Mark Holdaway asked: Folks, With 4 moves per day made 2 days per week, it takes about 18 weeks to go from COMPACT to Y+. Making an assumption about how long you sit in the Y+ and COMPACT arrays (15% in COMPACT, 15% in Y+), we end up with something like: 7.5 weeks COMPACT (including N-S extension) 18 weeks march OUT 7.5 weeks Y+ config 18 weeks march IN --------------------- 51 weeks Which, at first thought, is great (ie, drink vodka for a week and we've got 52 weeks even), but on second thought, is BAD. We want the arrays to cycle through the time of year. When we MARCH through the intermediate arrays, we'll hit them in SPRING and FALL, or in WINTER and SUMMER. It isn't quite that simple because small intermediate arrays will aloways be close to the compact array and will not be 6 months apart. BUT, we'd like to cycle through the seasons in a different way, so that EVERY configuration gets to see EVERY source with good TAU and PHASE STABILITY sooner or later. Questions: 1) What about the assumptions of 15% of the time in Y+ and in Compact? If these go down to 5%, the whole cycle is like 40 weeks, so we cycle through the seasons roughly every 3 years. I DISAGREE WITH THIS ONE. 2) If we sit in Y+, Compact for LONGER, ie, 22% each, we get something like a 1.25 year configuration cycle, which will cycle through in 5 years (similar to VLA, which is 4 years). I AM GUESSING THIS IS TOO LONG, BUT COULD BE RIGHT. 3) Other options include moving antennas 3 days a week or 3 days every 2 weeks to speed up or slow down the cycling; WHAT DO YOU GUYS THINK? 4) Or do like the VLA and MARCH the antennas one way as fast as you can, skipping stations, to get from say COMPACT to Y+ FAST, and going liesurely the other way. But I believe this goes against the ideas we've gone along with in the past: smooth reconfigurability. I AM AGAINST THIS IDEA, BUT FOR NO GREAT REASONS. Ultimately, observing pressure (perhaps with administrative nudges) will decide exactly what happens here, but it would be good to find a scheme that both matches observing pressure and fits nicely with our desire for cycling around the seasons (ie, is consistently either a fair bit SHORTER or LONGER than 1 year even). Bryan Butler replied: i would guess that we might not want to sit in the full resolution Y+ for a full 2 months. cut that in half, perhaps. just a thought. unfortunately, that doesn't solve the problem, since you only gain 4 weeks per year from that. i would attempt to up the number of moves per week, probably. how about 10 moves per week (3 each on M/W/F, and 1 extra one, whichever day allows it [i.e., if it's OK on monday, do it]). this buys you an extra 25% on the marching. this, along with the decreased time in the full Y+, might save you enough to cycle through sensibly (my rough numbers provide: 7.5 weeks COMPACT; 14.5 weeks march OUT; 3.5 weeks Y+; 14.5 weeks march IN = 40 weeks.). hmmm. as mark points out, this is a 3 year cycle to go through seasons... going to 12 moves per week, which is 4 per day on M/W/F gives you a total move time of 35 weeks, which gives you seasonal cycling in a 2 year timescale - that might be quick enough. how many intermediates are there between the outermost conway config (4 or 5 km max baseline) and the full Y+? we've discussed before skipping those intermediates. how much time is saved if you do that? -bryan Angel suggested: Hi Mark, Another idea: 4 moves/day, 3 times per week (Monday, Thursday, Saturday for example) Also assuming 6 weeks for compact and y+ configurations per cycle, we then get to the following situation: Compact (N-S sub-configuration included) 6 weeks March OUT 12 weeks Y+ Configuration 6 weeks March IN 12 weeks ------------ Total: 36 weeks (9 months). So in principle you will shift 1 season (quarter of the year) per year. In this case Compact and Y+ configurations will shift to the opposite seasons every two years. Sound this right? Cheers, Angel To which Mark said: It occurs to me: if we have a very fast cycle, this has implications on the proposing schedule, which the software people *think* they are in control of, but is probably an executive decision. If we cycle through very quickly, this may imply a fast proposal cycle and turnaround. Also, for a configuration in the middle of the "March", we will return to that configuration every 18 weeks. Given the redundancy in the IN and OUT marching, I would suggest that we either go for a cycle which is a bit longer than 1 year (also, less time