditions. Secondly, it may be possible to extract values for axial conduction or end effects. In addition, this extreme example shows exactly the origin of the problem with the transient experiment at low densities. The linear region in the temperature rise vs. the logarithm of time is extremely limited, degrading the accuracy of the resulting thermal conductivity data. In other words, a valid constant slope cannot be extracted from such a curve for temperature rise.

Transient Mode
Since transient measurements are increasingly unreliable as pressure decreases below 1 MPa, all assumptions made in the application of the theory of the instrument must be carefully examined. The theory for transient hot-wire measurements is well developed [5,6], although proper application of the theory requires significant care and judgment. The hot-wire cells are designed to approximate a transient line source as closely as possible, and deviations from this model are treated as corrections to the experimental temperature rise. The ideal temperature rise ⌬T id is given by where q is the power applied divided by length, is the thermal conductivity of the fluid, t is the elapsed time, a = / C p is the thermal diffusivity of the fluid, is the mass density of the fluid, C p is the isobaric heat capac-ity of the fluid, r 0 is the radius of the hot wire, C = e ␥ = 1.781... is the exponential of Euler's constant, ⌬T w is the measured temperature rise of the wire, and ␦T i are corrections to account for deviations from ideal line-source conduction [5,6]. The two most significant corrections account for the finite radius of the wire and for penetration of the fluid temperature gradient to the outer cell wall. The finite wire radius produces the short-time temperature lag relative to the ideal model, as shown in Fig. 1. Penetration of the temperature gradient to the outer wall produces the transition from the linear transient region to the steady-state conduction mode, which is also shown in Fig. 1. It is apparent in Eq. (1) that the thermal conductivity can be found from the slope of the ideal temperature rise as a function of the logarithm of elapsed time. The thermal diffusivity can be found from the intercept of this linear function. The uncertainty of the thermal conductivity at a level of confidence of 95 % is obtained from a linear fit of the ideal temperature rise data to Eq.
(1) and is characterized by the parameter STAT . A STAT of 0.003, for example, corresponds to a reproducibility of 0.3 % for the reported thermal conductivity. The principal corrections to the ideal model account for the finite dimensions of the wire ␦T 1 , penetration of the expanding thermal wave to the outer boundary ␦T 2 , and thermal radiation ␦T 5 . The relative magnitude of each correction depends on the fluid properties and the elapsed time in the experiment. In this work we apply the standard corrections [5,6], including the outerboundary correction and the thermal-radiation correction for a transparent gas. The correction for the finite wire dimensions requires careful examination.
The hot wires have finite diameters and their specific heat introduces a temperature lag from the ideal line source model of Eq. (1). The temperature response of an infinitely long wire of finite radius r 0 is given [5] by where (u ) = w a 1/2 J 1 (r 0 u )J 0ͩ r 0 u ͙a w /aͪ Ϫ a 0.5 w J 0 (r 0 u )J 1ͩ r 0 u ͙a w /aͪ, and where J n is the Bessel function of the first kind with order n , Y n is the Bessel function of the second kind with order n . In Eqs. (2)-(4) the thermal conductivity of the wire is w and the thermal diffusivity of the wire is a w . Eqs. (2)-(4) are defined for any point in the wire, and it is the volume-averaged temperature from r = 0 to r = r 0 which is required to correct our experimental observed rises in temperature. The resulting correction, ␦T 1f , for the finite wire dimension is Although it is fairly simple to implement this full solution to correct the experimental temperature rise, it is the first-order expansion of this solution which has been recommended because of its simplicity [5,6]. This first-order expansion for a bare wire is [5,6] Before an approximation such as Eq. (6) can be used, the truncation error for the case of our relatively large 12.7 m platinum hot wire must be evaluated. Figure 3 shows the full solution for the transient temperature rise [Eqs.
