You are on page 1of 4

Esso Research and Engineering Co., Brit. Patent 798,542 Shell International Research Maatschappiy N V , Brit.

(July 23, 1958). Patent 973,393 (Oct. 28, 1964).


Hollyday, W. C., Jr., Munsell, M. W. (to Esso Research Spindt, R. S.,et al., “Nitrogen Oxides, Combustion, and
and Engineering Co.), U. S. Patent 3,076,791 (Feb. 5, Engine Deposits,” SAE Golden Anniversary Fuels and
1963). Lubricants Meeting, Philadelphia. Pa., Sovember 1955.
Moyer, R. C . (to Pure Oil C o . ) , U. S. Patent 3,143,877 Verley, G. M. (to Sinclair Research), U. S.Patent 3,044,860
(Aug. 11, 1964). (July 17, 1962).
Rogers, D. T., et al., “Mechanism of Engine Sludge Forma- RECEIVED for review April 29, 1968
tion and Additives Action,” SAE Golden Anniversary ACCEPTED August 23, 1968
Fuels and Lubricants Meeting, Philadelphia, Pa.,
November 1955. Division of Petrochemistry, 155th Meeting, ACS, San
Rohm & Haas G.m.b.H., Brit. Patent 937,734 (Sept. 25, Francisco, Calif., April 1968. Permission to publish granted
1963). by the British Petroleum Co., Ltd.

GEL PERMEATION CHROMATOGRAPHY


Calibration Curue from Polydisperse Standards

S . T . B A L K E , A . E . H A M I E L E C , A N D B . P . LECLAIR
Chemical Engineering Department, McMaster Uniuersity, Hamilton, Ontario, Canada

S . 1. P E A R C E
Shell Oil,Ltd., Montreal, P. Q., Canada

A search technique and computer program (Fortran IV) have been developed
to determine the calibration curve for a gel permeation chromatograph using
broad molecular weight distribution standards. Two average molecular weights
(number, weight, or viscosity) for one or two polydisperse standards are required.
A Rosenbrock search is used to find the calibration curve constants. This search
has been tested for a variety of cases on an IBM 7040 computer. An evaluation
of these results is presented.

THEuse of the gel permeation chromatograph (GPC) data: one broad MWD standard and its number- and
for the determination of molecular weight distribution weight-average molecular weights; two broad MWD stand-
(MWD) of polymers is developing rapidly. T o interpret ards, given two number-average molecular weights, two
the GPC chromatogram it is necessary to establish a weight-average molecular weights, two viscosity-average
quantitative relationship between molecular weight and molecular weights, and one number- and one weight-
elution volume. The GPC is usually calibrated with narrow average molecular weight. Searches for nonlinear calibra-
MWD standards ( M , / M , 5 1.1) using the elution volume tions are also discussed.
a t the peak position. I t is not valid to use the peak
Development
position for broad MWD standards. For many linear
homopolymers narrow MWD standards are not readily Linear Calibration Curve. A linear calibration curve may
available, thus presenting the problem of determining the be expressed as
calibration curve with broad MWD standards. Rodriguez VOL = C1 - C2 logm(MW)
et ai. (1966) suggested a graphical procedure, but with
little elaboration. Cantow et al. (1967) proposed a pro- where
cedure for finding the calibration curve given the complete
MWD of one broad MWD standard. They are also able VOL = elution volume in counts
MW = molecular weight of a monodisperse
to correct their calibration curve for the effect of concen- sample
tration. C l and C2 = constants to be found by search
More recently, Frank et al. (1968) developed a technique program
for finding a calibration curve using a number- and a
weight-average molecular weight of one broad MWD The method involves the injection of one or two broad
standard. Their technique involves a graphical approxima- MWD standards into the GPC under conditions where
tion which makes it difficult to employ. the effect of correction for imperfect resolution on the
I n this paper a precise computer technique is presented, molecular weight averages of a polydisperse sample is
tested, and evaluated for several cases. These include negligible. [If skewing and symmetrical axial dispersion
searches for linear calibration curves given the following correction may be estimated, the true calibration curve