spent moving, hopefully more efficient telescope use) OR a burst reconfiguration from COMPACT to Y+ or vise versa. I don't think we've done a cost-benefit analysis of including a burst reconfiguration at one end. I guess the efficiency savings for the slower reconfiguration to get us to a reconfiguration cycle longer than 1 year would be close to negligible (like 1%). And Mark replied: There are now 42 antennas that must be moved from the 4km config to the Y+. I am guessing that we don't want to BOTH cut the time in the Y+ AND speed our way through the intermediates. Perhaps a better question to ask would be: how can we predict proposal pressure? or how can we make a reconfiguration scheme which will be flexible enough to respond to changes in proposal pressure, while maintaining certain properties we are happy with (ie, seaonal cycling). -Mark On 2002.12.11 08:36 Mark Holdaway wrote: > > > > how many intermediates are there between the outermost conway config > > (4 or 5 km max baseline) and the full Y+? we've discussed before > > skipping those intermediates. how much time is saved if you do that? > > There are now 42 antennas that must be moved from the 4km config to > the Y+. I am guessing that we don't want to BOTH cut the time in the > Y+ AND speed our way through the intermediates. i agree. one or the other. but you might get more time savings one way vs. the other... > Perhaps a better question to ask would be: > how can we predict proposal pressure? i submit that we can't. we've always (well, at least consciously since the grenoble meeting) been carrying around the idea of having the most flexible system possible because we *know* that it will change over time. now, that doesn't excuse us from the responsibility of coming up with the best scheme that we can to begin with, but we know that it will change, and almost certainly in unpredictable ways. > > or > > how can we make a reconfiguration scheme which will > be flexible enough to respond to changes in proposal pressure, > while maintaining certain properties we are happy with > (ie, seaonal cycling). yep - this is the key. but, i don't know that we have to make a scheme to begin with that is necessarily flexible intrinsically. the flexibility is built into the pad layout, in many ways, which was the big reason (in my mind, at least) why the 'zoom-spirals' won out over the nested donuts... -bryan On Wed, 11 Dec 2002, John Conway wrote (>), and Mark replied: > > Hi, > > I guess I was thinking of having a whole cycle > longer than a year and less time in the extreme > arrays giving a rate of one move day every four, > but Mark is right one probably wants quite a > bit of time in the extreme arrays. > > Of the suggestions given I like Angels the best > Doing an average rate of 12 moves per week is certainly > feasible given the transporter capacity though it > means moving almost every second day which seems > just a litle high for sustained operations. > I think there is an argument for going through > the arrays between the largest spiral and the extreme > Y+ faster than the rest. I agree that there are such arguments, but will also point out that these arguments have already been factored into the array design, as the resolution will be increasing with reconfiguration of 4 antennas by a multiplicative factor more than twice as large as the resolution increase factor for arrays smaller than 4 km. I can accept this suggestion at this time, with the provision that this is all flexible in responce to proposal pressure. > If in the move out from most compact to largest spiral > we make moves on 1 day out of every 3, then with 2moves/day/transporter > and 2 transporters there are an average fo 1.33 > antenna moves/day. It take 108 moves hence 81 days (11.5 > weeks) to go from compact to largest spiral. > Moving from largest spiral to largest Y+ if there are 42 pads > and we make moves on 2 days out of every 3 it > takes 16 days (2.3 weeks). > > Assuming 6 weeks stopped in the extreme arrays the cycle time > becomes > > Compact 42 days > Compact -max spiral 81 days > Max spiral - max Y+ 16 days > max Y+ 42 days > Max Y+ - max spiral 16 days > max spiral -compact 81 days > ---- > 278 days (39.7 weeks) I will take this is my straw-plan for the moment. > Note I make that 9 months equals 39 weeks > (not 36 as stated below) > > 1) It might be thought that the time to > 'cycle the seasons' for the compact array > is 4 years in this scheme, but remember > you have a