GPC Trace Analyzer

Compute Mₙ, Mw, Mz and Đ from a raw chromatogram, and see how far they move when you move the limits

What this does

Every other molecular weight tool on this site asks you for Mₙ and Mw as though those were things you measured. They are not. They are things your chromatography software computed, from a trace, using a calibration and a pair of integration limits somebody chose. This page does that computation in the open, on your own data, so you can see what the choices were worth.

The arithmetic is the standard slice summation. A GPC concentration detector reads a signal proportional to the mass of polymer eluting at each moment, so after subtracting a baseline, each slice i contributes a mass hi at a molecular weight Mi read off the calibration:

Mₙ = Σhi ÷ Σ(hi/Mi)     Mw = Σ(hiMi) ÷ Σhi     Mz = Σ(hiMi2) ÷ Σ(hiMi)

Nothing in those three expressions is difficult. What makes GPC numbers soft is everything around them: where the baseline sits, where the integration starts and stops, and whether the calibration was ever measured in the region you are reading. This page reports all three alongside the answer instead of behind it.

Analyze a trace

Chromatogram

Column 1 is always the x axis. Choose the concentration detector, usually RI. The slice sums assume the signal is proportional to mass, which a UV channel is only for a homopolymer with a constant chromophore, and a light-scattering channel is not at all.

Calibration
Column bank

Any vendor's datasheet quotes the two numbers this needs: a guaranteed plate count and a linear molecular weight range. Pick the closest or enter your own.

Describe the bank this trace came off. The calibration slope and the peak volume above already belong to it, so changing this on its own asks what a better-packed bank of the same length would give, not what buying more columns would give. The square-root rule further down answers that one.

A plate count quoted per column rather than per metre becomes plates per metre by dividing by the column length in metres: 16,000 plates per 30 cm column is 53,000 p/m.

Baseline and limits

What the limits are worth

Integrate down toWindowMₙMwĐArea kept

What the column bank adds

A worked example with a known answer

The trace the page loads with is not a real injection. It was generated from an exact log-normal distribution with Mₙ = 20,000 g/mol and Đ = 1.300, sampled every 0.025 mL, and given a gently rising baseline so the baseline correction has something to do. That matters: it means the right answer is known before the page computes anything, so the gap between the right answer and the reported one is measurable rather than arguable.

Integrate the whole trace and the summation returns Mₙ = 20,000 and Đ = 1.3000, which is the check that the arithmetic is right. Then cut the tails off, the way every real integration does:

Integrate down toMₙMwĐ (4 dp)Mₙ errorĐ error
10% of peak height20,40025,4001.2440+2.2%−4.3%
5%20,20025,7001.2689+1.2%−2.4%
2%20,10025,8001.2844+0.6%−1.2%
1%20,10025,9001.2914+0.3%−0.7%
0.5%20,00026,0001.2955+0.2%−0.3%
the whole trace20,00026,0001.30000.0%0.0%

Two things fall out of that table, and both are worth carrying to a real sample.

The error only ever runs one way. Integration limits discard tail, and a distribution with its tails clipped is narrower than the one you injected. So Đ is always reported low and Mₙ always high, never the reverse. A sample integrated at 10% of peak height reads Đ = 1.244 when it is really 1.300, and nothing in the chromatogram tells you that, because the trace looks perfectly clean. This is the mechanism behind the well-known observation that two analysts integrating one raw file disagree: they are not making random errors in opposite directions, they are both clipping, by different amounts.

Đ is about twice as sensitive as Mₙ. At every row above, the dispersity error is roughly double the Mₙ error, because Đ is a ratio of two moments that move in opposite directions when you clip: Mₙ rises as the low-M tail goes, Mw falls as the high-M tail goes, and the ratio takes both hits. If you are comparing dispersities between samples, matching integration limits matters more than anything else you could do to the method.

