The following is a list of publications where we have found incorrect ATNF data. We are using this as notes so we cant notify ATNF with a single email.
Bell_2016: their flux errors are an order of magnitude too small
Zhao_2019: found an error in a data point for J0953+0755 and J0835-4510. It looks like they input the average value, but it was off by an order of magnitude.
Mignani_2017: Off by three orders of magnitude as the paper records flux in micro Jy
Appear to be adjusted for the ATNF frequencies based on their spectral index
The reference frequency used in the models is currently constant at v=1.3 GHz for an unknown reason. To test how this affects the model fits, this should be instead calculated at the centre frequency of the frequency range of a particular fit.
Some papers include lower or upper limits on flux density measurements. This is an example of upper limits on a Jankowski 2018 plot. He plotted them but did not include them in the fit.
At some point, it may be useful to implement an option to include censored data in the fit using a statistical method such as the Tobit model. Alternatively, or perhaps as well, we could implement a feature to plot the data (with arrows indicating limits) whilst not including them in the fit, as seen in papers like Jankowski et al and Bilous et al.
The peak frequency of an LPS spectrum is not a fit parameter, however it is a useful value to know and to cite in a paper. It is calculated easily using the formula for the peak of a quadratic. We could report it separately to the fit parameters (perhaps in a dictionary alongside other values such as the AIC and p_best)
Our database lacks publications focusing on MSPs, high frequency measurements, and northern-sky pulsars. Here we are keeping a list of publications which may be useful additions to the database.
The uncertainties on the model parameters are calculated from the covariance matrix, which assumes that the likelihood is symmetric (i.e. normally distributed) around the best-fit value. This method can lead to underestimated and inaccurate uncertainties.
A more robust method would be to use a Markov Chain Monte Carlo (MCMC) algorithm to sample the probability distributions of the model parameters. It would be beneficial to implement this as an option in the software, although the higher computational cost warrants having option to not use it as well.
We currently have a working double BPL model. However, due to the lack of information available in the literature, this model has been excluded for now. It would be useful to have an option to fit the double BPL.
The Vela pulsar is a perfect example of a pulsar that may potentially be a double BPL. Here is an example fit: