![\"\"](\"https://i.imgur.com/l0DNOPe.png\")
Hey all,
\nSo I was looking a bit at the distribution of speeds between two detectors that activate during an event. I did that for all the detectors for ~1000 events.
\n
You can see that there is a cluster of speeds around 0.3 m/ns. This looks like the speed of the events we are trying to detect.
\nI had a look at some events for which the detectors with auxiliary = False match well the direction of the neutrino given. In those cases, we can in fact see a linear correlation between time and distance highlighting this speed around 0.3 m/ns
\n![\"\"](\"https://i.imgur.com/lP58Yz4.png\")
This is currently a piece of data I am trying to leverage to automatically discriminate non-relevant detectors.
", "messages": [ { "id": 2120580, "postDate": "2023-01-29T17:31:03.393Z", "content": "Hey all,
\nSo I was looking a bit at the distribution of speeds between two detectors that activate during an event. I did that for all the detectors for ~1000 events.
\n
You can see that there is a cluster of speeds around 0.3 m/ns. This looks like the speed of the events we are trying to detect.
\nI had a look at some events for which the detectors with auxiliary = False match well the direction of the neutrino given. In those cases, we can in fact see a linear correlation between time and distance highlighting this speed around 0.3 m/ns
\n
This is currently a piece of data I am trying to leverage to automatically discriminate non-relevant detectors.
", "rawMarkdown": "Hey all, \n\nSo I was looking a bit at the distribution of speeds between two detectors that activate during an event. I did that for all the detectors for ~1000 events.\n\n\n\nYou can see that there is a cluster of speeds around 0.3 m/ns. This looks like the speed of the events we are trying to detect. \n\nI had a look at some events for which the detectors with auxiliary = False match well the direction of the neutrino given. In those cases, we can in fact see a linear correlation between time and distance highlighting this speed around 0.3 m/ns\n\n\n\nThis is currently a piece of data I am trying to leverage to automatically discriminate non-relevant detectors.", "votes": 30 }, { "id": 2121628, "postDate": "2023-01-30T12:57:15.207Z", "content": "Could you please explain how do you calculate this speed?
\n0.3 m/ns is actually the speed of light, so the tail of the distribution in your plot seem to indicate speeds above c.
", "rawMarkdown": "Could you please explain how do you calculate this speed?\n\n0.3 m/ns is actually the speed of light, so the tail of the distribution in your plot seem to indicate speeds above c.", "votes": 2, "replies": [ { "id": 2121643, "postDate": "2023-01-30T13:12:22.843Z", "content": "*speed of light in a vacuum. Speed of light in ice is ~0.23m/ns
", "rawMarkdown": "*speed of light in a vacuum. Speed of light in ice is ~0.23m/ns", "votes": 3, "replies": [ { "id": 2121707, "postDate": "2023-01-30T13:57:42.503Z", "content": "Interesting, I would be curious to know why I find something slightly faster with my exploration. Any idea?
", "rawMarkdown": "Interesting, I would be curious to know why I find something slightly faster with my exploration. Any idea?" }, { "id": 2121855, "postDate": "2023-01-30T15:38:10.610Z", "content": "Appears to depends a lot on depth of ice, and a couple experiments had it approaching refraction index of 1.78 for deeper ice. But the source I was reading had a third reference that showed a peak and then back to much faster speed in deeper ice. So who knows.
\nBut if the 1.78 is somewhat close then 0.2998 / 1.78 ~= 0.168m/ns is about the speed of light in 1.5 to 2.5 km deep ice.
\nThe Cherenkov radiation effect gives a conic travel time between that speed and the higher speed of the particle.
\nAnd then there's a host of other factors in terms of light traveling from a more distant point not the nearest, so arrives later; or arriving at two points on the cone or sphere at nearly the same time (perceived speed much faster than light). Etc.
\nStill very interesting that there's a small bump at that 0.3 speed.
\nOn my phone so can't easily grab it, but one paper that was trying to reconstruct energy released showed arrival time as a probabilistic graph varying with a long tail. And that's from events where they were confident of the center. Using just time and distance between observations, without a known origin, obviously adds much more variability.
