GPC Peak Interpretation

Basic principles

GPC (gel permeation chromatography, a type of SEC or size exclusion chromatography) separates chains by hydrodynamic size, not directly by molecular weight or chemical identity. A column packed with porous beads lets small chains diffuse into the pores and take a longer path through the column, while large chains are excluded from the pores and elute first. More precisely, the small chains are not traveling a longer physical route; they spend extra time diffusing into and back out of the stagnant solvent held inside the pores, while the excluded large chains ride the faster flow in the interstitial volume between the beads. The size that decides this is the chain's hydrodynamic volume, which is why the same mass can elute at a different time for a different chemistry. A detector, usually refractive index, sometimes UV or light scattering, records signal against elution time, producing a chromatogram.

To turn that time axis into a molecular weight axis, the instrument needs a calibration curve, typically built by injecting a series of narrow molecular weight standards (often polystyrene) and fitting log(M) against retention time or volume. Every molecular weight your GPC reports for an unknown sample is only as good as that calibration curve and how closely the unknown's hydrodynamic behavior matches the standards used to build it. This is why the numbers are usually called a polystyrene equivalent molecular weight rather than a true molecular weight unless you are running true light scattering detection (MALS) or have converted using Mark Houwink parameters for your actual polymer.

Column selection

Pore size distribution sets the usable molecular weight range. A column optimized for small oligomers will let every large chain coelute at the exclusion limit with no separation between them, while a column built for very high molecular weight polymer will crowd small molecules together near the total permeation limit. Mixed bed columns blend several pore sizes to give a wide, reasonably linear calibration across several orders of magnitude, at some cost to resolution compared with a column matched tightly to your expected range.

Connecting two or three columns in series extends the usable range and improves resolution, at the cost of longer run times and higher back pressure. Smaller particle size packing gives sharper separation but needs more pump pressure and is less forgiving of particulates, which is why a guard column upstream is cheap insurance against slowly fouling the analytical columns. Column chemistry also has to match your solvent and sample: polystyrene divinylbenzene columns are the default for organic solvents like THF, while aqueous or polar samples need columns built and packed for water or DMF/DMAc service, since running the wrong solvent through the wrong packing can shrink, swell, or damage it.

Column banks in practice

Almost nobody runs analytical GPC on a single column. Two or three columns in series (a column bank) is the standard way to tighten resolution: each column adds its own pore volume, so the same difference in molecular weight is spread across more elution volume. The gain follows a square-root law. Band spreading adds in quadrature while separation volume adds linearly, so n matched columns give roughly √n better resolution for n× the run time and n× the back pressure. Two matched columns is the workhorse compromise; a third buys about 22% more resolution over two, and going past three is rarely worth the hour-long runs.

The columns in a bank only help each other where their selectivity windows overlap, which is why a matched bank beats a mixed grab-bag. Two identical mixed bed columns double the pore volume across the whole range, which is why 2× mixed bed (C/D class) is the default general-purpose bank for synthetic polymers from a few thousand to a couple million g/mol. Stacking individual pore sizes that bracket your sample (100 Å + 500 Å + 10³ Å for oligomers and prepolymers below ~30k; 10⁴ + 10⁵ + 10⁶ Å for high molecular weight work) resolves that region more sharply than any mixed bed, at the cost of a narrower usable window.

A mismatched column is worse than a wasted slot. A column whose pores are all wrong for your sample (say a 100 Å oligomer column in a bank running 500k material) contributes zero separation for your peaks but still contributes its full share of band broadening, dead volume, and back pressure. Resolution actually drops compared to running without it. The same logic explains the kinked calibration curve of a badly assembled bank: in the molecular weight region where one column has stopped separating and the next has not picked up, the combined curve flattens and everything in that window piles up poorly resolved.

Elution order does not depend on which column comes first; convention runs decreasing pore size downstream, with the guard column always first, where it can sacrifice itself to particulates. Every column you add also adds spreading from the connections, so use short, narrow-bore couplers. When switching solvents on a bank, flush at low flow and be patient: the packing in every column has to re-equilibrate, and a bank holds two or three columns' worth of the old solvent. And calibrate the bank as a whole. A calibration built on one column arrangement is invalid the moment you add, remove, or swap a column.

Plan a method from the polymer

Column choice is downstream of three things: whether the polymer dissolves, whether it carries charge, and how wide its molecular-weight distribution is. Set those and this names a starting method – mobile phase, salt, packing chemistry and a bank of columns – which you can then load straight into the simulator below to check the range is actually covered. It is a starting point to take to your column supplier's compatibility table, not a substitute for it.

