A New and Complete Rewrite
This page is intended to assist PeakFit v4 users in understanding the revisions that are present in PF Chrom v5. We begin with the understanding that two decades have passed since PeakFit was last updated and that PF Chrom v5 represents a complete rewrite of the PeakFit program for chromatographers.
Revised PeakFit Chromatographic Models
The HVL and NLC models remain an important part of PF Chrom v5 and were revised to consist of full range functions. Significantly, the NLC was also recoded as a specialization of the closed-form generalized NLC. Without the Bessel function integrals as used in PeakFit v4, the NLC is computed in PF Chrom with the same speed as the HVL.
The EMG, the model probably used historically far more in chromatographic modeling than any other, also remains in PF Chrom, but we suggest you regard it as obsolete in the current state of the science. As a PeakFit user, where the EMG was often the function of choice, you may disagree, but we will assert that we have never seen real-world chromatographic data where the core peak was truly a Gaussian and where the IRF was a single exponential decay.
New Core Chromatographic Models
PF Chrom offers all new chromatographic functions which are statistical generalizations of the HVL and NLC. These are once-generalized models which offer third moment adjustments, and twice-generalized models which add a fourth moment adjustment, important in HPLC gradient peak fits and in preparative peaks where a significant overload is present.
The once-generalized models, the GenHVL for diffusion-type statistical parameters, and the GenNLC for kinetic parameters, may be the only models you ever need in PF Chrom if you are only fitting analytic peaks. You will need the twice-generalized Gen2HVL or Gen2NLC only if you are addressing the additional complications of gradient separations or high overload.
The GenHVL and GenNLC fit identical shapes and differ only in whether a statistical or kinetic parameterization is used. Similarly, the Gen2HVL and Gen2NLC fit identical shapes and differ only in the parameterization. You may wish to refer to the information on GenHVL and GenNLC equivalence to better understand how a generalized chromatographic model can accommodate both GC and LC theory.
The revisions in the core models can be understood most easily by understanding chromatographic models in the context of a zero-distortion density or ZDD, the peak shape at near zero-concentration or near-infinite dilution. To fully understand the PF Chrom statistical generalizations, please look at the ZDD Concepts topic. This is central to the new fitting in PF Chrom and was never present in the PeakFit product.
Accurate Instrument Response Functions
When PeakFit was first written, the Fourier science of system identification was hardly where it is in present time. When you used PeakFit to fit an EMG, you fit an instrument response function or IRF, a first order exponential decay which convolved a core Gaussian peak. In an EMG fit, the simplest IRF and simplest peak were represented.
In PeakFit, the EMG had been modified to be bidirectional, allowing it to fit both tailed and fronted shapes, even though such a left-sided convolution for fronted peaks made no theoretical sense. The EMG had essentially become a purely empirical model.
In PF Chrom, we acknowledge the need for a core peak that manages the chromatographic fronting and tailing, and a separate IRF that manages the instrument-introduced tailing. The core peak must be intrinsically fronted or tailed as fitting the concentration and the sign and magnitude of the chromatographic distortion. Further, the IRF must address all instrument effects; the first order exponential is only part of the overall picture.
Please look at the Understanding PF Chrom’s Models in order to understand the importance of a two-component IRF in PF Chrom’s fitting. An impulse-type IRF component of narrow width is essential for fitting chromatographic peaks to near-zero error.
ZDD and IRF Enhancements
More than all other aspects of PF Chrom, the addition of zero-distortion density (ZDD) higher moment modeling with full accuracy instrument response function (IRF) convolution fitting, sets the product apart from PeakFit’s modeling paradigm. The benefit of fitting an appropriate ZDD and IRF in a chromatographic model is a many orders of magnitude reduction in error and the ability to characterize higher moment aspects of the chromatographic separation as well as all instrumental distortions. You will actually be able to map deterioration in column performance from fits of your standards.
Even if you don’t choose to explore the subtleties of the ZDD and IRF enhancements, we recommend that you do the HPLC Column Health and Overload tutorial.
True Deconvolution in Fitting
Even if you are an experienced PeakFit user, this will likely be new to you. Many chromatographers have defined deconvolution as the separation arising from fitting overlapping peaks or the estimation of small peaks hidden in the rise and decay of larger peaks. While this remains an important part of PF Chrom’s modeling, there is also true deconvolution.