", "rawMarkdown": "Appears to depends a lot on depth of ice, and a couple experiments had it approaching refraction index of 1.78 for deeper ice. But the source I was reading had a third reference that showed a peak and then back to much faster speed in deeper ice. So who knows. \n\nBut if the 1.78 is somewhat close then 0.2998 / 1.78 ~= 0.168m/ns is about the speed of light in 1.5 to 2.5 km deep ice. \n\nThe Cherenkov radiation effect gives a conic travel time *between* that speed and the higher speed of the particle. \n\nAnd then there's a host of other factors in terms of light traveling from a more distant point not the nearest, so arrives later; or arriving at two points on the cone or sphere at nearly the same time (perceived speed much faster than light). Etc. \n\nStill very interesting that there's a small bump at that 0.3 speed. \n\nOn my phone so can't easily grab it, but one paper that was trying to reconstruct energy released showed arrival time as a probabilistic graph varying with a long tail. And that's from events where they were confident of the center. Using just time and distance between observations, without a known origin, obviously adds much more variability. ", "votes": 1 }, { "id": 2121884, "postDate": "2023-01-30T16:01:36.553Z", "content": "I think speed of neutrino is almost 0.3m/ns even in ice (and it emits Cherenkov light).
", "rawMarkdown": "I think speed of neutrino is almost 0.3m/ns even in ice (and it emits Cherenkov light).", "votes": 1 }, { "id": 2121914, "postDate": "2023-01-30T16:15:38.927Z", "content": "Agree, but the travel time from closest point of the neutrino's path to the sensor will be a specific calculated value based on (and in between) both speeds (particle speed and speed of light in ice)
", "rawMarkdown": "Agree, but the travel time from closest point of the neutrino's path to the sensor will be a specific calculated value based on (and in between) both speeds (particle speed and speed of light in ice)" }, { "id": 2121984, "postDate": "2023-01-30T16:50:51.907Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2121990, "postDate": "2023-01-30T16:52:10.027Z", "content": "Here's a situation where this could happen, though keep in mind it is potentially too simplified.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13460982%2Ff82ebe8d663d932e0f41e731d14a13ed%2FScreenshot%20from%202023-01-30%2008-07-21.png?generation=1675094860326415&alt=media\")
Neutrino's move at very marginally below the speed of light in a vacuum, even when they are in a medium with a high index of refraction. This neutrino interacts with the ice and produces a charged particle from the interaction. Since all detections occur via Cherenkov radiation, we know the charged particle must be moving faster than the speed of light in ice (c/n, where n is the index of refraction in the ice at the south pole) in order to be detected / triggered. The charged particle then continuously emits Cherenkov radiation, which will be brightest along a cone oriented in the direction of travel of the charged particle.
\nIf the path of the charged particle is parallel to the line connecting the 2 DOMs, then the 'speed' between each DOM measurement will be related to the speed of the charged particle, since the photon travelling from the particle to DOM 1 and the analogous photon travelling from the particle to DOM 2 will both take the same amount of travel time.
\nCases where the particle is not travelling parallel to the line connecting the DOMs are a bit harder to conceptualize, since the difference in travel times of photons would also need to be incorporated. It could be that, for the radiation to be energetic enough to trigger a DOM, the distance from particle to DOM must be small relative to the distance between DOMs, so that differences in photon travel times can largely be neglected. But I haven't looked into the data to test this last theory.
", "rawMarkdown": "Here's a situation where this could happen, though keep in mind it is potentially too simplified.\n\n\n\nNeutrino's move at very marginally below the speed of light in a vacuum, even when they are in a medium with a high index of refraction. This neutrino interacts with the ice and produces a charged particle from the interaction. Since all detections occur via Cherenkov radiation, we know the charged particle must be moving faster than the speed of light in ice (c/n, where n is the index of refraction in the ice at the south pole) in order to be detected / triggered. The charged particle then continuously emits Cherenkov radiation, which will be brightest along a cone oriented in the direction of travel of the charged particle.\n\nIf the path of the charged particle is parallel to the line connecting the 2 DOMs, then the 'speed' between each DOM measurement will be related to the speed of the charged particle, since the photon travelling from the particle to DOM 1 and the analogous photon travelling from the particle to DOM 2 will both take the same amount of travel time.\n\nCases where the particle is not travelling parallel to the line connecting the DOMs are a bit harder to conceptualize, since the difference in travel times of photons would also need to be incorporated. It could be that, for the radiation to be energetic enough to trigger a DOM, the distance from particle to DOM must be small relative to the distance between DOMs, so that differences in photon travel times can largely be neglected. But I haven't looked into the data to test this last theory.", "votes": 2 }, { "id": 2122230, "postDate": "2023-01-30T19:17:32.333Z", "content": "For the parallel case, the interesting point I was trying to make is that the charge speed (labeled as Photon above) is faster than v = c/n, at least to the nearest (orthogonal) point of the origin line. Agree with the rest of that great picture, thanks.