compact array every 9 months > ( incidently a a lot better than the VLA at one > every 16 months), so I think the time between when > the starting date of say the compact array > starts on 1 Jan and when it starts again on > 1 Jan is 3 years not 4, see the schedule below. If the cycle is (1-1/n) years, it makes n cycles in n-1 years. If the cycle is (1+1/n) years, it makes n cycles in n+1 years. > 3) I'm not sure a rational mumber for the > ratio of cycle time to year is needed, perhaps > a cycle time of 40 or 41 weeks is as good > or better that 39 days (9 Months). Yes, as long as the seasons change. > 4) Are there times of year we definitely > DON'T want to be in the compact or extreme > arrays (the Bolivian winter pehaps) we can schedule > one of the intermediate arrays for this (since they > occur every 4.5 months anyways its not too much loss). > If we work such factors in this would argue for > a rational fraction between cycle time and 1 year > and for a particular phasing of that cycle. My opinion: all configurations need to share in the Bolivian winter, just as all configurations should share in the good times (at this point, you should picture an array of antennas drinking vodka and dancing -- don't ask me, I dislike vodka myself). Once I get a scheme for the reconfiguration cycle that we are all OK with (a hope), I will manufacture a complete dataset of TAU and RMS_PHASE (by filling in the gaps in the site testing data by hook or by crook), turn the crank to reconfigure for about 3 years, and then look at the distributions in TAU and PHASE for each configuration and each LST range, basically to verify that EACH source and EACH configuration is getting a fair shake. Meanwhile, i need to stop doing this and get back to other work! A principle which can guide flexibility: to let the time spent in each configuration be proportional to observing pressure (or high quality observing pressure). We have two knobs to turn: - speed of reconfiguration - amount of time spent on end points We start with OUR reconfiguration scheme (whatever that is), and then adjust the two knobs in responce to proposal pressure, perhaps in a week-by-week decision process (ie, if there is a large backlog of projects, the site director could choose to slow down reconfiguration; if there are gaps in ALMA's use because there are no projects at that LST and that resolution and those prevailing weather conditions, it is time to move antennas faster). If this flexible approach is chosen, then we have no reproducable cycling through seasons, it just comes out the way it comes out.... and we may move faster through summer than winter. -Mark And John chimed in: Hi, There is a question of how much flexibility in response to proposal pressure one wants. I myself was never a fan of changing the scheduling scheme on a week to week basis in response to proposal pressure. In my memo 283 I present what I had in mind. I considered that there would be differnt types of expt, some requiring an exact resolution others with looser requiements the more exacting ones would be scheduled first. In this way in a zoom array you could get the option of exactly the same resolution at different line transitions (which you didn't have with fixed arrays 2 or 3 apart in resolution). The felxibility in sheduling was based on -WHEN- a experiment was scheduled in the move schedled, NOT by changing the move schedule itself. I was thinking that one MIGHT change the overall cycle scheme every cycle or two in response to the statistics of proposals pressure, Thus if after 9 months or 18months it seemed that the statitsics showed a certain array size was less popular then for the coming 9 month schedule one would schedule there to be 2 moves days our of every 3 through that array size rather than 1 out of every 3 - and then move slower through the popular arrays, while maintaining the overall cycle time - something like that. Even in this scheme I did not envision allowing the whole configuration cycle time to be ajustable in reponse to proposal pressure. I think it would be good to consider the overall cycling with repect to seasons, try to optimise it and then keep it fixed no matter how much internal flexibilty is introduced within the cycle. Of course if people think it would be desirable to have more interaction between proposal pressure and the reconfiguration cycle- then I guess it can be built in but it gets complicated fast. John To which Bryan replied: i agree with