54 I & E C PRODUCT RESEARCH A N D DEVELOPMENT


where
Table I. Test Samples
Sample FMAX2 = objective function
No. M" M, M,IMn MW(1) = experimental weight-average
7A 48,000 55,000 1.14
108 197,400 250,000 1.26
molecular weight for sample 1
3A 333,000 430,000 1.29 MN(1) = experimental number-average
17 64,000 125,800 1.95
103 48,000 107,100 2.21 molecular weight for sample 1
VMW (1) = experimental viscosity-average
molecular weight for sample 1
may be obtained even. under extreme conditions (Balke T E M P 1 = calculated number-average
and Hamielec, 1968)]. This usually means residence times
of about 100 minutes or greater, according to Duerksen
molecular weight
and Hamielec (1968). This assumption of negligible correc- T E M P 2 = calculated weight-average
tion for imperfect resolution can be checked by injecting molecular weight
other broad MWD standards to test the validity of the
GPC calibration curve over a range of molecular weights. T E M P 5 = calculated viscosity-average
I n this program valu'es of C1 and C2 are guessed by molecular weight
the search and the calibration curve is used to convert
the GPC chromatogram to a differential MWD, assuming The computer program uses a Rosenbrock search as
infinite resolution. This differential M W D is then used described by Wilde (1964). The search is restricted to
to calculate the appropriate molecular weight averages the desired range of possible C1 and C2 by a "high failure"
(M,,, M,, or M u ) . One of the following objective functions technique.
is minimized. Nonlinear Calibration Curve. A nonlinear calibration
curve might be expressed as

FMAX2 = (MW(1) - T E M P 2 ) 2+ (MN(1) -


TEMP1)' (1)
FMAX2 = (MN(1:I - T E M P l ) ' + (MN(2) -
The search program can be easily modified to search for
TEMP1)' (2a) N constants given N bits of information. This involves
N / 2 to N broad MWD standards. An alternative approach
FMAX2 = (MW(1) - T E M P 2 ) *+ (MW(2) - might involve the search for straight-line segments of
TEMP2)' (2b) the nonlinear calibration using the existing program. No
attempt has been made here to evaluate this procedure.
FMAX2 = (VMW(1) - T E M P 5 ) ' + (VMW(2) - Effective Calibration Curves. An effective calibration
TEMP5)' (212) curve may be obtained where correction for imperfect
resolution is not negligible. This involves using the calibra-
tion curve constants obtained by search on the imperfectly
FMAX2 = (MW(1) - TEMP2)" (MiX(2) - resolved chromatogram of a broad standard only for
TEMP1)' (2d) unknowns with chromatograms similar to that of the

Table II. Input Data for Program Test


Search Parameters Initial
~ Estimate
Run Sample- First Sample Sample Second Sample Iststep, Accel, Deccel, ~

No. No. M, MI' ML No. M" Ma M, CC . A BE CI C2


1 17 64,430 125,800 0.005 2.5 -1 20 15
2 3A 333,000 430,000 0.005 1.5 -1 20 15
3 7A 48,000 3A 333,000 0.005 1.5 -1 50 6
4 7A 48,000 3A 333,000 0.002 2.5 -1 50 6
5 108 197,000 3A 333,000 0.005 1.5 -1 20 15
6 103 107,100 17 125,800 0.005 2.5 -1 80 5
7 103 107,100 17 125,800 0.1 2.5 -1 50 6
8 103 107,100 17 125,800 0.005 2.5 -1 50 6
9 103 107,100 17 125,800 0.005 1.5 -1 50 6
10 103 107,100 17 125,800 0.005 2.5 -1 20 15
11 103 107,100 17 125,800 0.05 2.5 -1 20 15
12 103 107.100 17 125,800 0.05 3.5 -1 20 15
13 7A 53,000" 3A 413,000" 0.005 1.5 -1 20 15
14 3A 333,000 17 125,800 0.005 1.5 -1 50 6
15 3A 333,000 17 125,800 0.005 1.5 -1 20 15
' a of 0.70 in Mark-Houwink-Sakurada relation used.