(2)-(5)], along with the first-[Eq. (6)] and secondorder approximations [7], for the case of a 12.7 m platinum hot wire in argon gas at 300 K and 0.1 MPa. It is apparent in Fig. 3 that the truncation error is quite large for dilute gases, so the first-order approximation of Eq. (6) cannot be used to correct the data. Both the magnitude of the correction for physical properties of the finite wire and the associated truncation error can be minimized by using thinner hot wires. Table 1 shows how truncation corrections depend on the wire diameter for argon gas at 300 K and 0.1 MPa. Also included in Table 1 are the relative uncertainties in thermal conductivity resulting from a relative uncertainty of 10 % in the thermal diffusivity of the fluid.
In the limit of zero density, the transient thermal wave will penetrate to the outer boundary. The outerboundary correction, which accounts for penetration of the transient heat pulse [5,6], is given by where g are the roots of J 0 (g ) = 0 and r b is the radius of the outer boundary. In the limit of infinite time, Eq. (7) approaches the steady-state solution given below. The final corrections, which must be considered because of their increasing significance at low densities, account for compression work ␦T 3 and radial convection ␦T 4 . A recent analysis by Assael et al. [8] concludes that ␦T 3 and ␦T 4 must be considered simultaneously and that Fig. 3. Calculated temperature rise for finite physical properties of the wire for argon gas at 300 K and 0.1 MPa using first-order and second-order approximations as well as the full integral solution. Table 1. Uncertainties due to corrections for finite wire diameter for argon gas at 300 K and 0.1 MPa. First, differences are examined between the actual heat transfer from a finite diameter wire and the first and second order series approximations of this solution for various wire diameters. Second, the effect of a 10 % error in the thermal diffusivity a used for finite wire diameter correction is calculated for various wire diameters. This correction can introduce significant errors in thermal conductivity measurements with large wires if the thermal diffusivity or wire diameter is not well known.

Fluid
Wire  [5] is in error. Based on the work of Assael et al. [8] we have set both ␦T 3 and ␦T 4 equal to zero in the present analysis. At low densities, the thermal diffusivity of the fluid increases almost linearly with inverse pressure. Large corrections to the experimental temperature rise are required for both heat-capacity and outer-boundary effects because of this large thermal diffusivity. Any uncertainties in the wire diameter and the fluid properties used in these corrections become increasingly important at low densities. The full heat-capacity correction must be used to correct the measured temperature rises at low densities since this correction is so significant. The measurements at low density must be carefully examined to verify that the thermal wave has not reached the outer boundary since the thermal diffusivity increases so dramatically in this region.
Transient results obtained by use of the traditional corrections employed in earlier work [5,6] and the revised corrections as discussed above are shown for argon at 300 K in Fig. 4. The differences are primarily due to setting the compression work ␦T 3 equal to zero [8], use of the full heat-capacity correction of Eqs. (2)-(5), and careful restriction of the regression limits to exclude times where the outer-boundary correction ␦T 2 is significant. The results for all power levels were averaged at each pressure level in Fig. 4. It can be seen that the results at low densities are more linear in terms of density with the revised corrections (hook due to increasing contributions from the outer boundary correction), while the results at the higher densities are not changed appreciably. This linear dependence on density is expected from the kinetic theory of low-density gases.