john here. adapting on that short a time scale to proposal schedule is not a good idea, IMHO. in my mind, the way to do it is come up with a decent scheme (the best we can given current thinking), and test it out for a long period (at least once through the complete cycle). if it turns out to work, fine, we're done for a bit. if it doesn't work, then we have to fix it, which may just be tweaking, or may be a major overhaul. for instance, we may find that the astronomers don't like a continual reconfiguration, and squawk about it enough, and we have to go back to burst reconfiguration... the fixing comes via input from operations and the user community. but, again, i think it would be a bad idea to start with the idea of immediate response to proposal pressure. that's something that will come over time... -bryan And Mark opined: It is not just proposal pressure, but also unlucky weather. Lets say you get two good storms in a row, basically taking out the good tau conditions for 2 weeks. After that, you are 24 moves in resolution down the road and have not been able to observe a whole host of proposals. OK, we wait for the Marching in the other direction. What if it happens again? There will be some irate non-observers. I agree with the "look back at the past years and make adjustments" approach, but with dynamic scheduling and incremental reconfiguration, it is easy for me to imagine getting a backlog, and it is also easy for me to imagine the array going underutilized, and I would be happier if the reconfiguration could respond to such pressures. One could build in such flexibility while still meeting a fixed overall cycle time by squeezing the entire cycle, not necesarily letting it slip. Anyway, this is likely an executive decision. -Mark Bryan agreed: agreed. it's for the ASAC + JAO + IPT leads, i guess... Al said: Interesting discussions. I think the best thing to do now is to settle on a reasonable plan--the one Mark proposed taking as a straw-plan seems good to me--and working out the ramifications. Personally, I don't think we should worry about the Bolivian Winter right now, except that I'd probably avoid positioning Y+ during it. I've always thought that it would be best to schedule more compact configurations and lower frequencies then, but I wouldn't make that a driver for now. I think that means avoiding 1 Jan - 15 Feb in Y+ configuration. I think that we need to begin at least with a fairly inflexible reconfiguration plan and feel our way into a plan with parameters as variable as making a decision a few weeks ahead of schedule. The dynamic scheduler should of course be built with that notion considered and operations plans should as well. Mark noted BTW: I was speaking with Fred Lo yesterday, and he strongly favors a reconfiguration cycle which is shorter than a year. -Mark Al pointed out: I think we all agreed with a cycle of less than a year, so if Fred makes that decision it will be because he had good advice. and John agreed: (1) Yes I think with a cycle time of 9 months you get quite good properties, the time to cycle the seasons is 3 years, so 18 months to go to being the opposite season - but since the compact array and most extended arrays are of order 6 weeks long, the time between the end of one compact array and the start of another is 7.5months (similiary for the most extended array), so you get an approximate shift from day to night observations for your favorite source on the shortest possible timesclale. (2) A year can be divided into 8 periods of 6.5 weeks With an exact 9 month schedule, the compact array only every occurs in 4 out of these 8 calendar periods. (similary for the most extended array) This can be used to advantage if there were periods of order 6.5 weeks long one wanted to avoid for particular arrays. If on the other the object is to give every configuration a 'fair shake' and sample all parts of the year equally then you would like to advance the cycle by about 6.5 weeks over 3 years, so a cycle period of around 41 weeks would accomplish this. This would ease transporter load but it would then be longer between end of a compact array and the start of one (35 weeks) but still give some day/night cycling within a single year. John Mark characterized the Bolivian Winter: Taking all data from 1996 to 2002, TAU: The Bolivian Winter is very obvious in TAU, with median opacities (median of all measurements on a given day, over all hours and years of that day) jumping a factor