Optimization of Integrated Luminosity of the Fermilab Tevatron Collider
aa r X i v : . [ phy s i c s . acc - ph ] O c t Proceedings of the DPF-2009 Conference, Detroit, MI, July 27-31, 2009 Optimization of Integrated Luminosity of the Fermilab Tevatron Collider
M.E. Convery
Fermi National Accelerator Laboratory, Batavia, IL 60510, USA
We present the strategy which has been used recently to optimize integrated luminosity at theFermilab Tevatron proton-antiproton collider. We use a relatively simple model where we keep theproton intensity fixed, use parameters from fits to the luminosity decay of recent stores as a func-tion of initial antiproton intensity (stash size), and vary the stash size to optimize the integratedluminosity per week. The model assumes a fixed rate of antiproton production, that a store isterminated as soon as the target stash size for the next store is reached, and that the only downtimeis due to store turn-around time. An optimal range of stash sizes is predicted. Since the start ofTevatron operations based on this procedure, we have seen an improvement of approximately 35%in integrated luminosity. Other recent operational improvements have been achieved by decreas-ing the shot-setup time and by reducing beam-beam effects by making the proton and antiprotonbrightnesses more compatible, for example by scraping protons to smaller emittances.
I. INTRODUCTION
The Fermilab accelerator complex (Fig. 1) providesbeam to two collider experiments, CDF and D0 at theTevatron, two neutrino experiments, MiniBooNE andNuMI, and 120 GeV fixed-target experiments.
FIG. 1: The Fermilab accelerator complex.
The proton source consists of a Cockroft-Waltonwhich accelerates H − ions to 750 keV, Linac whichaccelerates the ions to 400 MeV, and the Boosterring. Multiple turns worth of beam are loaded intothe Booster, with the electrons being stripped off theions on the first turn, and then the accumulated pro-ton beam being accelerated to 8 GeV. MiniBooNEreceives 8 GeV protons on target from the Booster.The protons otherwise continue to the Main Injector,where they can be accelerated to 120 GeV and sent tothe Pbar Target for antiproton production, to NuMI,or through the Transfer Hall in the Tevatron ring tothe Switchyard fixed-target area. Protons for the col-lider are accelerated to 150 GeV in the Main Injectorand then transferred to the Tevatron, where, once an- tiprotons are also loaded, they are accelerated to 980GeV.Antiprotons are produced by sending 120 GeV pro-tons into a target. The particles produced are fo-cused using a lithium lens, separated by charge us-ing a bend magnet, and then negatively-charged par-ticles are sent down a transport line, at the end ofwhich only antiprotons remain, to the antiproton De-buncher and Accumulator. Antiprotons are stored ona short timescale in the Accumulator, where they arecooled stochastically, and then are transferred to theRecycler storage ring in the same tunnel as the Maininjector, where they are stored until enough antipro-tons have been produced for a new store of protonsand antiprotons in the Tevatron. One distinct advan-tage of the Recycler is electron cooling, where a beamof electrons passes along side of the antiproton beamin a portion of the ring. Coulomb scattering betweenthe antiprotons and electrons brings the antiprotonsinto thermal equilibrium with the electrons, and bycontinually refreshing the electron beam, brings theantiproton momenta into a tighter range around thedesired momentum. Besides this advantage for storingantiprotons in the Recycler, antiprotons are also pro-duced much more effectively when only a small num-ber is present in the Accumulator. Antiprotons areinjected into the Tevatron by transferring the 8 GeVantiprotons from the Recycler to the Main Injector,accelerating them to 150 GeV, and then transferringthem to the Tevatron, 4 bunches at a time.Antiprotons in the Accumulator are referred to asthe stack , and stacking is the term we use for pro-ducing them. A pbar (¯ p ) transfer is the process ofmoving antiprotons from the Accumulator to the Re-cycler, and once they are in the Recycler, they arereferred to as the stash . A store is a colliding set ofprotons and antiprotons in the Tevatron (36 bunchesof each), and collider shot setup is the process ofloading a store into the Tevatron.The luminosity is a measure of the number of col-lisions expected. For an intersecting storage ring col- Proceedings of the DPF-2009 Conference, Detroit, MI, July 27-31, 2009 lider, the instantaneous luminosity is given by L = f nN N /A , where f is the revolution frequency ofa bunch, n is the number of bunches in each beam, N is the number of particles in each bunch [pro-tons(antiprotons) in the case of the Tevatron], and A is the cross section of the beam. The instantaneousluminosity decays over time as the number of parti-cles decreases due to collisions and other losses andthe size of the beam in phase space increases. At theTevatron, the instantaneous luminosity is in the rangeof 10 cm − s − , and the integrated luminosity overthe course of a store is on the order of a few pb − . II. MODEL FOR OPTIMIZATION OFINTEGRATED LUMINOSITY