Try it: build a column bank

Pick up to four columns and watch the combined calibration curve, the separation of a chosen peak pair, and the costs. The model uses typical 300 × 7.5 mm analytical columns (polystyrene-equivalent molecular weights, 1 mL/min): each column adds its pore volume where its pores are the right size, band spreading adds in quadrature, and resolution R_s = ΔV_e / 4σ. Values are representative, not a substitute for a vendor's data sheet; the scaling behavior is the point.

Combined calibration curve (bold) and each column's own curve (faint). A flat stretch in the bold curve is a dead zone: molecular weights there co-elute.

Simulated chromatogram of the two peaks on this bank. Baseline resolution needs R_s ≥ 1.5.

Real column banks people actually run

Here are real, currently sold series banks and the window each one targets. Two design philosophies show up. A stack of individual pore sizes concentrates resolution across a defined window and resolves it sharply, which is what you want once you know roughly where your sample sits. A stack of mixed bed columns (each already a blend of pore sizes) trades some of that sharpness for a wide, near-linear calibration and is the safe default for unknowns or broad distributions. A guard column always goes first. The molecular weight ranges below are the manufacturers’ published linear operating ranges in the standard eluent, polystyrene-equivalent for the organic banks and PEG/PEO- or pullulan-equivalent for the aqueous ones. They are the region of shallowest calibration slope, not hard exclusion limits, and the analytical columns are the usual 300 mm length (7.5 mm i.d. for Agilent and Tosoh, 7.8 mm for Waters). Confirm against the current data sheet before you buy.

Organic phase (THF or chloroform, calibrated with narrow polystyrene standards):

Aqueous phase (water with salt/buffer, for water-soluble polymers; calibrated with PEG/PEO or pullulan, not polystyrene, which does not dissolve in water and adsorbs to the packing):

A few patterns worth noticing in the table. The mixed bed banks (PLgel MIXED, TSKgel GM, Styragel HR 5E) all cover several decades from a single linear calibration, which is why they are the go-to when you do not yet know the distribution. The matched individual-pore stacks (the Styragel HR and TSKgel G-series rows) cover a narrower window but resolve it more sharply, which is the point of choosing them. And the aqueous ranges are meaningfully lower than they look next to the organic ones, because a PEO or pullulan chain of a given molecular weight is a different hydrodynamic size than a polystyrene chain of the same mass, so the very same column reports a different number depending on the standard, which is exactly the standards-dependence covered next.

Targeting a molecular weight window: standards, cutoffs, and lab-to-lab spread

Modern instruments are sold as banks aimed at a target window. A GPC configured for polyolefins, for controlled radical products, or for oligomer analysis is really the same hardware with a different set of series columns whose pore sizes bracket the molecular weights the lab expects to run. That is the practical way to think about method setup: decide the window your samples live in, then stack columns whose selectivity overlaps across that window, so the pore volume, and therefore the resolution, is concentrated where your peaks actually elute. Two matched mixed beds give a wide general-purpose window; a bank of individual pore sizes tuned to the target (500 Å + 10³ Å for a 5k–50k RAFT polymer, 10⁴ + 10⁵ Å for a 100k–1M free radical product) resolves that window more sharply at the cost of everything outside it. The bank builder above shows both behaviors.

Calibration kits are narrow-dispersity standards, usually made anionically, in sets of ten or so spanning roughly 500 to a few million g/mol; you run them, fit log M against elution volume, and every sample is then read against that fit. They are that narrow because living polymerization (usually anionic), run under clean conditions, approaches a Poisson distribution with Đ below about 1.1 to 1.2, narrow enough that each standard reads as effectively a single molecular weight; polystyrene, PMMA, poly(2-vinylpyridine) and others are all made this way specifically for use as SEC calibration standards. Polystyrene is the default in THF, toluene, and chloroform: cheap, stable, available in the narrowest dispersities. PMMA standards are the workhorse for polar eluents, DMF, DMAc, and HFIP, where polystyrene is a poor choice because it interacts with the column packing or is not even soluble, which is why polyamides and fluoropolymers run in HFIP are almost always reported against PMMA. The number your software prints is therefore a PS-equivalent or PMMA-equivalent molecular weight: the M a polystyrene or PMMA chain would need to reach the same hydrodynamic size, not the true M of your polymer. The two scales disagree with each other, and both disagree with the true value, by amounts that depend on the chemistry; the GPC Convert page translates between scales when Mark–Houwink parameters are known. Whichever set you use, the standard identity, eluent, and column set belong in every report; a molecular weight without them is not reproducible information.

Cutoffs silently truncate a distribution. A bank only separates between its total-permeation and exclusion limits, and the calibration fit only spans the standards you actually ran. Three failure modes follow. Chains larger than the exclusion limit are not separated at all: they co-elute in a stack right at the exclusion volume, so a high-M tail or a crosslinked fraction shows up as a sharp spike at the front of the peak, and the M_w your software integrates under it is capped at whatever the calibration says at that volume, underreporting the true high end and flattering Đ. At the other end, oligomers and residual monomer crowd into the total-permeation region among solvent and system peaks, where the integration choice decides whether they count, moving M_n substantially. And in between, software will happily extrapolate a polynomial calibration fit past the last standard you ran; a value read beyond the highest standard is an extrapolation, not a measurement, and third-order fits can bend sharply out there.