\nIf you assume charged particle is traveling at c, and a known ice refraction of n of say 1.8, then that 'photon' speed is a calculable value based on red line I added to below picture, though I haven't taken the time to do the math yet.
\n![\"cherenkov\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6648125%2F628ede4064c6cf84559794b2a20cb378%2FCherenkov.svg.png?generation=1675105726577650&alt=media\")
My understanding and the picture come from wikipedia.
\nThe apparent speed relative to the nearest point of origin could potentially be helpful if doing any kind of estimation of origin point for a subset of points, and trying to see if the origin point 'answer' fits with the expected speed value.
", "rawMarkdown": "For the parallel case, the interesting point I was trying to make is that the charge speed (labeled as Photon above) is faster than v = c/n, at least to the nearest (orthogonal) point of the origin line. Agree with the rest of that great picture, thanks.\n\nIf you assume charged particle is traveling at c, and a known ice refraction of n of say 1.8, then that 'photon' speed is a calculable value based on red line I added to below picture, though I haven't taken the time to do the math yet.\n\n\n\nMy understanding and the picture come [from wikipedia](https://en.wikipedia.org/wiki/Cherenkov_radiation).\n\nThe apparent speed relative to the nearest point of origin could potentially be helpful if doing any kind of estimation of origin point for a subset of points, and trying to see if the origin point 'answer' fits with the expected speed value.", "votes": 1 }, { "id": 2125355, "postDate": "2023-02-01T15:55:21.327Z", "content": "I might misunderstand your point but i think that the neutrinos might continuously (roughly speaking) emits particle that can be detected. Therefore in the ideal case you will have several detector separated by a time/space intervall coherent with neutrino's speed in ice which might be what we see on the distribution.
\nI might also misunderstand the physics behind it ahah.
I am using this formula to calculate the \"activation speed\" between two sensors:
\n ():\n\n time1,_,_, x1,y1,z1 = sensor1\n time2,_,_, x2,y2,z2 = sensor2\n\n dist = np.sqrt((x2-x1)**+(y2-y1)**+(z2-z1)**)\n dist/(time2-time1))\nThe idea is that in the case of activations made purely randomly, we expect to see the distribution in red in the figure in my last message. But I also expect the \"activation speed\" to highlight a cluster of \"sensor pairs\" all in the same range of \"activation speed\", which could be directly connected to the neutrino passage.
\nFrom the above figure, we see there is a second distribution \"mixed\" with the random one centered around 0.25-0.26 m / ns, which is what I was hoping to see to confirm my \"theory\". If the \"activation speed\" between two sensors is too low or too high, it is probably due to at least one of the sensors being activated purely randomly, but if it falls in the range around 0.25-0.26 m/ns, it can also be due to the path of the neutrino traveling inside the IceCube.
\nNow I find this interesting because it can give another way of filtering noise out of the activated sensors. We can, for example, build a connexion graph with a link between two sensors for which we see an \"activation speed\" in the good range, and check the biggest cluster of sensors (which are supposedly all activated by the same particle). I ran some experiments around this and even if I don't manage to consistently beat the sensors filtered in the dataset, I get sometime much better results by using the detectors highlighted with my method.