At this point in the collider Run II, all major up-grades have been incorporated and the antiprotonstacking rate is not expected to undergo any morelarge increases. With relatively stable and repro-ducible conditions, we ask ourselves how to make themost of what we have. We take the antiproton pro-duction rate to be the limiting factor in integrated lu-minosity, and by using recent historical data to modelthe performance of the accelerator complex, find theoptimal use of antiprotons for maximizing integratedluminosity.In the model, proton parameters are kept fixed sincethey have little variation in practice, with intensitiesaround 320 × per bunch and emittances (a measureof their area in phase space) around 16 − π mm mradat 8 GeV in the Main Injector.Luminosity parameters are obtained using datafrom recent stores. The dependence of initial lumi-nosity on the number of antiprotons in the stash isquite reproducible, as is the luminosity lifetime be-havior, which also happens to be roughly independentof initial luminosity.For the antiprotons, the model takes into accountthe effective production rate, including the stackingrate, the pbar transfer efficiency, lifetimes in both theAccumulator and Recycler, and any interruption tostacking during pbar transfers, and also the efficiencyof antiproton transfers to the Tevatron. The modelcalculates antiproton production and integrated lumi-nosity over the course of a week given a target stashsize at which the existing store is terminated and theantiprotons are transferred to the Tevatron for a newstore.Figure 2 shows the number of antiprotons in thestack and stash as well as the integrated luminosityover the course of a week for a target stash size of ∼ × , a peak stacking rate of 30 × / h r ,and a collider shot setup time of 1.5 hrs. Figure 3shows the output weekly integrated luminosity as afunction of input target stash size. The model indi-cates that the integrated luminosity is maximized by FIG. 2: Number of antiprotons in the Accumulator (blue)and Recycler (red) along with integrated luminosity (navy)in a modeled week.FIG. 3: Output of the model showing predicted weeklyintegrated luminosity as a function of target stash size. a target stash size of ∼ − × . These are theconditions under which the accelerator complex oper-ated for the last year or so. The predicted maximumweekly integrated luminosity of ∼
74 pb − was foundto be quite accurate when compared with two “per-fect” weeks in which we had very little downtime; theluminosity integrated in one of those weeks, totaling FIG. 4: Luminosity integrated over the week of April 13,2009, which totaled 73 pb − . roceedings of the DPF-2009 Conference, Detroit, MI, July 27-31, 2009 FIG. 5: Output of a tool used to determine store duration showing instantaneous and integrated luminosity over thecourse of a collider store along with the number of antiprotons available, which predicts future integrated luminosity ofthe current store and of a potential new store from the antiprotons available.
73 pb − , is shown in Fig. 4.An additional tool is available which is used forstore-by-store decision making and which also con-firms the conclusion drawn from the model discussedabove. Figure 5 shows the output of that tool, whichis updated hourly during the course of a store. Redand blue points show the instantaneous luminosity asmeasured by CDF and D0, and curves fitting the lu-minosity decay are also shown. Green points showthe integrated luminosity as determined from the de-cay fit, and as measured by CDF is shown in cyan.The number of antiprotons in the stack is shown inorange, in the stash in purple, and total in pink. Thetool predicts the initial luminosity of a new store fromthe number of antiprotons available based on recentstores from similar stash size and plots that as a sep-arate red point. The integrated luminosity over thefirst hour of that potential new store is also predictedbased on the same historical data, and is comparedto the luminosity which would be integrated over thenext 2.5 hours of the current store based on the lumi-nosity decay fit (comparing 2.5 hours of the existingstore versus 1.5 hours of shot setup plus 1 hour of thethe store).In the left plot of Fig. 5, the number of antipro-tons available is around our typical target stash of375 × , the predicted luminosity of the currentstore over 2.5hrs is about 600 nb − , while the pre-dicted luminosity over the first hour of a new store isabout 800 nb − , confirming that when we reach thetarget stash size determined from our model, puttingin a new store will lead to more integrated luminositythan continuing to run the existing one. The plot on the right of Fig. 5 shows that this tool is especiallyuseful during non-standard running conditions, suchas when we have long stacking downtimes. We findthere that although the stash is only about 275 × ,the luminosity has decayed enough that we would beintegrating more with a new store than with the ex-isting one (790 nb − vs. 750 nb − ).This tool is used to make store-by-store operationaldecisions, especially in response to interruptions ofstandard operating conditions, while the model di-rects our general plan to maximize integrated lumi-nosity over the time scale of multiple stores. By vary-ing conditions in the model, we also gain insight intoareas to attack in order to improve integrated lumi-nosity. The model is also rerun whenever changes inthe performance the accelerator complex occur in or-der to ensure that our operations are optimized. Themodel is described in more detail in Ref. [1].The model determines the best target stash sizegiven certain conditions, such as antiproton produc-tion, shot setup duration, proton intensity and emit-tances, and the behavior of the Tevatron in initial lu-minosity vs. stash size and luminosity decay. If wecan improve these conditions, we can make additionalgains. Section III describes additional efforts to op-timize antiproton production, and Sec. IV describessome recent operational improvements in these otherareas. Proceedings of the DPF-2009 Conference, Detroit, MI, July 27-31, 2009
FIG. 6: Antiproton stacking rate, stack, and stash size as a function of time over the course of a collider store, a Tevatrondown time, and another collider store, with and without “partial mining” of antiprotons from the Recycler.