Send the same sample to two competent labs and the reported M_w values routinely come back 10–20% apart, with M_n and Đ spreading even wider, and that is with everyone doing their job properly. The gap is systematic, not sloppiness: different column banks and their band broadening, different standard sets and lots, PS versus PMMA scales, different calibration fit orders, different integration limits, and percent-level flow rate differences all stack in one direction or another. Repeatability on a single instrument, same day and same calibration, is far tighter, a few percent, which is exactly why GPC molecular weights should be treated as comparative numbers. Track trends on one instrument against one calibration, report the full method, and when an absolute number genuinely matters, use light scattering detection or an absolute technique rather than arguing over whose conventional calibration is right.

Solvent system

The mobile phase has to fully dissolve your polymer with no aggregation, and it has to be compatible with the column chemistry and detector. THF is the default for many organic soluble polymers because it is a good general solvent, low viscosity, and works cleanly with refractive index detection.

Polar or hydrogen bonding polymers (polyamides, some polyurethanes, polyacrylonitrile, polyacids) often need DMF or DMAc with a lithium salt such as LiBr added, since the salt screens ionic and hydrogen bonding interactions with the column that would otherwise cause tailing or irreversible adsorption. Aqueous SEC for water soluble polymers usually needs added salt too, for the same reason: without it, polyelectrolytes and charged polymers can show anomalous elution from electrostatic interaction with the column packing rather than pure size exclusion.

Flow rate stability matters more than people expect, since the whole molecular weight axis is built from retention time. A flow rate that drifts between your calibration run and your sample run shifts the apparent molecular weight even though nothing about the sample changed, which is why many setups include an internal flow marker. Finally, remember that Mark Houwink parameters (see the calibration tool on this site) are solvent and temperature specific, so a K and α measured in THF do not apply to a DMF or aqueous run.

Try it: flow rate drift and calibration

A calibration curve is built once, from a set of standards run at a known flow rate. If your sample runs at a slightly different flow rate, even by a percent or two, its peak lands at a different retention time than it "should," and reading that shifted time off the original calibration line gives the wrong molecular weight, even though the polymer itself never changed.

Peak shouldering

A shoulder or a poorly resolved second peak usually means there is a second population of hydrodynamic sizes in the sample, but the cause could be chemical or it could be an artifact of the measurement.

Real causes include chain coupling by radical termination (combination roughly doubles the chain length for the coupled fraction, showing up as a smaller peak near twice the molecular weight of the main peak), a blend of two batches or a chain extension that did not fully consume the first block, star or branched contamination alongside linear chains of similar mass but different hydrodynamic volume, or a genuinely broad or multimodal distribution from the polymerization chemistry itself.

Artifacts include physical aggregation (chains sticking together in solution, especially near their solubility limit or if the sample was not fully dissolved before injection, showing up as a small hump at an unrealistically high apparent molecular weight), column overloading from injecting too much mass at once, which broadens and can skew a peak with no real change in the underlying distribution, and residual low molecular weight material such as unreacted monomer, initiator fragments, or a system peak eluting at the low molecular weight end.

A useful check is to run the same sample at two different injected concentrations. Aggregation and overloading artifacts are concentration dependent and shrink or shift at lower loading, while a real second population stays put.

Correct way to integrate peaks

Where you set the baseline and the start and end cutoffs has a large, and often underappreciated, effect on the reported M_n, M_w, and Đ, because the two averages are not equally sensitive to the same parts of the distribution.

M_n weights every chain equally, so even a small amount of area at low molecular weight, unreacted monomer, a short chain population, a poorly resolved solvent peak, pulls M_n down hard, since it is essentially an average of 1/M. M_w weights by mass, so a small amount of area at high molecular weight, an aggregate shoulder, a coupling peak, can pull M_w up disproportionately even though it represents only a few percent of the sample by chain count. A worked case makes the asymmetry concrete: a mixture that is 95% by weight at M = 10,000 and just 5% by weight at M = 100 comes out at M_w = 9,505 but M_n = 1,680, so the tiny low mass fraction barely touches M_w yet collapses M_n to about a sixth of the real bulk value. That is exactly why where you cut the low molecular weight baseline matters so much for M_n. Try this yourself in the demo below.