", "rawMarkdown": "I am using this formula to calculate the \"activation speed\" between two sensors:\n\n```python\ndef get_speed(sensor1, sensor2):\n \n time1,_,_, x1,y1,z1 = sensor1\n time2,_,_, x2,y2,z2 = sensor2\n \n dist = np.sqrt((x2-x1)**2+(y2-y1)**2+(z2-z1)**2)\n return dist/abs(time2-time1))\n```\n\nThe idea is that in the case of activations made purely randomly, we expect to see the distribution in red in the figure in my last message. But I also expect the \"activation speed\" to highlight a cluster of \"sensor pairs\" all in the same range of \"activation speed\", which could be directly connected to the neutrino passage.\n\nFrom the above figure, we see there is a second distribution \"mixed\" with the random one centered around 0.25-0.26 m / ns, which is what I was hoping to see to confirm my \"theory\". If the \"activation speed\" between two sensors is too low or too high, it is probably due to at least one of the sensors being activated purely randomly, but if it falls in the range around 0.25-0.26 m/ns, it can also be due to the path of the neutrino traveling inside the IceCube.\n\nNow I find this interesting because it can give another way of filtering noise out of the activated sensors. We can, for example, build a connexion graph with a link between two sensors for which we see an \"activation speed\" in the good range, and check the biggest cluster of sensors (which are supposedly all activated by the same particle). I ran some experiments around this and even if I don't manage to consistently beat the sensors filtered in the dataset, I get sometime much better results by using the detectors highlighted with my method. \n\n\n\n", "votes": 4 } ] } ], "comments": [ { "id": 2121628, "author_name": "Alex Z", "author_url": "", "post_date": "2023-01-30T12:57:15.207000", "content": "Could you please explain how do you calculate this speed?
\n0.3 m/ns is actually the speed of light, so the tail of the distribution in your plot seem to indicate speeds above c.
", "votes": 2, "replies": [ { "id": 2121643, "author_name": "datasaurus", "author_url": "", "post_date": "2023-01-30T13:12:22.843000", "content": "*speed of light in a vacuum. Speed of light in ice is ~0.23m/ns
", "votes": 3, "replies": [ { "id": 2121707, "author_name": "Jacky", "author_url": "", "post_date": "2023-01-30T13:57:42.503000", "content": "Interesting, I would be curious to know why I find something slightly faster with my exploration. Any idea?
", "votes": 0, "replies": [] }, { "id": 2121855, "author_name": "Robert Hatch", "author_url": "", "post_date": "2023-01-30T15:38:10.610000", "content": "Appears to depends a lot on depth of ice, and a couple experiments had it approaching refraction index of 1.78 for deeper ice. But the source I was reading had a third reference that showed a peak and then back to much faster speed in deeper ice. So who knows.
\nBut if the 1.78 is somewhat close then 0.2998 / 1.78 ~= 0.168m/ns is about the speed of light in 1.5 to 2.5 km deep ice.
\nThe Cherenkov radiation effect gives a conic travel time between that speed and the higher speed of the particle.
\nAnd then there's a host of other factors in terms of light traveling from a more distant point not the nearest, so arrives later; or arriving at two points on the cone or sphere at nearly the same time (perceived speed much faster than light). Etc.
\nStill very interesting that there's a small bump at that 0.3 speed.
\nOn my phone so can't easily grab it, but one paper that was trying to reconstruct energy released showed arrival time as a probabilistic graph varying with a long tail. And that's from events where they were confident of the center. Using just time and distance between observations, without a known origin, obviously adds much more variability.
", "votes": 1, "replies": [] }, { "id": 2121884, "author_name": "Yosshi999", "author_url": "", "post_date": "2023-01-30T16:01:36.553000", "content": "I think speed of neutrino is almost 0.3m/ns even in ice (and it emits Cherenkov light).
", "votes": 1, "replies": [] }, { "id": 2121914, "author_name": "Robert Hatch", "author_url": "", "post_date": "2023-01-30T16:15:38.927000", "content": "Agree, but the travel time from closest point of the neutrino's path to the sensor will be a specific calculated value based on (and in between) both speeds (particle speed and speed of light in ice)
", "votes": 0, "replies": [] }, { "id": 2121984, "author_name": "", "author_url": "", "post_date": "2023-01-30T16:50:51.907000", "content": "", "votes": 0, "replies": [] }, { "id": 2121990, "author_name": "Jacob HDS", "author_url": "", "post_date": "2023-01-30T16:52:10.027000", "content": "Here's a situation where this could happen, though keep in mind it is potentially too simplified.
\n
Neutrino's move at very marginally below the speed of light in a vacuum, even when they are in a medium with a high index of refraction. This neutrino interacts with the ice and produces a charged particle from the interaction. Since all detections occur via Cherenkov radiation, we know the charged particle must be moving faster than the speed of light in ice (c/n, where n is the index of refraction in the ice at the south pole) in order to be detected / triggered. The charged particle then continuously emits Cherenkov radiation, which will be brightest along a cone oriented in the direction of travel of the charged particle.
