Intro to Scintillation Studies by AGSL
AGSL defines scintillation as “The white or colored sparkles that are seen when the observer and/or the light source and/or the diamond moves.” Using their ray-tracing engine and computer muscle the lab is beginning to quantify scintillation events in a diamond by number, location and size. Since scintillation is dynamic AGSL has started to add animated analysis in planning for future grading systems.
Click on this link
to see a screen capture of a 9mm Tolkowsky close to face-up: There are seven graphics. The first three are ASET 30, ASET 40 and Fire Map (covered in prior threads). The next three show Small Scint Events, Medium Scint Events and Large Scint Events independent of light source. The last is modeled using a structured light source: A 60 degree pie sector going out to the hemisphere; light coming in 45-75 degree angular range.
This static view is only 1/45th of the overall impression these graphics give. When the animation is seen the diamond graphics rotate through 45 degrees of tilt. As they rotate you can see the different appearance of angular spectrum and dispersion in the first three maps. The last four are incredibly dynamic because they are showing where scintillation events (regardless of whether they are white or colored flashes) are taking place in the diamond.
Click on this link
for a face-up example of the difference in an extra-faceted cut, a Tolk and an OEC. Check out the number of small scint events in the Leo, the balance of small and medium in the Tolk and the predominance of large and medium events in the OEC.
1. The graphics above are one static view only. They are only 1/45th of an overall puzzle. Without rotation animation the overall meanings can’t be conveyed in one photo; these are just examples to share the “gist” of this: Even a few degrees + or – makes vast differences in the distribution and number of events seen. For example, the Tolk shows no large events in this view, but a degree or two of tilt changes that, as well as overall distribution. So it goes with the other configurations. Also bear in mind that these are simulated models with perfect optical symmetry, etc.
2. The color coding in the scintillation event graphics - yellow, cyan, magenta etc. - just identifies angular spectrum in 5 degree increments from 45 to 75 degrees. It has nothing to do with the color of the flashes.
3. Many unknowns and questions remain. Some have to do with distance (AGS is currently using 25 cm), stereoscopic vision, etc. They continue to research the frequency and visibility of scintillation events; relative to size for instance.
The best thing about this system is that it is light source independent (more below) with the exception of the final graphic which is a kind of standardized litmus test. Prior systems fire lights at a diamond and count pixels to arrive at a ‘score.’ This metric evaluates scintillation potential. Sergey Sivovolenko's work with ETAS handshakes in many ways with portions of what AGSL is doing.
This is all in its infancy, but it has been born and will be a precursor of things to come.
QUESTIONS AND ANSWERS
Q: How is this different from mechanical light boxes produced by companies who try to evaluate “fire/dispersion” and “scintillation/sparkle.”
A: There are significant differences between AGSL and prior approaches, with regard to both fire and scintillation.
Some virtual facets have a higher potential to produce fire than others because some have higher dispersion than others. Just because you see fire in a specific VF in a diamond photo it doesn’t tell you the total fire potential for that facet: A VF with high dispersion could appear white and one with low dispersion could appear colored in a given lighting environment. For example, a VF with low dispersion might be drawing light from the edge of a source, appearing colored, which would be misleading in many cases. This is just one example of the problem with assessing a diamond in a fixed lighting environment (or two, or five, or 25). It’s why AGSL has worked for some years to develop metrics and animations independent from lighting environments.
The AGSL approach to fire is to measure the dispersion of each VF directly via ray-tracing. They are not measuring the quantity/quality of the particular fire observed in one specified environment. They are assessing the potential of each point in the stone to display fire, completely independent of environment. In essence, this is factoring lighting environments out of the equation to provide equal footing to all diamonds regardless of lighting environment.
This difference is very important, since a diamond can be designed to do well in a light-source-dependent environment. Making it independent removes this kind of mechanical bias.
Q: What does light source independence mean with regard to scintillation?
A: Here’s an analogy: Suppose you’ve never played poker before but want to know how often a particular hand is dealt (3 of a kind vs 2 pair for instance).
You deal yourself a dozen hands and count how many times 3 of a kind is dealt vs 2 pair. Then you do it again, but you get a different result. Repeat and you get a third result. It becomes clear that the result is not going to be constant. The more dozens of hands you deal the closer you get to a predictive answer, but you soon realize that any hand dealt is a function of the shuffle (or in our case the lighting environment). For some shuffles 3 of a kind appears more often. For others 2 pair appears more. For some neither appears. This is how a fixed lighting environment works and why labs and experts have rejected systems like GemEx which are ‘weighted’ towards a particular shuffle. In reality, the world’s lighting environments are made up of an infinite number of shuffles.
So…you begin to look for a better way of knowing how often a certain hand will be dealt in poker. You realize you need a way to analyze the game that is shuffle-independent and will give you the greatest chance of evaluating overall potential. The poker channels do this by calculating the odds for a given results with all possible shuffles considered.