VOL. 8 NO. 1 M A R C H 1 9 6 9 55
standard. The effective calibration curve constants
Table 111. Results of Program Test automatically accomplish necessary resolution correction.
output Another possibility is to use a series of broad standards
Cl c2 over the range of elution volumes desired and combine
(actual (actual the linear segments so obtained to obtain a complete
Run value value No. of APP~X effective calibration curve.
No. = 37.55) = 3.62) iterations time Remarks
A nonlinear calibration curve search would provide even
1 37.54 3.62 166 2 min. more flexibility in the above procedures. None of the
28 sec. above are evaluated in this paper.
2 37.56 3.62 409 4 min.
8 sec. Evaluation
3 37.52 3.61 209 3 min.
36 sec. I n this evaluation samples of high and low polydispersity
4 41.34 4.32 195 3 min. Slowb were used (Table I). Maximum computer time permitted
27 sec. per run was approximately 6 minutes. Previous programs
5 37.43 3.60 348 4 min.
0 sec. using a Hooke and Jeeves pattern search had required
6 72.31 9.42 198 4 min. Very slowb 7 to 30 minutes to reach an optimum. Tables I1 and
21 sec. I11 show that, in general, convergence to the correct
7 47.71 5.59 136 4 min. Very slowb optimum (&0.7%) occurred after about 200 iterations
44 sec.
8 49.79 6.00 58 5 min. Very slow* within approximately 3% minutes using an initial step
11 sec. size of 0.005, an acceleration factor of 2.5 or 1.5, and
9 49.79 6.00 219 5 min. Very slow6 a deceleration factor of -1.
1 2 sec. I n the only exceptional case slow convergence occurred
10 42.11 4.50 250 6 min. Very slowb
15 sec.
when two very broad molecular weight samples, each of
11 40.66 4.22 219 4 min. Lax criterion" known but of almost equal weight-average molecular
6 sec. weights, were used. These samples were examined in runs
12 36.74 3.47 260 5 min. Slow6 6 to 13. In none of these instances was the search criterion
6 sec. for compfetion, Y1 - Y2 = 0.1, reached although values
13 37.53 3.62 470 5 min.
30 sec. as low as Y1 - Y2 = 2.0 were obtained a t values far
14 37.35 3.58 234 3 min. from the correct C1 and C2-that is, the criterion was
39 sec. almost satisfied far from the optimum of these runs. I t
15 37.33 3.59 403 5 min. was evident from examination of the calculated molecular
16 sec.
weights that the objective function was not satisfied. This
situation was attributed to the automatic step size reduc-
tion in the search, which usually resulted in very rapid
"Search criterion was 20 rather than 0.1. bSearch criterion of
objective function variation (Y1 - Y2) = 0.1 not reached. convergence. Attempts to remedy this situation by
changing the three search parameters, the search criterion

Table IV. Input Data


Polydisperse PVC sample 1
GPC response
Counts 23.0 23.5 24.0 24.5 25.0 25.5 26.0 26.5 27.0 27.5 28.0 28.5 29.0
Height 0.0 0.6 1.9 4.8 11.9 24.2 43.9 66.1 74.2 111.1 118.8 117.8 108.8
Counts 29.5 30.0 30.5 31.0 31.5 32.0 32.5 33.0 33.5 34.0 34.5 35.0 35.5
Height 90.9 70.8 51.6 38.7 27.2 18.0 11.5 7.9 4.7 2.9 1.6 0.3 0.0
Average molecular weight
M," = 67,800
M.b = 28,200
Search parameters
CC = 0.005
A = 1.5
BE = -1
First guess of C1 = 20
First guess of C2 = 15
Output data
C1 = 48.87
C2 = 4.36
Time = 2 min. 52 sec.
No. of iterations = 305
a Measured by light scattering. Measured by osmometry.