Use a consistent baseline method (usually a straight line from a stable point before the peak to a stable point well after it, not a curve that follows drift) across every sample you plan to compare. Decide up front whether a shoulder is part of the real distribution or an artifact you intend to exclude, and apply that decision consistently rather than choosing cutoffs sample by sample to make the numbers look better. Report where you set your integration limits, since two people integrating the same raw chromatogram differently can get meaningfully different Đ values from identical data. The trace analyzer puts a number on it for your own chromatogram, and the direction is always the same: clipping narrows a distribution, so tighter limits always report a lower Đ – a sample that is truly 1.300 reads 1.244 when integrated down to 10% of peak height.

Try it: peak shape and integration limits

These are idealized, simulated distributions built from Gaussian components in log(M) space for illustration, not raw instrument data. Pick a scenario, then drag the integration limits and watch M_n, M_w, and Đ recalculate against the full curve versus your selected range. For a sense of scale, theory pins a few Đ values: a living or well controlled polymerization approaches Đ = 1 + 1/X_n, about 1.0 to 1.1; a conventional free radical polymer runs about 1.5 when chains join by combination and 2.0 for disproportionation or chain transfer; and a step growth polymer taken to high conversion tends toward 2.0. So a measured Đ near 1.05 reads as well controlled, 1.5 to 2.0 as conventional, and much above 2 says something extra broadened the distribution.

Try it: upload your own trace

Upload a screenshot or photo of a chromatogram and get a rough, automatic read on its shape: how many peaks it has, whether there's a shoulder, and which direction any tailing points. This runs entirely in your browser using pixel analysis, not a trained model. It reads shape only, not real M_n/M_w/Đ values, and your image is never uploaded anywhere.

Inconsistencies between users and GPC setups

Absolute molecular weight numbers from GPC are much less portable between labs than people assume, even for the same physical sample. The single biggest source of disagreement is calibration standard identity: unless everyone is running true light scattering detection (MALS), the reported M_n/M_w is a polystyrene equivalent (or whatever standard was used) value, not necessarily the polymer's true molecular weight. Two labs calibrated against different standards, or standards from different suppliers or lots, will report different numbers for the same sample even on identical instruments. Use the GPC Calibration Converter on this site to translate between them when you know both polymers' Mark Houwink parameters.

Detector choice adds another layer. Refractive index detection responds to concentration and to each polymer's specific refractive index increment (dn/dc), UV detection responds to chromophore content, and light scattering responds to molecular weight and concentration directly, so a copolymer or a polymer with an unusual dn/dc can give different apparent distributions on different detector types even on the same column.

Band broadening, meaning peak spreading caused by the instrument itself from tubing, fittings, and column efficiency, varies between instruments and increases the apparent Đ of a narrow sample, so a column and system combination in poor condition can make a genuinely narrow, well controlled polymerization look broader than it is. Flow rate reproducibility, column aging, mobile phase batch variation, and even ambient temperature control all shift retention time calibration in ways that are usually small individually but add up.

None of this means GPC data is unreliable, but it does mean comparing M_n/M_w/Đ values across different labs, instruments, or runs months apart on the same instrument should be done cautiously. As quantified in the standards-and-cutoffs section above, a 10–20% spread in reported M_w between two well-run instruments is normal, not a red flag. Relative comparisons, same instrument, same calibration, same day, are far more trustworthy than absolute ones.

Multiple detector methods

Everything above assumes a single concentration detector (refractive index or UV) and a calibration curve built from standards. Adding more detectors changes what the instrument can measure directly instead of by comparison.

A light scattering detector (MALS) measures molecular weight directly from how much light a chain scatters, with no calibration curve or standards needed at all, though it does need an accurate refractive index increment (dn/dc) for your specific polymer to convert scattering intensity into a molecular weight. A viscometer measures intrinsic viscosity directly, which combined with light scattering gives a true Mark Houwink plot for your actual sample and can reveal branching, since a branched polymer has a smaller hydrodynamic size, and therefore lower intrinsic viscosity, than a linear chain of the same true molecular weight. Combining a concentration detector, a light scattering detector, and a viscometer (often called triple detection) gives absolute molecular weight, size, and branching information without relying on a polystyrene equivalent conversion at all.

The tradeoff is practical, not fundamental: light scattering signal is weak for low molecular weight material and very sensitive to dust and filtration, viscometers need careful baseline and pressure stability, and both need a correct dn/dc to mean anything. That low molecular weight weakness has a physical basis with a number attached: light scattering intrinsically reports a weight-average molecular weight (the signal grows with molecular size), so it becomes more accurate the larger the chains and hits a floor around 5,000 to 10,000 g/mol, below which the scattered signal is too small to measure reliably. MALS is therefore least dependable at the low molecular weight end, exactly where oligomer and residual monomer peaks live. For routine relative comparisons within one lab, conventional calibration is usually good enough. For a true, absolute molecular weight, especially on a branched or unusual polymer, multiple detector methods are worth the extra setup.

Troubleshooting quick reference

A fast lookup for the discussion above. Start here, then read the relevant section for the full reasoning.