One thing does not carry, and it is the band broadening. This trace was generated from the exact distribution, so it has no instrument in it: the peak is as narrow as the maths says and not one bit wider. Apply the column correction above to it anyway and it moves Đ from 1.284 down to 1.276, away from the 1.300 it was built from rather than towards it. That is the correction working exactly as it should on data it was not meant for. On a real injection the two effects run in opposite directions, clipping narrowing the distribution and the columns widening it, and on a trace like this one they are close enough in size to hide each other. Which is the argument for measuring both rather than hoping they cancel.

It is worth putting that next to the opposite result from the calibration converter. There, Đ is the robust number: converting between polymer scales transforms it as Đtrue = Đapparent(1+αstd)/(1+αsample), which contains no K at all, so a badly chosen calibration constant leaves dispersity untouched. Here it is the fragile one. The two are not in tension: dispersity survives a wrong molecular weight axis and does not survive a wrong integration window, because the first rescales the axis and the second throws material away.

The error is a property of your column bank

Band broadening is usually treated as an instrument constant: run a narrow standard once, note how much wider it comes out than it really is, and apply that forever. That is the right procedure, but it hides where the constant comes from. It is set by two numbers printed on the column datasheet, and you chose them when you bought the columns.

A bank of plate count N spreads a peak eluting at VR into a Gaussian of standard deviation σV = VR ÷ √N. The calibration then converts that spread in volume into a spread in molecular weight, and the exchange rate is the local slope of the calibration itself:

σ = ln(10) · B · VR ÷ √N     where B = −d(log10M)⁄dV at the peak

A Gaussian of width σ in ln M multiplies a dispersity by exactly eσ², so the number your software prints is Đreported = Đtrue · eσ², and the excess over 1 that you actually reason about picks up σ² on top of whatever the polymer really has. Note that B·VR is dimensionless, so this works unchanged whether your x axis is millilitres or minutes.

Both numbers on the right trace back to the datasheet. N is the guaranteed efficiency times the packed length. B is fixed by how many decades of molecular weight the series spreads over that same length: a mixed bed sold as covering a wider range has to fit more decades into the same separation volume, which is a steeper calibration, which is a larger σ for identical volumetric broadening.

Range you do not need is not free

Column seriesLinear range, PS equivDecadesEfficiencyσĐ inflated by

Specifications are Agilent's own, from the Organic GPC/SEC Columns product guide (publication 5990-7994EN): linear molecular weight operating range in polystyrene equivalents, and guaranteed column efficiency. Holding the bank geometry fixed, each row's slope is scaled from yours by the ratio of the stated ranges, on the assumption that columns of the same dimensions have about the same separation volume. PLgel 3 µm MIXED-E is left out because the guide states only an upper limit for it, up to 30,000 g/mol, so its span cannot be computed; being both the narrowest in range and the highest in efficiency, it would sit below every row here.

Adding columns helps as the square root

Put a second identical column in series. VR doubles and N doubles, so σV grows by √2. But the same molecular weight range is now spread over twice the volume, so B halves. The two effects do not cancel: σ falls by √2, and the excess dispersity the columns add, which goes as σ², halves outright. That is the familiar result that resolution in size exclusion scales as the square root of column length, arrived at from the dispersity side rather than the peak-width side. It is also why the fourth column is a much worse buy than the second.

The practical rule is to run the narrowest range that brackets your sample. A wide mixed bed is insurance against a sample you have not characterised yet, and it is paid for in dispersity. Once you know roughly where your polymer sits, the narrower series measures the same material more precisely, on the same instrument, with no change to the method.

Two cautions about the number this page gives you. It is a floor: the guaranteed efficiency is the column's contribution alone, measured under the vendor's conditions, and the real bank also carries extra-column dispersion from the injector, the detector cell and every length of tubing between them. A narrow standard measures the whole system and will always come out worse; when the two disagree, trust the standard and read the difference as what your plumbing is costing you. And a column that has been in service for a year is not the column on the datasheet.

What this cannot do for you

Related tools