Anal. Chem. 1980, 52, 813-817
813
Optimum Conditions for Crude Oil and Petroleum Product Analysis by Carbon- 13 Nuclear Magnetic Resonance Spectrometry S. Gillet and J-J. Delpuech" Laboratoire de Chimie Physique Organique, ERA CNRS 222, Universit6 de Nancy I, C.O.140, 54037 Nancy Cedex, France
P. Valentin and J-C. Escalier Centre de Recherche Elf-Solaize, E.P. 22, 69360 Saint-Symphorien d' Ozen, France
A general procedure to determine the optimum instrumental parameters in quantitative Fourier transform nuclear magnetic resonance (FT-NMR) analysis is described. Four couples of solvent and relaxation reagent have been tried; the couple C,D,CI,/Fe(acac), is shown to be about 3-5 times more efficient than the usual CDCI,/Cr(acac), couple. Graphs are shown which allow the determination of the best conditions for obtaining either a given resolution in a minimum time, or the minimum time of operation, whatever the resolution may be. The procedure is applied to the carbon analysis of three crude oil heavy ends (Light Arabian). The time of operation may be reduced to about 1.5 and 12 h for a vacuum distillate ( E = 370-535 "C) and an asphalt. I n any case, the reagent concentration must not be greater than 0.1 M.
Carbon-13 quantitative analysis requires operating conditions which are a t some variance with those generally used in routine qualitative analysis. The relevant instrumental parameters in Fourier transform nuclear magnetic resonance (FT-NMR)-the flip angle 8, the acquisition time t 2 ,and the pulse delay t3 (using proton gated decoup1ing)should be set so as to obtain the highest S I N ratio subject to the condition that the sensitivity obtained by spectral accumulation be uniform throughout the whole spectrum. In this respect, optimum values of 8 and t 3 were determined in a previous paper ( I ) so as to ensure an accuracy of 1%: 8 = 90" and t , = 4.6 to 6.5 Tic, depending on whether the I3C relaxation time TICis larger or smaller than 20 s. For complex samples such as petroleum products, these values lead to recommended pulse delays of about 200 and 270 s for the titration of nonquaternary or quaternary (aliphatic or aromatic) carbon atoms, respectively. Such long pulse delays, however, would result in time-consuming operations when a high sensitivity enhancement is required, e.g., for heavy crude oil residues. This drawback has been partially circumvented by using paramagnetic relaxation reagents (2-5) which, without shifting resonance frequencies, strongly decrease both carbon and proton relaxation times Tlc and TIHand cancel the Nuclear Overhauser Effect (NOE) factor 9. The reagents most used a t the present time are the acetylacetonates of chromium(II1) and iron(II1): Cr(acac)gand Fe(acac), (6, 7). The procedures described in each specific example from literature (8-13) cannot, however, be easily extended to any other sample owing to the necessity for diluting the sample of volume V E with a convenient volume VS of an appropriate solvent so as to ~ )the dissolve the required concentration [Q] (in m ~ l - d m - of relaxation reagent. No comparison between the various solvent-reagent systems has been made regarding the efficiency of the procedure, Le., the reliability of the 13C titration, 0003-2700/80/0352-0813$01 .OO/O
the gain of time, and the subsequent degradation of the resolution. These points are examined with some detail in this paper. First of all we show how obtaining a practical procedure may be rationalized in each case by an analysis of the instrumental and nuclear parameters.