56 I & E C PRODUCT RESEARCH A N D DEVELOPMENT


Table V. Verification of GPC Calibration Curve
P vc (MJos (MnJcPc[ ( M n ) c P C- (Mn)as] (MY)LS ( M ~ ) [ ( ~M s I~c p c~- ( M ~ L s ]loo
Sample No. (Osmometry) GPC ( M d O S (Light Scattering) (GPCI (Mi,
2 38,000 43,000 +13.2 128,000 115,000 -10.2
3 49,000 50,000 +2.1 148,000 133,000 -10.2
4 31,000 37,000 +19.4 99,000 98,000 -1.0
5 60,000 57,000 -5.0 176,000 141,000 -19.9

for convergence and particularly the initial guesses of C1 technique has been developed and used to determine a
and C2, met with considerable success. Run 12, although linear calibration curve for a wide number of cases (Tables
slow, was very near convergence when the time limit I1 and 111).
expired. The program may be very easily modified to search
Using the values of C 1 and C2 obtained with the search for parameters of a nonlinear calibration curve. Results
technique and PVC sample 1 (Table IV) the GPC results of program evaluation and specific recommendations for
calculated assuming infinite resolution (Table V) differ in its use are presented. A source program listing is available
M , from -5.0 to +19.4’% compared with osmometer values, on request.
and in M, from -19.9 to -1.070, compared to SOFICA
Acknowledgment
light scattering measurements. Considering the sources
of uncertainty in all three methods of measurement as The authors appreciate financial support from the Chi-
well as the fact that the search technique was used only nook Chemicals Corp., Ltd., and the National Research
for sample 1 (no attempt was made to optimize a C1 Council of Canada. Data on PVC from Imperial Oil Enter-
and C2 for the range of molecular weights of interest), prises, Sarnia, Ontario, and the original Rosenbrock search
these results were comidered satisfactory. program from John D. Sheel of McMaster University
are also gratefully acknowledged.
Recommendations for Use crf Search Program
literature Cited
Use search parameter values of
Balke, S. T., Hamielec, A. E., Reprints, Sixth International
CC = 0.005 Seminar, Gel Permeation Chromatography, Miami
Beach, Fla., 1968.
A = 2.5 Cantow, M. J. R., Porter, R. S., Johnson, J. F., J .
BB = -1 Polymer Sci. 5 , Part A-1, 1391 (1967).
Duerksen, J. H., Hamielec, A. E., J . Polymer Sci. 6,
Use a t least two different sets of initial guesses of C1 Part C, 83 (1968).
and C2 to test each case. Frank, F. C., Ward, I. M., Williams, T., Reprints, Fifth
International Seminar, Gel Permeation Chromatog-
Use a search criterion of Y1 - Y2 = 0.1 and a computer raphy, London, 1968.
time limit of 6 minutes.
Rodriguez, F., Kulakowski, R., Clark, 0. K., IND. ENG.
Adjust initial guesses of C1 and C2 to speed convergence. CHEM.PROD. RES.DEVELOP. 5,121 (1966).
Altering search parameters, or criterion for convergence, Wilde, D. J., “Optimum Seeking Methods,” p. 151,
are less effective solutions to the problem. Prentice-Hall, Englewood Cliffs, N . J., 1964.
Conclusions
A program to determine a linear calibration curve from
polymer samples of known weight number- or viscosity- RECEIVED
for review May 20, 1968
average molecular we:ights utilizing a Rosenbrock search ACCEPTED November 4, 1968

VOL. 8 NO. 1 M A R C H 1969 57

You might also like