Steady-State Mode
The working equation for the steady-state mode is based on a different solution of Fourier's law but the geometry is still that of concentric cylinders. The solution can be found in standard texts for the case of constant thermal conductivity (see, for example, Reference [9], page 114). This equation can be solved for the thermal conductivity of the fluid ; = q lnͩ r 2 where q is the applied power divided by length, r 2 is the internal radius of the outer cylinder, r 1 is the external radius of the inner cylinder (hot wire), and ⌬T = (T 1 Ϫ T 2 ) is the measured temperature difference between the hot wire and its surrounding cavity. For the concentric-cylinder geometry described above the total heat flux divided by length, q , remains constant and is not a function of the radial position. Assuming that the thermal conductivity is a linear function of temperature, such that = 0 (1 + b T ), it can be shown that the measured thermal conductivity is given by = 0 (1 + b (T 1 + T 2 )/2). Thus, the thermal conductivity that is measured corresponds to the value at the mean temperature of the inner and outer cylinders, where This assumption of a linear temperature dependence for the thermal conductivity is valid for experiments with small temperature rises. The density of the fluid assigned to the measured thermal conductivity is taken from an equation of state [10] using the temperature from Eq. (9) and the experimentally measured pressure. Equation (8) assumes that the dimensions of the wire and cavity are well known and that the wire is perfectly concentric with the outer cylindrical cavity. The diameter of our wire is 13.14 m and it is known with a relative uncertainty of 0.5 % at a level of confidence of 95 %. This uncertainty contributes a component of relative uncertainty of 0.07 % to the uncertainty of the measured thermal conductivity. Since it is nearly impossible to keep the wire perfectly concentric with the outer cavity, the uncertainty associated with the eccentricity of the wire with respect to the cavity must be assessed. For a wire that is eccentric, the thermal conductivity is given by where b is the distance between the wire's axis and the axis of the outer cylinder. The eccentricity correction is shown in Fig. 5 for a 12.7 m diameter wire in a 9 mm diameter cavity. The hot wires in the present cell are concentric with the cavity within 0.5 mm, so it is apparent from Fig. 5 that the relative uncertainty is about 0.2 % due to misalignment of the wire in the cavity. The combined relative uncertainty, due to both the diameter and eccentricity of the wire is 0.3 % in the measured thermal conductivity. While 40 s may seem to be a very short time in comparison to normal steady-state measurements, it still allows the very small wires used in transient hot-wire systems to equilibrate in the gas phase. The time of the steady-state experiments is restricted to 40 s since the temperature of the cell wall T 1 is assumed to be the initial cell temperature. Both transient and steady-state measurements of thermal conductivity should be made at several power levels. The thermal conductivity should be valid and free of convection if a plot of the measured values of thermal conductivity as a function of applied power is constant. The present measurements are made over a large range of applied powers, and the powers of the transient measurements overlap those used for the steady-state measurements as much as possible.

Data Reduction
Three isotherms were measured for gaseous argon at 300 K, 320 K, and 340 K. There were 13 to 14 different pressure levels covering a density range from 120 kg m Ϫ3 down to 2.4 kg m Ϫ3 . At each pressure level, experimental results were collected at 7 to 11 different applied powers. Transient measurements were made for an ex-perimental time of 1 s, while for the steady-state measurements the total elapsed time used was 40 s. In either case, 250 measurements of the bridge imbalance voltage were obtained. To elucidate the end effects in the experiment, a special series of runs were made using an additional digital voltmeter to measure directly across the long hot wire, the short hot wire, and the bridge. Finally, an abbreviated set of measurements was made for pressures below 1 kPa. In all, we made 883 measurements. The temperatures were measured on the International Practical Temperature Scale of 1968 (IPTS 68) but the effect of converting the temperatures to the International Temperature Scale of 1990 (ITS 90) on the reported thermal conductivity is less than 1 W m Ϫ1 K Ϫ1 .
The steady-state measurements required the development of a new data-analysis procedure. The rises in steady-state voltage as a function of time were always examined to select reliable measurements. Five typical profiles of the bridge imbalance, used to select the appropriate range of power levels, are shown in Fig. 6. Trace b in Fig. 6 is considered reliable since it is nearly horizontal after 20 s. In trace a, the power level is too low, so electronic noise is significant in the imbalance voltages. In traces c, d, and e the power levels are too large, so convection makes a visible contribution. The next step was to determine the experimental temperature rise ⌬T . The experimental voltage rises were averaged over a time interval where they were nearly constant. To find the optimum time to begin the averaging, the last 50 bridge imbalance voltages were averaged to find a reference imbalance voltage. The actual average, V ave , is obtained by averaging the points from the first voltage, which is 0.5 % below this reference imbalance voltage up to the final data point. Solving the bridge equation with V ave yields the change in resistance in the variable arms of the bridge. The resistance change is finally converted into ⌬T using the calibration of the wire resistance as a function of temperature. The maximum and minimum values of the voltage rises were also obtained over the range averaged. The difference between them was expressed as a percentage of V ave , and is designated by the parameter TBAND . TBAND is a direct measure of the precision in ⌬T at the level of 3 standard deviations. In Fig. 7, the values of TBAND are plotted for all of the steady-state measurements made near 320 K. A final selection of valid measurements was made by rejecting all points with a TBAND larger than 2 %. This is equivalent to rejecting those points that have voltage traces similar to traces c, d, and e in Fig. 6.