OPTIMIZATION PROCEDURE The optimum conditions, i.e , the whole set of operating factors t2,tS,8, and [ Q ] will be defined in either of two ways: (i) The resolution r , i.e. the linewidth of a reference signal, should not be larger than a minimum value rm. (ii) The time of operation T should be reduced to a minimum for a given final signal-to-noise ratio (S/NII. First of all, the optimum flip angle Om may be estimated using a theoretical treatment from a previous publication (I). Assuming that the residual NOE factor is nearly zero, and the relaxation times T1c of the carbon atoms are reduced to values below ca. 20 s, an optimum value 0, = 80" gives a slightly improved sensitivity as compared to 0 = 90" which was the value recommended for carbons with large 9 factors. If (S/ N),900is the S / N ratio obtained after one 90' pulse for the pure sample E containing the same quantity of'the relaxation reagent, the S / N ratio after one 80" pulse for a solution of E in solvent S is
(S/N), = --.(S,/N),900 VE sin d,, VT where VT = VE + VS. Performing YI scans in EI time T = n(tz + t3)results in a sensitivity enhancement of d;,and therefore
Obtaining a minimum amount of time T i s therefore equivalent to obtaining a maximum value of the variable V
V = (S/N),'/(t,
+ t3) =
with V , = ( ( S / N ) l g 0 0 4 n The sample volume fraction b'E/VT is subjected to some instrumental constraints which are graphically expressed in Figure 1: (a) ( V E I V T ) < 1 (line d); (b) ( V E I V T ) < E , (line e), so as to maintain a deuterium signal sufficiently intense to allow a long term field-frequency stabilization (lock); (c) ( V E I V T )> F1 (line f), to avoid any line-base distortion due to an intense 13C signal from the solvent (this constraint could be removed by using expensive 13C depleted solvents, but these are not practical for routine analysis). Another restriction may arise from the solubility of the reagent in the solvent, which decreases by the addition of the C 1980 American Chemical Society
814
ANALYTICAL CHEMISTRY, VOL. 52, NO. 6, MAY 1980 -'biz
Figure 2. The optimization variable V / V , = h([Q]); [ 0 ] = concentration of the relaxation reagent in m ~ l - d m - ~
LL
J
-
_________-_ concentration [c:
3'
Figure 1. Schematic plots of the possible sample volume fractions VEI VT(shaded area I) and pulse delays t , (area 11) as a function of
the relaxation reagent concentration [ 01 (m~l-dm-~). Other notations are explained in the text sample. A solubility curve VEIVT = g([Q]) is schematically drawn on Figure 1 (curve b). T h e variation domain of ( V E / V T ) is thus defined by the shaded area I in Figure 1 enclosed between the ordinate axis (segment E I F J and lines b ( E z F ~e ,(E1E2),and f ( F I F J . If we turn now t o parameters t z and t3 in Equation 3, we may first observe that tz is determined by the digital resolution r = l / t z alld in practice by the computer memory area which is reserved for data acquisition. Typical values are t z = 0.6 and 1.3 s for spectra a t 80-100 MHz and 200-300 MHz, respectively. We are then left with the problem of adjusting t3 for a given t z value. This requires observing the minimum value t3, of the waiting time t3 from which a correct 13C analysis of a template molecule (containing several types of carbon nuclei with TlC and 7 values covering approximately the same range as the sample to be examined) can be obtained within experimental uncertainty. The latter value is defined as the standard deviation s (70)obtained when performing a series of analyses of a reference compound (using the same relaxation reagent and a time much larger than t3,). The measurement of t3, is carried out for several concentrations [Q] of the reagent. The form of the curve obtained by plotting t 3 as a function f([Q]) of [Q] is shown in Figure 1 (line a). f ( [ Q]) is a function continuously decreasing to zero, as concentration [Q] is increased from zero to a given value [QM]. ([QM] N 0.1 m ~ l - d r n -for ~ petroleum products and the aforementioned relaxation reagents.) The possible values of t3are thus