iiinova — where your digital self keeps evolving

Pedro Bernardinelli @pbernardinelli.com · 20/05/2026

I was looking for a paper from @teddykareta.bsky.social whose details I didn't know, so I decided to Google it. Gods, I love AI search so much, it's so, so useful...

131

Pedro Bernardinelli @pbernardinelli.com · 15/01/2026

Bye, Seattle! For those who didn’t know: as of today I’m moving out of the US to start a faculty job at the University of São Paulo, in Brazil. I’m excited to be back home, but it’s sad to leave this amazing region of the US

050

Pedro Bernardinelli @pbernardinelli.com · 21/07/2025

Lots of science aside, I love this (unintentionally!) out of focus image of 3I taken on June 24

130

Pedro Bernardinelli @pbernardinelli.com · 04/06/2025

The best part of the main sorcha paper is the Acknowledgements section ( @megschwamb.bsky.social , I had forgotten we had this paragraph!)

000

Pedro Bernardinelli @pbernardinelli.com · 19/02/2025

It'd be nice to see if their two period solution reproduces our Fig 6 (which includes the same data they used, and a lot more) ui.adsabs.harvard.edu/abs/2023PSJ.... 1/2

100

Pedro Bernardinelli @pbernardinelli.com · 08/01/2025

Joao and I spend a lot more time than we should discussing how to make pretty plots. Look at this beauty!

140

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

Now, with all of these ingredients at hand, we can derive a complete set of population estimates, splitting along each property (color, dynamics). So, in our chosen absolute magnitude range, if our analysis is right, there are ~80k TNOs out there.

Table 4, showing the per-class population estimates

100

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

Now, let’s look into the inclination distributions. Here, we ask: is there evidence that indicates that, for a dynamical class, the NIRB and NIRF inclination distributions are different? In the case of the HCs, but also the Plutinos, and resonances internal to the Kuiper belt, the answer is yes!

Posterior distributions of the inclination concentration parameter as a function of color and dynamical class

100

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

We can also split the color distributions. These are essentially 1D tracks, so we can ask what happens on each half of the distribution. We see (almost) no physical differences, but the NIRB HCs are bluer than the detached NIRBs, whereas the NIRF CCs are redder than the other NIRF populations.

Figure 10, ternary diagram of occupation on the NIRB, as well as the redder and bluer halves of NIRB for the HCs, detached, and scattering objects

100

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

We can break things down even further: the Classical population goes from 100% NIRF to 100% NIRF as it increases in (free) inclination, whereas we don’t see that in scattering or detached pops, they seem well mixed — later on we’ll properly quantify this trend.

Figure 8. As before, but in bins of inclination

100

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

Now, we can achieve our initial goal: which percentage of each dynamical class is composed of each color family? CCs are 100% NIRF, whereas the detached, scattering and hot Classicals are ~70% NIRB. Resonances vary, with things inside the 3:2 and 2:1 being more like the CCs.

Figure 7, showing posteriors of NIRB occupation percentage per TNO dynamical class

100

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

Something similar happens with the absolute magnitudes: each color family in all dynamical classes have the same abs mag distribution. Even more surprisingly, these two distributions are indistinguishable from each other: there seems to be a universal TNO size distribution in our size range.

figure 9, showing the data and the cumulative H mag distribution for six different TNO dynamical classes

121

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

So we asked: are there relations between color and lightcurves? We saw dynamical differences in paper I, so we asked if colors also mattered. It turns our, colors are what drive the dynamical differences: NIRF objects, independent of dynamical class, are more variable, as are the cold Classicals.

Posterior samples for the LCA distribution, and the LCA distribution itself. Figure 6 from the paper

110

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

We took a completely different approach, and arrived at a simpler form of our Equation 8. This form neatly relates TNO orbital parameters, colors, absolute magnitudes and lightcurve amplitudes. This was when things got fun!

Equation 8, a Poisson likelihood for a general mixture model in TNO colors, sizes, lightcurves and orbital elements

110

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

Now, we wanted to ask how represented each color family was in each dynamical class. I visited Penn after Thanksgiving 2022, and Gary and I spent an afternoon brainstorming how to do this. Gary thought my GMM likelihood (Equation A8) was wrong, so we rederived it.

Equation A8: a mixture model likelihood penalized by a selection function

100

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

With this model, we can probabilistically assign a class to each object. Doing that, we found a few interesting things: the cold Classicals are all NIRF, while the dynamically excited objects are primarily NIRB. This is similar to previous works, so we’re not completely crazy.

Diagram of semi-major axes, eccentricities and inclination for all DES TNOs, color-coded by their color assignment and with symbols indicating dynamical class

110

Pedro Bernardinelli @pbernardinelli.com · 06/01/2025

Applying this GMM to our data, we found a two component model, separated by their near infrared colors. Surprisingly, this is quite similar to what our Col-OSSOS friends had recently found! We’re calling these components NIRB and NIRF: at a fixed g-r, NIRF is redder.

Figure 1 of the paper, showing 800+ colors and the result Gaussian mixture model

111

Pedro Bernardinelli @pbernardinelli.com · 26/11/2024

I can't stop looking at this image... I thought I was the one who liked playing with bad data! Congrats on the paper - I think you told me about it @ ACM last year, good to see it out there

120