A correction for radiation was also applied to ⌬T . The radiation correction for transparent fluids ␦T 5 was used as given in Ref. [4]. The maximum effect of this correction was 0.13 % at 340 K. Additional corrections have been considered by other authors for steady-state hotwire systems (see for example Refs. [9,11]). These include corrections for temperature jump, end conduction in the wire, lead-wire conduction, and temperature rise in the outer wall. The temperature-jump correction does not apply because the present pressures are not low enough. The correction for end conduction in the wire and the lead-wire correction were found to be negligible in our experiments because a bridge with a compensating hot wire was used. A special series of measurements was made to determine the size of the end effects by directly measuring the temperature rise of each wire. Temperature gradients in the outer wall were considered negligible for our thick-walled pressure vessel. The primary platinum resistance thermometer (PRT) was mounted on the outside of the pressure vessel. The temperature T ref of the PRT increased by about 30 mK for a series of measurements at a single pressure level. The temperature of the long hot wire, which is inside the cell, increased by a nearly identical amount. Since a Fig. 7. Temperature uncertainty, TBAND , for the 320 K isotherm of argon. Data with TBAND values greater that 2 % were considered invalid due to free-convection contributions. measurement series normally consisted of about 20 measurements at time intervals 1 min apart, an average change of 1 mK per measurement was negligible in comparison to the measured ⌬T , which was typically from 1 K to 4 K.

Free Convection
Convection has always been a problem in measuring thermal conductivity; its onset has been associated with the Rayleigh number. One of the major advantages of the transient method is the ability to detect and avoid contributions from convection. The present measurements using the steady-state mode also show the evolution of convection very clearly. Figure 8 shows the deviations between the steady-state measurements near 320 K, calculated before application of the correction for convection, and the thermal conductivity surface for argon [12]. Since the diameter of the hot wires is comparable to the boundary-layer thickness for heat transfer, the standard engineering models for vertical flat plates are not applicable, and so an empirical expression was developed for the thin-wire geometry.
The dimensionless Rayleigh number is commonly used to characterize the onset of free convection. For a concentric-cylinder geometry, the Rayleigh number is given by where g c is the local acceleration of free fall, and is the fluid viscosity. The correction for natural or free convection was obtained from two equations given by Le Neindre and Tufeu [13] for a concentric-cylinder apparatus: and where q is the applied power divided by length of Eq.