represented by the shaded area I1 in Figure 1which is bordered by curve a, the vertical line A passing by Fz (whose abscissa is the highest permissible value of VEI V T ) and the horizontal axis. For the last step, we plot in the same figure the resolution r (half-height linewidth of a reference signal) as a function of [Q]. T h e representative curve c is in fact a straight line as expected from the theory of paramagnetic relaxation. Graphs of Figure 1 are now used to visualize the wsly to obtain the desired optimum conditions. For optimum condition (i), the desired minimum resolution r , is reported on the ordinate axis of Figure 1, and the corresponding value 4 , of [Q] is obtained as the abscissa of the point c1 of line c whose
ordinate is r,. The optimum values t3, and V E I V T , of t3 and V E I V T are similarly obtained as the ordinates of the intersections al and bl of lines a and b with the vertical line passing by cl. If a minimum time of operation is desired whatever the resolution may be, we refer t o the graph representing the variable V / V , as a function of [Q] (Figure 2 ) , when the highest and lowest values of V E / V T and t3,respectively, are introduced into Equation 3
V/Vo 3 h(lQ1) = g([Ql)'/(tz
+ f[Q1)
(4)
This graph is limited on the right by the vertical line A. T h e first part of the curve (small [Q] values) is approximately linear due t o the fact that V E I V T is constant (segment E I E z of Figure 1) and that l / ( t 2 + t 3 ) = l / t 3 0: [Q] as long as t3 >> t2. This linearity ceases when the point representing VEI V T in its variation domain is running along E,Fz (curve b, Figure l), i.e., when the corresponding abscissa [Q] is contained within the segment E z fFzfof the abscissa axis of Figure 2. The curve then goes through a maximum for a value [Q,,] < [QM] of [Q] (in fact, in our experiments [QM] E [Q,,]). [Q,] is then reported on the abscissa axis of Figure 1 to obtain the corresponding VEI VT,, and tamaxvalues of V E / V T and t3from lines b and a. These settings correspond to a time T and a resolution r which are derived from Equation 3 and curve a. Some difficulty may arise in practice in the definition of the final signal-to-noise ratio to be introduced for the computation of T. If the spectrum consists of well-separated lines, refers to the smallest line to be analyzed. If the spectrum consists of large envelopes of overlapping lines-as is the case with heavy crude oil residues-the integration is performed over a characteristic frequency range, e.g., over the aliphatic carbon region or the aromatic carbon region or over segments from these regions. In this case, the signal-to-noise ratio should be considered over the integral and should refer to the portion of the spectrum whose integrated intensity is smaller, In practice, relative gains of time only need to be considered, and spectra accumulation will be carried out until a satisfactory quantitative analysis is possible, i.e., as soon as the signal-to-noise ratio of the digital-analogic integration curve over the smallest frequency range of interest is larger than ca. 60, a value which ensures a standard deviation of the integral less than 3 % . The choice of the set of parameters defined above then ensures the operator that this will occur in a minimum time. We have thus defined a rational procedure for optimizing the instrumental parameters in any case. We shall now focus attention on the specific case of petroleum products analysis using the two aforementioned relaxation reagents and either
-
ANALYTICAL CHEMISTRY, VOL. 52, NO. 6, MAY 1980
815
'E/VT 1.01
j {\ 0
0.2
OX
Q6
0.8
CQi
1.0
Flgure 3. Solubility curves showing the maximum sample volume fraction as a function of the Concentration [ Q] of the rebxation reagent 0. Sample: Light Arabian petroleum vacuum distillate ( E = 370-535 "C); (a)Solvent S = CDCI, and reagent Q = Cr(acac),; (b)S = CDCI, and Q = Fe(acac),; (c)S = CpzCl, and Q = Cr(acad),; (d) S = C$&I4 and Q = Fe(acac),
of two convenient solvents: chloroform-d, (CDC13) and 1,1,2,2-tetrachloroethane-d2 (C2D2C14).
