Hello: Have you considered correcting both the phase and baseline? This might improve the results. Cheers

Your typical proton spectrum is not usually fully quantitative, and it is not uncommon to have discrepancies of 10-20% in the integrals (and even bigger are possible).

To get accurate integration you need to make sure that the acquisition time (AQ on bruker systems) and recycle delay (D1) are long enough so that your sample has time to completely relax between scans. Depending on the compound, the T1 relaxation of protons can be 15 s or more, and to get 99.7% relaxation you need to wait 5 * T1 between the scans. Your acquisition time should be at least 3 * T2 (i.e. transverse relaxation, which is the exponential decay you see in the FID).

I would start by increasing the acquisition time to at least 5 s and D1 to 10 or 20 s to see if that helps.

Quantitative NMR can be a challenge.

The DFT is only mathematically suitable to estimate FID signal frequencies, amplitudes, decay rates and phases when all the in information in the FID consists of exponentially decaying sinusoids. This is rarely the case. Accurate frequency-domain parameter estimates also require adequate peak separation.

The FID usually contains time-independent components (DC offsets) and errors in the initial data values. Frequency-domain, signal area estimates are particularly vulnerable to these components. This remedy is usually empirically modifying the frequency domain data to minimize the problem. Most of the time ad-hoc approach is adequate. Sometimes it isn’t.

Peak overlap in 1H spectra can come from the sample itself. Any type of Intermediate chemical exchange can broaden some of the lines. Sources other than the sample can make it difficult to achieve a frequency-domain baseline suitable for integration. The most common culprit is water contamination in solvents. For low-concentration samples, polymer impurities can have amplitudes similar to the sample’s.

Finally, the frequency response must be linear. A simple test is to take an array of spectra of a sample with a strong, single peak. The excitation frequency is arrayed +/- 0.5 ppm over the spectral width of interest. The peak amplitude should be independent of peak position. If the amplitudes at the far edge of the array are attenuated, increasing the spectral width eliminates the issue.

Dear Korban,

if have a well concentrated NMR sample, but not much NMR time, I normally acquire only a single scan (NS=1) with zero dummy scans (DS=0). This has normally good quantitative response, because insertion of the sample, temperature equilibration, shimming (watch out NOT to use 1H shimming here!) takes long enough for full relaxation of magnetisation.

BUT: do not use RGA to optimize reciever gain or automated pulse determination - all this also destroys the equilibrium of the magnetisation...

Alfred

P.S. what has been said by others (baseline, long d1,....) above is of course also good advice!

Dear Korban

all the above answers will help you to identify deviations of the integral larger 1 - 2 %.

In precision quantification you should also consider

- the isotope satellites, they account for 1 % for protons attached to carbons, but only 0.3 % for protons attached to nitrogens and typically are omitted for integration

- possible isotope exchange if working in deuterated solvents (depend on exchange rates)

- and some funny deviations in 1H intensities stemming from the enrichment of (natural) deuteration, e.g. if you use material highly purified by destillation.

All this may cause deviations up to 1 % and typically are considered as exotic but are a problem when using e.g.: di-methyl-terphthalate as standard...