(1), q meas is the experimental heat flow determined from the measured voltage and current, q c is the heat transfer by natural convection, K is a numerical apparatus constant, and Ra is the Rayleigh number. Le Neindre and Tufeu use a numerical constant of 720, but also values of d , the thickness of the fluid layer, and l , the length of the internal cylinder. For our system, both d and l are constant and our ratio of d /l is much larger than that of Le Neindre and Tufeu. Since the aspect ratio d /l is constant in our apparatus, it is incorporated into the experimentally determined apparatus constant K for our hot-wire cell. Equations (12) and (13) together give Next, Eq. (8) is applied twice, once for uncorrected conditions and once for corrected conditions. Forming a ratio we can solve for the corrected thermal conductivity corr = ͩ q q meas ͪmeas , (15) or with the use of Eq. (14), The best value for K in Eq. (16) was 1.8435 ϫ 10 Ϫ6 and was obtained by comparing the experimental points for each isotherm against a parabolic fit of the companion transient measurements. This procedure is justified because the deviations of the companion transient measurements from the thermal conductivity surface of argon are less than 1 % [12]. After applying this correction for free convection, Eq. (16), to all of the steady-state measurements, the resulting deviations are plotted in Fig. 8 for the 320 K isotherm. Figure 8 contains three different regions. For densities from 0 to 40 kg m Ϫ3 , convection contributes very little to the measured conductivity, i.e., the corrections for convection are less than 1 %. In this region, the Rayleigh numbers range from near 0 to 17 000. For densities between 40 kg m Ϫ3 and 80 kg m Ϫ3 the corrections given by Eq. (16) gradually increase to about 5 %, while the Rayleigh numbers range up to 65 000. The correction is highly successful, as the band or range of measured values at each pressure level decreases. For steady-state thermal conductivities at densities above about 80 kg m Ϫ3 , for which some of the Rayleigh numbers range well above 65 000, the corrections for simple natural convection increase to about 12 %. However, the uncertainty bands associated with the uncorrected thermal conductivities no longer decrease. The measured voltage rises suggest that there is flow turbulence in the cells at the larger power levels. This could be along either the short hot wire, the long hot wire, or both. Measurements associated with Rayleigh numbers greater than 70 000 have to be rejected, and are omitted from Fig. 8.

Results
The transient measurements for all three isotherms are given in Table 2. All of the transient points with a STAT greater than 0.003 were eliminated; thus, 98 points remain at a nominal temperature of 300 K, 105 points remain at 320 K, and 102 points remain at 340 K. Figure  9 shows the deviations between all transient points and the thermal-conductivity surface of argon [12]. Based on Fig. 9 it can be concluded that, over the range of densities shown here, the deviations fit within a band of Ϯ1 %. The problem with the transient measurements at low densities shows up clearly as a systematic deviation. However, as shown later, this systematic deviation must be ascribed to a difference in the dilute gas 0 values used for the surface [12].
The steady-state measurements are listed in Table 3. As explained before, all steady-state results with TBAND greater than 2 % and all points with Rayleigh numbers greater than 70 000 were not considered. Thus, there are 104 points at a nominal temperature of 300 K, 120 points at 320 K, and 119 points at 340 K. The deviations between all steady-state points and the thermal-conductivity surface of argon [12] are plotted in Fig. 10. It can be concluded from Fig. 10 that over the range of densities shown, the deviations fit within a band of Ϯ2 % at a level of confidence of 95 %. The        9 and 10 shows that the systematic deviations at low densities are also seen in the steady-state measurements, which is why we ascribe this systematic deviation to a difference in the values for dilute gas thermal conductivity 0 . The agreement between the transient measurements and the steady-state results is quite good; the mean deviation between the two methods is 1 %. The steady-state single-wire results are given in Table  4 under several different headings. The single-wire experiments are an attempt to measure the end effects in each wire. At the ends of each wire, heat is flowing from the wire ends to the wire supports. In addition, a part of the applied heat is flowing from the end of the wire through the fluid to the cell ends, a geometry quite different from the center portions of the wire. The lines in Table 4 are in pairs by run and point number. The first line is the regular or normal result calculated from the measured bridge imbalances, while the result on the second line uses the voltages measured directly across the individual hot wires. As a check the added voltmeter was also connected across the bridge. In this case the two lines given in Table 4 are virtually identical. Relative deviations of the data from the thermal conductivity surface of argon [12] are also provided in Table 4. These deviations are shown in Fig. 11 as a function of the applied power q . We see that for the short hot wire the thermal conductivity results are about 20 % above the normal steady-state bridge values, while for the long hot wire the deviation is around 4 % for the higher powers. An inspection of the single-wire voltage profiles indicates that turbulent convection (see trace c of Fig. 6) is first seen in the short-wire cell; however, it carries over into the full-bridge measurement. For the low-temperature system [1], using wire lengths of 0.05 m and 0.10 m, end effects of 8.3 % for the short wire and 4.8 % for the long wire were observed in nitrogen gas. The present wires have lengths of 0.05 m and 0.20 m, with end effects of 16 % and 4 %, respectively. For equivalent wire lengths (short wires), the present end effects are larger by about a factor of 2, quite reasonable given that the present steady-state measurements run considerably longer in time on a different gas.