L-'J
Figure 4. Optimized curves of t h e pulse delay t , (a) and resolution r(c)as a function of the concentration [ Q] (m~l-dm-~) of the relaxation reagent (Cr(acac), or Fe(acac),) using a solution of toluene in C,D,Cl4
as a reference system Table I. Relaxation Times and NOE Factors of Carbon-13 Nuclei of Toluene ( 1 4 )
EXPERIMENTAL Materials. The relaxation reagents, the solvents, and the toluene sample used as a reference are Merck products used without further purification. The petroleum products were Light Arabian petroleum fractions furnished by Elf refining company. Three fractions (VD, VR, and AS, respectively) were analyzed: the vacuum distillate ( E = 370-535 "C), the vacuum residue ( E > 535 "C), and the pentane-precipitated asphalt. Chlorinated solvents seemed to be the most appropriate as they are known to dissolve significant amounts of petroleum products. Fully chlorinated compounds, such as CC4, were discarded since they cannot provide a deuterium lock signal. The solvents used should preferably give one carbon signal lying outside the frequency ranges of interest, 10-60 and 110-160 ppm for hydrocarbons. CDCl, and CzD2C14,two commercially available solvents, were chosen along these lines. Deuterium chloroform (CDCl,) is the most commonly used solvent in quantitative analysis at the present time. We however selected 1,1,2,2-tetrachloroethane-d4 which has the net advantages of a higher boiling point ( E = 146 "C instead of 60 "C) and better dissolving properties toward hydrocarbons (see below). Solubility Measurements. Seven mixtures were prepared by mixing known volumes VS and VE of solvent and sample. Increasing amounts of relaxation reagent were added to these mixtures up to saturation at room temperature. The saturated complex was filtered off, and the chromium(II1) or iron(II1) content of the filtrate was measured by neutron activation or by X-ray fluorescence, respectively. Solubility curves are shown in Figure 3 for the following systems: CDCl,/Cr(a~ac)~/vD; CDCl,/Fe(acac),/VD; CzDzC14/Cr(acad),/VD; CzD2Cl4/Fe(acac),/VD. Solubility is clearly much larger for the Fe( a c a ~ ) ~ / C ~ Dpair ~ C (four l ~ times as large as for the usual CDCl,/Cr(acac), pair. The F e ( a c a ~ ) ~ / C ~ pair D ~ Cwas l ~ consequently examined along with the remaining two samples (VR and AS). It should however be noted that the formation of a complex between the chlorinated solvent and the relaxation reagent (and also with the dioxane internal standard reference) is possible after a few days, especially for the chosen pair C,D,Cl,/Fe(a~ac)~( 5 ) . Solutions to be analyzed should therefore be freshly prepared before use. In this respect, we have checked, using a known molecule (toluene), that the agreement between experimental and calculated carbon percentages is independent of the volume ratio
VE/ V T . NMR Spectroscopy. Proton spectra were taken on a Cameca 250 superconducting spectrometer operating a t 250 MHz in the continuous wave mode. Carbon-13 Fourier transform spectroscopy was performed with the same apparatus a t 62.86 MHz. Experimental conditions were as follows: memory capacity used for data acquisition: 32K words of 20 bits; sweep width: 12500 Hz
carbon
relaxation time, Tic, s
NC)E factor, q
c, c2
58.0 20.0
0.43 1.32
21.0
1.32 1.70
c 3
C,
c,
15.0 16.3
-
0.61
( N 200 ppm) and digital resolution: 0.76 Hz; flip angle: 80"; pulse width: 10 ps; acquisition time t 2 = 1.3 s; pulse delay t 3 = 0 to 270 s; delay-time: 40 ~ s frequency ; filter: 12400Hz; scan number: 100 to 10000; digital-analogic integration; sample temperature: 25 "C (toluene), 40 "C (VD), and 70 "C (VR and AS); sample tube diameter: 8 mm (13C);deuterium internal lock; standardization of frequencies using a reference signal (from dimethyl sulfoxide) and the frequency meter built into the spectrometer; standardization of errors using a solution of tetralin in C2D2C14( V E / VT = 0.66) and a 0.025 M concentration of reagent: the maximum standard deviation of experimental carbon molar ratios to the calculated values over 20 independent measurements was s =
2.7%.