\nIf the path of the charged particle is parallel to the line connecting the 2 DOMs, then the 'speed' between each DOM measurement will be related to the speed of the charged particle, since the photon travelling from the particle to DOM 1 and the analogous photon travelling from the particle to DOM 2 will both take the same amount of travel time.
\nCases where the particle is not travelling parallel to the line connecting the DOMs are a bit harder to conceptualize, since the difference in travel times of photons would also need to be incorporated. It could be that, for the radiation to be energetic enough to trigger a DOM, the distance from particle to DOM must be small relative to the distance between DOMs, so that differences in photon travel times can largely be neglected. But I haven't looked into the data to test this last theory.
", "votes": 2, "replies": [] }, { "id": 2122230, "author_name": "Robert Hatch", "author_url": "", "post_date": "2023-01-30T19:17:32.333000", "content": "For the parallel case, the interesting point I was trying to make is that the charge speed (labeled as Photon above) is faster than v = c/n, at least to the nearest (orthogonal) point of the origin line. Agree with the rest of that great picture, thanks.
\nIf you assume charged particle is traveling at c, and a known ice refraction of n of say 1.8, then that 'photon' speed is a calculable value based on red line I added to below picture, though I haven't taken the time to do the math yet.
\n
My understanding and the picture come from wikipedia.
\nThe apparent speed relative to the nearest point of origin could potentially be helpful if doing any kind of estimation of origin point for a subset of points, and trying to see if the origin point 'answer' fits with the expected speed value.
", "votes": 1, "replies": [] }, { "id": 2125355, "author_name": "Aleb", "author_url": "", "post_date": "2023-02-01T15:55:21.327000", "content": "I might misunderstand your point but i think that the neutrinos might continuously (roughly speaking) emits particle that can be detected. Therefore in the ideal case you will have several detector separated by a time/space intervall coherent with neutrino's speed in ice which might be what we see on the distribution.
\nI might also misunderstand the physics behind it ahah.
I am using this formula to calculate the \"activation speed\" between two sensors:
\n ():\n\n time1,_,_, x1,y1,z1 = sensor1\n time2,_,_, x2,y2,z2 = sensor2\n\n dist = np.sqrt((x2-x1)**+(y2-y1)**+(z2-z1)**)\n dist/(time2-time1))\nThe idea is that in the case of activations made purely randomly, we expect to see the distribution in red in the figure in my last message. But I also expect the \"activation speed\" to highlight a cluster of \"sensor pairs\" all in the same range of \"activation speed\", which could be directly connected to the neutrino passage.
\nFrom the above figure, we see there is a second distribution \"mixed\" with the random one centered around 0.25-0.26 m / ns, which is what I was hoping to see to confirm my \"theory\". If the \"activation speed\" between two sensors is too low or too high, it is probably due to at least one of the sensors being activated purely randomly, but if it falls in the range around 0.25-0.26 m/ns, it can also be due to the path of the neutrino traveling inside the IceCube.
\nNow I find this interesting because it can give another way of filtering noise out of the activated sensors. We can, for example, build a connexion graph with a link between two sensors for which we see an \"activation speed\" in the good range, and check the biggest cluster of sensors (which are supposedly all activated by the same particle). I ran some experiments around this and even if I don't manage to consistently beat the sensors filtered in the dataset, I get sometime much better results by using the detectors highlighted with my method.
", "votes": 4, "replies": [] } ] } ], "raw_markdown_by_id": { "2120580": "Hey all, \n\nSo I was looking a bit at the distribution of speeds between two detectors that activate during an event. I did that for all the detectors for ~1000 events.\n\n\n\nYou can see that there is a cluster of speeds around 0.3 m/ns. This looks like the speed of the events we are trying to detect. \n\nI had a look at some events for which the detectors with auxiliary = False match well the direction of the neutrino given. In those cases, we can in fact see a linear correlation between time and distance highlighting this speed around 0.3 m/ns\n\n\n\nThis is currently a piece of data I am trying to leverage to automatically discriminate non-relevant detectors.", "2121628": "Could you please explain how do you calculate this speed?\n\n0.3 m/ns is actually the speed of light, so the tail of the distribution in your plot seem to indicate speeds above c." } }