In line with the above, one way to arrive at a scintillation result would be to illuminate the stone in one lighting environment and count the number of sparkles. You could count the white sparkles, colored sparkles, sparkles of various sizes and anything else you can think of related to scintillation. You could then “shuffle” your lighting environment and repeat, over and over again.
AGSL approaches the problem from the other direction. They ask, what is the total number of potential sparkles this stone has when it is tilted along such and such an axis, with all shuffles considered? What is the probability of seeing at least 3 or more sparkles when a x mm light source is randomly placed in the hemisphere? On average how far are these sparkles distributed across the crown of the stone? On average, how large are the sparkles? What are the odds of seeing a sparkle of such and such size? The answers to all of the questions are not calculated from one specific lighting environment, or two, or twenty: They are the results you would get when you consider all possible lighting environments.
Knowing the answers to these questions gives you expectations of what you will likely see in the stone. Considering that the number of the world's lighting environments are closer to infinite than not, decoupling the lighting environment from the metric (as much as possible) seems the way to go. Fortunately, the information now possible by ray tracing the stone allows these sorts of probabilities to be computed.
Q: The AGS animated metric is based on 45 degrees of study with assessments at each degree of tilt. What happens if you move it over one degree, won’t things change?
A: Definitely, the animations and data change when the tilt axis changes. What’s more, remember these are virtual models with perfect optical symmetry. AGSL is just scratching the surface. They are in the process of increasing the speed of the tools and the granularity of the tilt studies. With regard to animations it will be useful to show the tilts along a particular axis or along other patterns like a figure-8, as Sergey has programmed into DC. I think it will be interesting to see what they come up with to show several tilts at once. This is going to be a huge job and data compilation will be massive. Logically there will come a point where the data doesn’t shift much with the addition of additional tilts. A sweet spot between speed and accuracy will need to be decided-on; something the lab has done a good job with before in my opinion.
That’s another important distinction between this and other metrics: Tilt is critical to an evaluation of scintillation. Moving lights inside a box is different from tilting the stone. Stones can differ in several different ways when tilted. One critical difference that has a huge impact on appearance is the size of the virtual facets. The “cracked glass” scintillation that you see on the side of an oval looks a lot different than the large bold virtual facets observed in the bow tie areas. The distribution of small, medium, and large virtual facets across the crown also has a huge impact. For example you don’t see broad flashes of fire in stones that lack broad virtual facets. The AGSL scint maps which partition the crown according to scint event size gives the assessor this kind of critical information and insight.
By contrast, looking at a collection of five images only tells you what the stone looks like with the light in five different positions. You can’t create meaningful potential maps with such a limited lighting condition. Also, the photos give you no information on what happens with tilt: Usually when people are observing scintillation it’s the diamond that is in-motion, not the lights in the room.
It’s important to reiterate that, like the fire potential maps, the scint maps are abstractions and not meant to be realistic depictions of scintillation. They have to be interpreted properly in the overall sense of potential to be understood. Remember, this is not a grading metric yet (not even close) but when one is presented it will be built on non-variable, repeatable research.
LIGHT SOURCE INDEPENDENCE
Here’s an example to further demonstrate the need for light source independence:
Click on this link
to see a test for light return. Check the scores. You may chuckle when you realize there is no difference in the images. In fact, all five are of the same stone. It was done to prove a point: The same image was simply rotated on axis 5 degrees in each progression. Intuitively one would assume the readings would be identical with no deviation (which should be the case), especially since this image was given a face-up analysis only. Considering that, there should have been no deviation. Now project this forward and you realize the use of a changing lighting environment on even a static subject will logically reveal more dramatic deviation; results that are NOT repeatable unless you truncate to 10 (not 00.10 but 10.00). This LR test illustrates the variables possible when image analysis or pixel counting are used. The only way to develop a metric for this is to allow for a range of non-repeatability in the result and confine that non-repeatable result to one (or several) lighting conditions at fixed ranges. Unfortunately our world has a near-infinite number of panoramas of illumination, so the results are both non-repeatable and limited in what they can tell us.
Per the above post, AGSL is working towards quantification based on results that are not variable and are repeatable. Comparing their current studies to existing “light box” technologies is like comparing an analysis of tens of thousands of individual rays (in the neighborhood of 60k) at any orientation, subtending to any angle - to counting colored pixels on a face-up image. We feel the AGSL, Dr. Sasian, Sergey Sivovolenko and Martin Haske are far ahead when it comes to this area of study.
This article is not intended to disparage any system. Researchers in all fields are to be commended because everyone contributes to our knowledge base. Still, technological advancements in many disciplines can make what we are doing today obsolete next week. You never know, and we in the trade are humbled by the efforts of all the researchers out there.