Determination of Pulse Delays and Resolution. Solutions of toluene in CzDzC14were added with a variable amount of relaxation reagent. The intensity ratios of the carbon nuclei of toluene were measured as a function of the pulse delay which was progressively increased from zero. These ratios were compared with those computed from the toluene molecular formula. The waiting time t3 was increased until the two sets of numbers, experimental and calculated, were coincident within experimental uncertainties (s = 2.7%). This limit yields the desired t8mvalue which is plotted on curve a, Figure 4. It was further checked that the two sets of numbers remain in close coincidence as the pulse delay is increased beyond the boundary value t3m. Simultaneously the mean linewidth r of the toluene signals was measured, thus yielding a plot (c) of P as a function of [Q] (Figure 4). Curves for the two relaxation reagents Cr(acac& and Fe(acac), in CzD2C14were found to be coincident within experimental errors, thus showing a fortuitously identical intrinsic effectiveness of both reagents. The results were shown to be independent of the volume fraction VE/ V T . Toluene was chosen as a template compound for hydrocarbons because of a wide range of relaxation times and NOE factors (Table I). This point was checked using two other reference compounds: tetralin and 2-ethylnaphthalene, which gave results identical to toluene within experimental errors. Limiting Values of the Volume Ratio VEI V T . These values are highly dependent on the operator's appreciation of the quality of the spectrometer used. A minimum value VE/ VT 2 0.25 (line f of Figure 1) was found necessary to avoid distortions of the base line far away on each side of the solvent signal. These
816
ANALYTICAL CHEMISTRY, VOL. 52, NO. 6, MAY 1980
Table 11. Operating Conditions for Three Light Arabian Petroleum Fractions sample
r , Hza
vacuum distillate
2 3.5 5.0
vacuum residue
asphalt
9, m 0 1 , d m - ~ ~ t,, s c 0.025
0.050 0.075
2.0 3.5 5.0
0.025 0.050
2.0 3.5 5.0
0.025 0.050
0.075
0.075
r = resolution. Q = concentration of Fe(acac),. of operation. f By NMR. By elemental analysis. a
VEIVTd
T,h e
C/ Hf
ratiog
16 6 2
0.66 0.66 0.64
13.5 5.5 2.6
0.58
0.57
16 6 2
0.66 0.64 0.52
13.5 6.2 3.8
0.67
0.68
16 6 2
0.30
57.0 30.0 18.0
0.99
0.94
0.30 0.24
t , = pulse delay.
VE/VT = sample volume fraction.
e
T = time -
0.25
0.20-
0.15
!
0.10
0.05-
0
0.05
0.I
LO1
0.15
Figure 5. Optimization variable V/ V, = ([ Q]) using a Light Arabian petroleum vacuum distillate and four solvent-reagent pairs: (a) CDCI,/Cr(acac),; (b) CDCI,/Fe(acac),; (c) C2D2CI4/Cr(acac),;(d) C,D,CI,/Fe(acac),
distortions were of the same magnitude as the signals of the sample beyond this dilution limit. On the other side, a maximum value V E j V T I 0.66 was found necessary to prevent fortuitous “delocking” of the spectrometer. (No improvement in the unfavorable effect of dilution could be obtained by using an internal capillary filled with the lock substance.)