The Deconvolution Levels available in the Review are specific to PF Chrom. For a single peak, you can see the fit of the peak as eluted, but you can also visualize the peak absent all instrumental distortion, as if the instrument added no distortion. Further, you can see that peak absent all ZDD non-idealities (which certainly address multiple-site adsorptions). You can even see that pure theoretical HVL or NLC peak as the Gaussian or Giddings peak that would exist at infinite dilution, the concentration-independent peak.
A Far More Capable Fourier Deconvolution
In PeakFit v4, there were separate options to remove an exponential or Gaussian instrument response from the data prior to fitting.
For PF Chrom v5, there is a single new Fourier Deconvolution procedure with a large number of new IRFs and a genetic optimization procedure for alternate estimation of the instrument distortions. The magnitude of new technology to properly manage IRFs is vastly greater. We encourage you to explore the IRF Deconvolution and Fitting tutorial.
Fitting HPLC Gradient Peaks
By having the strength of the gradient change in chromatographic separations, the complexity of the modeling problem dramatically increases. The modeling in PeakFit was never capable of managing gradient peaks. In PF Chrom, we have found three highly effective approaches. If you are fitting gradient peaks, we suggest that you explore all three of the tutorials specific to these types of peaks:
Fitting Preparatory Peaks
Similarly, PeakFit never had any capability of modeling peaks with even a modest measure of overload. The unusual preparatory peak shapes were impossibilities. PF Chrom’s twice generalized models, with the simple addition of a fourth moment parameter, will allow you to see the peak that would exist if your column had infinite capacity, where no overload was present.
If you are using preparative chromatography we encourage you to explore the Fitting Preparative (Overload) Peaks tutorial. You should be able to fully characterize every aspect of the the preparative process, including the column and instrument health since this will accurately estimate the measure of overload present in any separation.
Advanced Fitting and Hidden Peak Detection
This was always PeakFit’s strength. In addition to enabling this technology to be simultaneously applied to as many as 25 distinct data sets, PF Chrom has adopted immense revisions in the fitting process. In PeakFit, there was one progression from a set of starting estimates to a final solution. If a fit did not reach a satisfactory endpoint, it was necessary to manually adjust parameters and restart the fitting.
In PF Chrom, we sought to make those restarts a thing of the past. In the course of a PF Chrom fit, you may see upward of a hundred or more discrete fits, each with a specific tailored purpose. These were designed to manage the ZDD and IRF innovations in the new models. When a fit is deemed finished by the program you should have the global minimum or values exceptionally close to such, with no intervention.
PF Chrom also employs Fourier fitting for the IRF-bearing convolution models, and uses all cores and threads in your machine to maximize performance. PeakFit was a single core product with no Fourier fitting functionality for non closed-form convolution models.
For hidden peaks, PF Chrom adds a new layer of automation where the segments from the baseline processing are treated as separate data sets, each possibly having their own overlapping or hidden peaks which will share parameters with that segment. Since fitting overlapping and hidden peaks is one of the principal reasons this type of software is purchased, we strongly encourage you to explore the Fitting Hidden Peaks tutorial.
The Subtract Baseline procedure has been completely rewritten to offer far more extensive baseline preprocessing. In the early PeakFit era, we recommended users fit baselines together with the peaks instead of removing them prior to fitting. We no longer recommend this.
For PF Chrom v5, we strongly encourage you to subtract the baseline prior to fitting. There is now a non-parametric baseline fitting algorithm that will manage virtually any form of baseline. We recommend that you explore the Chromatography Peak Modeling tutorial. In it you will see the importance of the non-parametric baseline fitting. When one is fitting the final peak data to a few ppm error, the baseline correction step is not a matter to take lightly. The difference can be as great as that of 100 ppm error for a sloppy baseline correction vs. that of 5 ppm for a meticulous one. This makes a difference when higher moments or instrument parameters are being estimated as part of a determination of column health.
Finally, as a PeakFit user, PF Chrom will appear to be an entirely distinct product with respect to the UI. The main software interface was rewritten to simplify and streamline the process when processing and fitting up to twenty-five different data sets simultaneously. PeakFit was designed to fit one data set a time.
You will see the AISN graphing engine largely intact with only a few additions. The most obvious will be the 3D graphing additions in the Visualization procedure, and in the Explore option for evaluating experimental studies. The graphical enhancements are appreciable ones, such as being able to rotate 3D surfaces in real time.
If you are a long-time PeakFit user, we suggest having a look at The Quest for a Universal Chromatographic Fit. You will likely have a sound statistical foundation to understand our motivations for building PF Chrom.