The very last segment in Table 4 shows a series of runs made at a pressure considerably below ambient with the added voltmeter connected across the short hot wire. The exact pressure was difficult to determine because we did not have an appropriate pressure gage. We estimate that the pressure must be between 2 Pa and 1300 Pa, somewhere between the limit of the forepump and the rather approximate reading of the regular pressure gage. We can be certain that these measurements fall into the Knudsen region where the thermal conductivity is proportional to pressure. The deviations for the two highest power levels in this series fall between the ones for the long and the short hot wires. We might expect the end effects to be smaller under these conditions because the heat flow from the wire ends to the cell ends is reduced.
It is shown that for all conditions, except perhaps turbulent convection in either cell, having the short hot wire in the bridge insures sufficient compensation for end effects. This seems to be in agreement with the conclusions of Taxis and Stephan [7]. Finally, the steady-state results, the first line of each pair, are independent of applied power, an excellent verification of the key requirement that the results are free from the influence of natural convection.

Analysis
Argon was selected as the test fluid because the dilute-gas thermal conductivity 0 and the first density correction 1 = (Ѩ /Ѩ ) T are well known from Chapman-Enskog theory, as is the pair interaction potential for argon. The present results must be compared with the values derived from theory to validate the technique. To make the analysis easier, the large number of points was reduced by averaging as follows. All of the results were shifted from their experimental temperatures to the even temperatures of 300 K, 320 K, and 340 K using the        11. End effect ratios from "single wire" measurements to determine the effectiveness of end effect compensation by the bridge circuit. The baseline is the correlation of Younglove and Hanley [12]. thermal conductivity surface for argon [12]. The mean of this adjustment is Ϫ0.35 %, the maximum is Ϫ1.25 %. The next step is to average the results for the various power levels at each pressure. This step gives us 13 or 14 points per isotherm. Theory indicates that the thermal conductivity is a nearly linear function of density at low densities. Since values are now available, measured by two different methods in the same apparatus, it is reasonable to combine both transient and steady-state values into one result. Thus the final step is to obtain averaged straight lines for each isotherm from the experimental results. These averaged straight lines become the basis for deviation plots to assess the accuracy of the present steady-state and transient results.
The averaged thermal conductivities adjusted to nominal isotherm temperatures are given in Table 5 along with the deviations of these values from the straight lines. From Table 5 it is easy to establish that the mean difference between transient and steady-state measurements is about 1 %, with the steady-state values nearly always higher. While transient and steady-state values agree to within their combined uncertainties, we cannot exclude the possibility that a systematic difference of about 1 % may exist between the two methods. Our straight-line intercepts are the values of the dilute-gas thermal conductivity 0 and the slopes are values of the first density correction 1 . The coefficients of the lines with their calculated expanded uncertainty at a level of uncertainty of 95 % are given in Table 6.
The reason that we have used the thermal conductivity surface of Younglove and Hanley [12] in this paper for comparisons, etc., rather than some of the other possible choices, will now become clear. The dilute-gas thermal conductivities of the Younglove and Hanley model [12] are equivalent to the theoretically derived values of Kestin et al. [14]. We conclude from Table 6 that our dilute-gas thermal conductivities are lower than the theoretical values of Kestin et al. [14] as well as those of Aziz [15]. However, they appear to be in better agreement with those of Aziz. Our first density corrections are seen to depend slightly on temperature, unlike the theoretical results, which are constant. They appear to be in better agreement with the values of Rainwater and Friend [16,17] than with those of Bich and Vogel [18]. Good agreement is found with the previous work using the NIST low temperature instrument [19]. Finally, in Fig. 12 the present results are compared with our earlier ones [1,4,[19][20][21][22], and with those of other authors [23][24][25][26][27][28] for a temperature of 300 K, where the baseline is the present least-squares fitted line (coefficients in Table 6). The present results are connected by lines to set them off from the others. All other results were shifted to a temperature of 300 K by using the thermal-conductivity surface of [12]. The largest shift is Table 6. Experimental and theoretical dilute-gas thermal conductivity and first-density coefficients a for argon. Here, U is the expanded uncertainty (coverage factor k = 2, and thus a 2 standard deviation estimate) Ref. [22] c Ref. [14] d Ref. [15] e Refs. [16,17] f Ref. [18] around 2.3 %, because a few of the original experimental temperatures are as high as 308 K. Figure 12 shows that the present results, including the new steady-state data, are in good agreement with our earlier results [1,4,[19][20][21][22]. Figure 12 also shows that all of the values assembled here, which include those from transient experiments, steady-state concentric cylinders [24], and steady-state parallel plate systems [23], agree to within 1 %, a truly remarkable result. Comparisons made at the other two temperatures, 320 and 340 K, are quite similar to Fig. 12.

