Distribution Simulator

The whole molecular weight distribution your polymerization should produce, and the trace your GPC would show of it

Why a shape and not a number

The dispersity predictor tells you what Đ a controlled polymerization should reach. That is the right number to design against, but it is a summary, and two very different reactions can summarize to the same one. A living population at Đ = 1.24 and a tight population at Đ = 1.05 carrying a fifth of its chains dead both read 1.24 on a report and look nothing alike on a trace. One is a slow-exchange problem you fix with a better RAFT agent; the other is a termination problem you fix with lower radical flux. The single number cannot tell you which you have. The shape can.

This page builds the distribution from three pieces, each with a closed form behind it, and then shows you what the instrument would do to it.

Simulate a distribution

The polymerization

C in the equation is 1/Cex. Above ~20 gives Đ near 1.05; below 1 means the agent is mismatched to the monomer.

What went wrong, and what the instrument adds

As a fraction of all chains, not of mass. Dead chains are short, so they weigh far less than they count.

This is how you measure your own band broadening. A standard that is genuinely 1.00 would read 1.00 on a perfect instrument.

Hand it to the trace analyzer

This converts the simulated distribution into the two-column chromatogram a GPC would have exported, using log₁₀M = 10.5 − 0.70·V. Paste it into the trace analyzer to see what integrating it would have reported – which is the round trip: a distribution you know, turned into a trace, integrated back into numbers that no longer quite match.

Three things the shape tells you that Đ does not

Termination hides from Mw. Take a RAFT polymerization sitting at Đ = 1.05 and kill a tenth of its chains. Đ goes to 1.131 and Mₙ falls by 8.1% – but Mw moves only 1.0%, because the dead chains are short and weigh almost nothing. If you are tracking a reaction by Mw, a termination problem is close to invisible; it shows up in Mₙ and in Đ, which is a good reason to report all three rather than the one the instrument prints largest.

Band broadening multiplies Đ. Convolving with a Gaussian in log molecular weight is exact for a log-normal distribution, and it gives a closed form: Đobserved = Đtrue · eσ², where σ is the broadening width in ln M. The factor does not depend on the sample at all, which is why you can measure it once with a standard and apply it to everything. What it does to your conclusions, though, depends entirely on how narrow the sample is:

True ĐObserved ĐTrue Đ − 1Observed Đ − 1Excess inflated by
1.021.0620.0200.062208%
1.051.0930.0500.09386%
1.101.1450.1000.14545%
1.501.5610.5000.56112%
2.002.0821.0001.0828%

That table uses σ = 0.20 in ln M, a middling bank. The dispersity itself is inflated by the same 4.1% in every row – but the quantity you reason about is the excess over 1, and that is inflated by 208% for a genuinely narrow polymer and 8% for a broad one. Band broadening is a narrow-sample problem. It is also why a commercial Đ = 1.02 standard is not really telling you the column is that good, and why chasing Đ below about 1.05 by conventional calibration is chasing your own instrument.

A dead-chain tail and a broad living population are not the same distribution. Set the dispersity to 1.24 two ways – once by dropping Cex to about 4, once by killing 20% of the chains at Cex = 20 – and the summary statistic is the same while the curves separate clearly: the first is a single symmetric peak that got wider, the second is a tight peak with a shoulder running down to low molecular weight. On a real trace that shoulder is the thing to look for, and it is the reason to keep the low-molecular-weight end inside your integration limits rather than clipping it off, which the trace analyzer shows the cost of.

What this model does not include

Related tools