RESULTS AND DISCUSSION Curves representing the optimization variable V / V , as a function of the concentration [Q] of the relaxation reagent are shown in Figure 5 for a vacuum distillate using either of the two aforementioned solvents and of the two relaxation reagents. The couple C2D,C14/Fe(acac)3clearly has the highest effectiveness due t o a better solubility of the reagent in this solvent. The time of operation is thus divided by a factor of about 3.5 with respect to the usual CDCl,/Cr(acac), pair. The former system alone will therefore be examined below. T h e optimum concentration [Q] is equal to 0.1 m o l ~ d m - ~ , in which case the optimum pulse delay t , is zero. I t should be pointed out that using higher concentrations does not result in a further decrease in the time of operation, and on the contrary brings about an unwanted degradation of the resolution. Graphs analogous to those of Figure 1 are represented in Figure 6 for the three petroleum fractions studied. Graphs are limited to the useful range of concentration 0 5 [ Q ] 5 0.1
i
“TI 1 * 3 c
-I 10 3 keel
Flgure 6. Optimum operating conditions as a function of t h e concentration [ Q] of reagent Fe(acac),: pulse delay t, (curvea), resolution r (c),sample volume fraction V € / VT(e, e’, e”, b, b’, b”), optimization variable V/ V, (v, vr, v”) for a carbon-I3 analysis of three Light Arabian petroleum fractions: vacuum distillate (b, e, v), vacuum residue (br, e’, v’) and asphalt (b”, d’, v”). Solvent: C2D2C14. Flip angle: 80’. Temperature: 40, 70, and 70 O C for the three fractions, respectively
m ~ l e d m - so ~ , that truncations of solubility curves b may occur before their intersection F2 with line f. In the case of asphalt, another factor appeared to limit the possible range of V E / V T values, namely the solubility of asphalt in CzDzC14,which imposed V E I V T I0.30 (line e“) instead of V E / V T I0.66 (lines e, e’). These curves show that the optimum conditions for a minimum time of analysis are 0 = 80”; [Q] = 0.1 moldrn-,; t , = 0 s; V E I V T = 0.58, 0.47, and 0.17 for the three fractions studied VD, VR, and AS, respectively. Convenient times of analysis with our spectrometer were found to be, respectively: 1.2, 1.9, and 11.2 h. Under these conditions, the resolution is r = 6.5 Hz, which is too poor for an integration of individual lines, but is sufficient for an integration over a frequency range. If we rather look for conditions ensuring a better resolution r , a concen- ~ be used, and a convenient tration [Q] < 0.1 m ~ l - d m must pulse delay should be computed to obtain a minimum time of operation according to the procedure described above. The optimum setting parameters are reported in Table I1 for a set of three resolution values: r = 2.0, 3.5, and 5.0 Hz. The times of operation used are also shown for convenience. In fact, most of them are computed values since, for a given compound, the operation time T i s proportional to the optimization variable
817
Anal. Chem. 1980, 52. 817-820
V; this allows the computation of any time T from one experimental T value (if curves of Figures 2 to 4 have been previously determined). Results of experiments using relaxing agents may indeed be vitiated by associations of the relaxation reagent with the sample molecules, thus resulting in a nonuniform sensitivity enhancement all along the frequency range explored. The absence of such effects has been checked with three reference compounds: toluene, 2-ethylnaphthalene, and tetraline assumed to be representative of hydrocarbons found in petroleum products (see Experimental section). Such a control is indeed impossible with petroleum products for which a parallel NMR analysis without relaxation reagent is impossible. Some evidence for the reliability of the method is provided by comparison of the carbon-to-hydrogen ratio determined either by NMR or by elemental analysis. The NMR method in this case utilizes a reference compound, 1,4-dioxane,containing carbon and hydrogen, which is added to the sample of interest. Results from the two methods (Table 11) show a satisfactory agreement within experimental error (s = *3%) and are a good indication of the reliability of the NMR method. ACKNOWLEDGMENT We thank E. Dumortier and E. Eppiger for their assistance in recording spectra at 250 MHz.
LITERATURE CITED (1) S. Gillet and J J . Delpuech, J. Magn. Reson., 38, 3 (1980). (2) M. S. Gutemsky and D. E. S. Natusch, J. Chem. phvs , 57, 1023 (1972). (3) D. E. S. Natusch, J. Am. Chem. Soc.. 93, 2566 (1971). (4) G. N. La Mar, J. Am. Chem. SOC.,93, 1040 (1971). (5) G. C. Levy and J. D. Cargioii, J. Magn. Reson., 10, 231 (1973). ( 6 ) 0. A. Gensow, A. R. Burke, and W. D. Vernon, J . Am. Chem. SOC., 94, 2550 (1972). (7) S. Barcza and N. Engstrom, J . Am. Chem. Soc., 94, 1762 (1972). (8) J. N. Shoolery and W. C. Jankowski, Varian Application Note, Varian Associates. Palo Alto. Calif.. number NMR.73.4 (1973). (9) J. N. Shooiery and W. L. Budde, Anal. Chern., 48, 1458 (1976). (IO) B. Thiault and M. Mersseman, Org. Magn. Reson., 8 , 28 (1976). (11) H. C. Dorn and D. L. Wooton, Anal. Chem., 48, 2147 (1976). (12) H. L. Retcofsky, F. K. Schweighardt, and M. Hough, Anal. Chem., 49,
585 (1977). (13) D. L. Wooton and W. H. Coleman, Fuel. 57, 17 (1978). (14) A. Aiierhand and D. Doddrel, J. A m . Chem. Soc.. 93, 2777 (1971).