of 2-5 over the regular good times. There WILL be some good conditions mixed in (ie, the Bolivian Winter "breathes" in and out, and the details of how it breaths will vary from year to year). Opacity shows essentially no diurnal effect during the good conditions, but shows a significant diurnal effect during the southern summer, so there will be some good TAU times, especially at night. It starts suddenly around Day 350 (Dec 15) and slowly comes back to normal by day 100 (April 10). This is like 115 days, or 16 weeks. The WORST of the Bolvian Winter is Days 355 - 80 (95 days, or 14 weeks). And THE ABSOLUTE WORST of the Bolivian Winter is Days 360 - 60 (65 days or 9 weeks). I won't tell you when the TRULY MOST HORRIBLE TERRIBLE part of the Bolivian Winter is. PHASE: The Bolivian Winter is much less obvious in phase, amounting to roughly a factor of 2 larger "median phase per day". It begins around day 350 and ends around day 100. It seems to end less gradually than the TAU. DEMANDS of Y+ Y+ will be used at all frequencies. There may be a bias towards higher frequencies to get the highest possible resolution, or towards low frequencies so that the resolution isn't so great that it resolves out everything. I don't know. Phase stability will be an issue, but fast switching theory basically predicts that the residual phase will be the same as if it were a 500m array. One effect that goes against this is that at such high resolution (ie, 50 mas at 90 GHz), some of the calibrator sources may stop looking like point sources, or more importantly, may lose some flux and not be as efficient in calibrating. However, as the jet components (ie, > 50 mas) are steep spectrum and the compact core is flat spectrum, I expect this to be a minor effect. SO MY GUESS: there is no significant reason that the Y+ array will be more adversely affected by the Bolivian Winter than, say, the Marching Through Intermediate Arrays will be. DEMANDS of COMPACT ARRAY As the resolution increases with frequency, we will be pushed to smaller arrays (ie, THE COMPACT ARRAY) for a lot of the high frequency work. As the opacity goes to hell at high frequencies even in the best of conditions, I think there may be a good argument for trying to avoid the Bolvian Winter for the COMPACT ARRAY. Phase stability: fast switching is not so effective for the compact array because the crossing time is comparable to the switching time. You won't wreck the observations by doing fast switching, and it won't be worse than longer baseline observations; its more like you are stuck with what the atmosphere provides you. I don't think the phase stability pushes us one way or the other. The wind speed is pretty much anti-correlated with the Bolvian Winter. BIG WINDS begin gradually around day 100 and come down around day 330. AS POINTING IS ALSO IMPORTANT FOR THE COMPACT CONFIGURATION (for both mosaicing and for high frequency work), THIS seems to give some merit to having some COMPACT array time during the Bolivian Winter. NOTE that this analysis DOES NOT look at diurnal effects, and there are always good calm (low wind) conditions at night, even in the windiest winter times. GIVEN THE FAIRLY BROAD NATURE OF THE BOLIVIAN WINTER I don't think we can avoid putting any confifuration ever in the Bolvian Winter. If any configuration is justified in being spared, I think it is the COMPACT CONFIG. (Al should make his arguments for sparing the Y+ configuration.) We can probably arrange for the COMPACT CONFIG to fall on the part of the Bolivian Winter that is usually not the worst. Take care, -Mark And Min suggested a summary: Hi Gang, Most of these issues were discussed at some level before, and it is interesting which idea has survived the test of time. My recollection as an early participant in this work and as an ASAC observer is that the reconfiguration cycle time discussed in the past (at least when the configuration style and reconfiguration modes were being decided) were shorter. There were discussions of going to Y+ configuration only every other times or so, because of the large overheads in the reconfiguration, etc. I don't think the whole ASAC would care to look at all these e-mail exchanges, but I suspect many members of ASAC have an opinion on this. If someone can put together a straw reconfiguration plan and a summary of the relevant discussions (e.g. Bolivian winter), I would like to forward it to the ASAC and hear some of their voices. What do you think, Al? -- Min