Uncertainty
The measurements were deliberately made over a wide range of power levels. For both transient and steady-state results those power levels which were either too small or too large were eliminated. For power levels that are too low, the bridge imbalance becomes comparable to the background noise level and there is significant uncertainty in the measured temperature rises. In general the instruments require a temperature rise of at least 2.5 K to obtain accurate transient results (STAT < 0.003). With power levels that are too high, curvature is found in the relation for ⌬T vs ln(t ) for the transient measurements, as is typical of convection. The onset of natural convection is also observed in the steady-state measurements as a time-varying steady-state temperature rise. Imposing certain limits on the experimental uncertainty parameter-a maximum STAT of 0.003 for the transient points and a maximum of 2 % in TBAND for the steady-state points-seem to be appropriate restrictions. The uncertainty in transient thermal conductivity data increases at densities below 28 kg m Ϫ3 (p = 1 MPa). For valid steady-state measurements of the thermal conductivity, the Rayleigh number must be less than 70 000. With these restrictions it is found that the relative expanded uncertainty of the transient thermal conductivity is 1 % (k = 2, see Fig. 9), the uncertainty of the steady-state thermal conductivity is 2 % (k = 2, see Fig.  10), while the agreement between the two methods is 1 % (see Table 5). The overall agreement between our present results, our earlier transient results and the results of many other authors is a truly remarkable 1 % (see Fig. 12).
The steady-state results have a greater uncertainty than the transient ones. This can be attributed to a number of factors. First, the steady-state experiment requires an accurate measurement of the temperature rise ⌬T quite similar to the measurement of thermal diffusivity in the transient hot-wire system [20,21]. Second, the steady-state measurement requires accurate determination of the cell geometry: wire radius, cavity radius, eccentricity. Due to the constraints imposed by the onset of convection, the valid temperature rises for steadystate measurements are half those of the transient ones. Table 7 shows the mean temperature rises for both modes of operation. This may serve as a guide for operating hot-wire cells in either the transient or steady-state mode.

Summary
It has been demonstrated that the thermal conductivity of argon can be measured with a relative expanded uncertainty (k = 2) of 2 %, using a transient hot-wire system operating in an absolute steady-state mode. The bridge arrangement used in this experiment provides sufficient compensation, eliminating the problem of end effects for both transient and steady-state measurements. The selection of valid steady-state results is based on the shape of the curve of measured voltage rises, on the size of the error in the temperature rise, TBAND , and on the magnitude of the Rayleigh number. Since the fluid gap is quite large in transient hot-wire cells, the steady-state mode is restricted to the low-density gas. The use of the absolute steady-state mode requires no information on the thermophysical properties of the fluid of interest. This is a significant advantage for the measurement of thermal conductivities in the vapor phase of refrigerants where the fluid properties are not known well enough to obtain accurate corrections for transient hot-wire measurements. From a fundamental point of view it can contribute to the determination of accurate pair-interaction potentials for diatomic molecules, especially the long-range weak interactions, a very active research field, by measuring the thermal conductivity of low-temperature vapors.