GPC if the sample were monodisperse. STEP 1. I found it easiest to drop the DRI vs. Ve data into Origin one cocktail at a time. Origin has a nice feature for scanning along a dataset, as shown by the arrow below. So, I just read the Ve value for each peak and associate it with the M values supplied by the calibration standard vendors. Of course, the biggest M value is the lowest Ve (assuming the GPC is working in size exclusion mode, which is the case here). STEP 2. I like Origin for making the M vs Ve calibration plot. 7.0 6.5 6.0 5.5 log10M 5.0 4.5 4.0 3.5 3.0 2.5 2.0 15 20 25 30 35 40 Ve/mL Of course, Origin is good at fitting this, too (but maybe not—read on!). You can try various polynomial fits. 7.0 B Polynomial Fit of Data1_B Upper 95% Confidence Limit Lower 95% Confidence Limit Upper 95% Prediction Limit Lower 95% Prediction Limit 6.5 6.0 5.5 log10M 5.0 4.5 4.0 3.5 3.0 2.5 2.0 15 20 25 30 35 40 Ve/mL In this case, I show a fifth-order polynomial. Origin computes confidence and prediction limits, too. The arrows are discussed below. Confidence limit: If you did the experiment 100 times, 95 of them would fall between the green confidence limit lines drawn. Prediction limit: You have a 95% chance that the next measurement would fall between the blue prediction lines. Of course, Origin provides all kinds of info about the fit, as shown on the next page. The most important are the equation of fit, uncertainties in the various parameters to that equation, and the confidence interval data. These are highlighted in the box below. [11/23/2008 20:51 "/Graph1" (2454793)] Polynomial Regression for Data1_B: Y = A + B1*X + B2*X^2 + B3*X^3 + B4*X^4 + B5*X^5 Parameter Value Error t-Value Prob>|t| --------------------------------------------------------------------------A 184.88389 27.54607 6.71181 <0.0001 B1 -30.96298 5.38691 -5.74782 <0.0001 B2 2.13122 0.41545 5.12995 1.00793E-4 B3 -0.07323 0.0158 -4.63561 2.74872E-4 B4 0.00125 2.96308E-4 4.22659 6.41816E-4 B5 -8.53934E-6 2.19452E-6 -3.89121 0.0013 --------------------------------------------------------------------------R-Square(COD) Adj. R-Square Root-MSE(SD) --------------------------------------------------------------------------0.99949 0.99933 0.02822 22 --------------------------------------------------------------------------- N Parameter LCI UCI --------------------------------------------------------------------------A 126.48888 243.27889 B1 -42.38271 -19.54326 B2 1.25051 3.01192 B3 -0.10672 -0.03974 B4 6.24228E-4 0.00188 B5 -1.31915E-5 -3.88718E-6 --------------------------------------------------------------------------ANOVA Table: --------------------------------------------------------------------------Degrees of Sum of Mean Item Freedom Squares Square F Statistic --------------------------------------------------------------------------Model 5 24.85808 4.97162 6240.8574 Error 16 0.01275 7.96624E-4 Total 21 24.87082 --------------------------------------------------------------------------Prob>F --------------------------------------------------------------------------<0.0001 --------------------------------------------------------------------------Fit and Residual data have been saved into Data1 worksheet. STEP 3. Put the calibration results back into Excel because it’s easier to calculate in that program than it is in Origin. Best Fit Results Value Error Value Error A 184.88389 27.54607 184.88389 27.54607 B1 -30.96298 5.38691 -30.96298 5.38691 B2 2.13122 0.41545 2.13122 0.41545 B3 -0.07323 0.0158 -0.07323 0.0158 B4 0.00125 2.96E-04 0.00125 2.96E-04 B5 -8.54E-06 2.19E-06 -8.54E-06 2.19E-06 Enter a Ve in mL Best fit M 15 <--This cell is defined to be Ve 4.07E+09 <--This cell reads: =10^(M3+M4*Ve+M5*Ve^2+M6*Ve^3+M7*Ve^4+M8*Ve^5) You can put any value you wish into Ve. For Ve = 25 mL, you get M = 3110….BUT THIS IS TOO LOW. SOMETHING IS WRONG! Drawing arrows on the calibration clearly shows that Ve of 25 corresponds to 104.3 20,000. The problem is that Origin is not reporting out the data to the precision it uses internally to compute the plot. This is actually the worst mistake I have seen Origin make. Maybe it has been fixed in later versions. This figure shows the plot YOU generate from Origin’s printed fit compared to the one ORIGIN generates using the better numbers it has internally: 14 12 log10M 10 8 6 4 2 0 15 20 25 30 Ve/mL 35 40 45 So,that’s the hazard of trusting someone else for software. In the bad old days, we had our own polynomial fit routine, which undoubtedly would have done this right. Today, we have Google, which quickly leads us to a program named POLYSOLVE: http://www.arachnoid.com/polysolve/index.html PolySolve produces these results, which seem more accurate as a reasonable M is found at Ve=25: Best fit results from POLYSOLVE: http://www.arachnoid.com/polysolve/index.html 179.76398827770200000 A -29.90223290944740000 B1 2.04508951275564000 B2 -0.06979938640524170 B3 0.00118523157897783 B4 -0.00000802259892519 B5 Ve 25 Best fit M from POLYSOLVE 25650 Nice of POLYSOLVE to provide so many significant figures, even though it fails to provide error estimates. We could probably use those from Origin as a guide, but let’s just do the uncertainty graphically to get an idea. The arrow drawn in the Origin figure with confidence limits shown above gives us a low estimate of 104.3 20,000. Another arrow drawn to the upper confidence limit comes closer to 104.5 30,000. In consideration of the POLYSOLVE result, we might quote out, for Ve = 25 mL, M = 25,650 ± 5000. Almost 20% uncertainty! GPC is one of those experiments that is more precise than accurate.