RECEIVED for review November 16, 1979. Accepted January 29, 1980. One of us (S.G.) wishes to thank the Elf Society, Solaize, France, for a research grant. Financial support of DBlBgation GBn6rale 5 la Recherche Scientifique et Technique (DGRST) for this work (contract no. 77-7-1034) is gratefully acknowledged. The Cameca spectrometer of the Centre de Mesures Physiques de l'Acad6mie d e Nancy-Metz was purchased with funds from the Centre National de la Recherche Scientifique (CNRS),DGRST, the University of Nancy-Metz, and local public or private authorities.
Derivation of Structural Parameters for Coal-Derived Oil by Carbon- 13 Nuclear Magnetic Resonance Spectrometry Tadashi Yoshida" and Yosuke Maekawa Government Industrial Development Laboratoty, Hokkaido, 4 7-2Higashi- Tsukisamu, Toyohira, Sapporo 06 7-0 7, Japan
Hiroyuki Uchino and Susumu Yokoyama Coal Research Institute, Faculty of Engineering, Hokkaido University, Sapporo 060, Japan
Aromatic carbon signals in the 13C NMR spectra of coalderived oil can be divided into three groups, each being assignable to protonated, bridgehead, and substituted carbons, based on the chemical shifts of model compounds and the characteristic spectra of "ring-type''fractions of coal-derived oil. Relative areas of these groups afford direct calculation of BrownLadner's structural parameters without any additional assumptions. The determination of protonated carbon by the Knight method is also dlscussed.
T h e structural analysis of complex mixture such as coalderived oil by the 'H NMR method was first reported by Williams (1) and Brown et al. ( 2 ) . Although the method was later modified ( 3 , 4 ) ,it always requires an assumption on the H / C atomic ratio of aliphatic portions to determine the average skeletal structure of mixture. On the other hand, 13C NMR spectrometry enables direct measurement of the hydrocarbon skeleton. Quantitative analysis by I3C NMR is now possible by employing sufficient pulse repetition time to avoid progressive saturation of signals and the gated-decoupling technique to suppress nuclear Overhauser enhancement ( 5 , 6 ) . The modern 13C NMR can be 0003-2700/80/0352-0817$01.OO/O
a powerful tool for the structural investigation of coal-derived oil (7-9). However, the useful structural analysis of coalderived oil by the I3C NMR method has rarely been reported, probably because the spectra are too complicated. Fortunately, NMR spectra give valuable information with regard to the types of aromatic carbons. The aromatic region in the spectra can be divided into three groups, each being assignable to protonated, bridgehead, and substituted carbons, respectively. The structural parameters of any mixture can be calculated by the relative amounts of these groups of carbons. It is therefore necessary to determine their chemical shift ranges accurately. The purpose of this work is to determine the chemical shift ranges of protonated, bridgehead, and substituted carbons of coal-derived oil on the basis of the chemical shifts of model compounds (10-13) and the characteristic 13C NMR spectra of the ring-type fractions of coal-derived oil, and thus derive the equations for structural parameters. The determination of protonated carbon by the Knight method ( I a) is also discussed.
EXPERIMENTAL Coal from Akabira, Hokkaido (Japan), was hydrogenated at 400 "C for 60 min under the initial hydrogen pressure of 100 kg/cm* over Adkins catalyst. The product was extracted with B 1980 American Chemical Society