III. OPTIMIZATION OF ANTIPROTONPRODUCTIONA. Optimizing pbar transfers
One feature of overall antiproton production overwhich we have a lot of control is the stack size atwhich a pbar transfer from Accumulator to Recycleris initiated. In favor of frequent transfers from smallstacks is the fact that the stacking rate declines as thestack size increases. Up to stack sizes of ∼ × ,the rate is around 25-30 × / hr, while at a stacksize of 80 × , the rate drops to about 20 × / hr[2]. However, frequent transfers would not be opti-mal if there were long durations in which we were notstacking during the transfer process. Other factorsrelated to the transfer process are the percentage ofantiprotons removed from the stack and the transferefficiency to the Recycler, both of which depend on thestack size and the number of transfers in a set. Thelifetime of antiprotons in the Recycler is also a factor,and we must take into consideration the need for ad-equate cooling between the last pbar transfer and acollider shot. Given the current conditions, we foundthat a set of two transfers initiated when the stackreached 25 × was more optimal than our previousmode with a varying number of transfers from a stackof 40 × . B. Rapid transfers
As mentioned, non-stacking time during a pbartransfer makes frequent transfers inefficient. Mucheffort has been put into speeding up pbar transfersand improving transfer efficiency over the past sev-eral years [3]. The time to prepare and execute the transfer has been reduced from as much as an hourdown to less than 5 minutes, with the non-stackingtime now being negligible. At the same time, transferefficiencies have increased from 80-90% to an averageof 95%.
C. “Partial mining”
Another gain in overall antiproton production hasbeen achieved though a new operational method inthe Recycler referred to as “partial mining”, whichenables only a percentage of the stash to be extracted(“mined”) for a collider shot, without compromise ofcooling or lifetime of the antiprotons. In the newmethod, RF manipulations separate the beam to ex-tracted from the beam to be left behind in the Recy-cler. Because of the RF bucket size, there are limita-tions such that the fraction of beam extracted must bein the range 20-80%, and no more than 150 × canbe left behind. This capability allows us to maintainregular pbar transfers even after the target stash sizehas been reached, thus maintaining small stack sizesin the Accumulator and the associated higher stack-ing rates and better pbar transfer efficiencies. In addi-tion to the gains in overall antiproton production, thisgives us more flexibility in scheduling collider shots towork around problems, and can give us a backup sup-ply of antiprotons if we have a failure during collidershot setup.Figure 6 shows the effect of a few hour Tevatrondowntime with and without partial mining. On theleft, we see that without partial mining, the targetstash size of 375 × is reached and instead of contin-uing pbar transfers, the stack size grows to 120 × with the stacking rate dropping from 27 × / hrdown to 16 × / hr. Once the Tevatron is ready, the roceedings of the DPF-2009 Conference, Detroit, MI, July 27-31, 2009 × , andpartial mining allows us to extract only 370 × for the collider shot. Regular pbar transfers continue,and we have reached the target stash for yet anothercollider shot about 2hrs sooner than in the previousexample, thus trading 2hrs of lower luminosity at theend of a store for the higher luminosity of an newstore, leading to more integrated luminosity overall. IV. OTHER OPERATIONALIMPROVEMENTSA. Reducing collider shot setup time
When we implemented the model described inSec. II, where the store duration was determined bythe target stash size of ∼ × , the average storeduration dropped from ∼ B. Increasing proton brightness
Brighter beam means more particles in a smallerarea; it is defined as intensity over emittance. Increas-ing brightness leads to higher instantaneous luminos-ity. In the Tevatron, because of very small antiprotonemittances achieved in the Recycler, the antiprotonbeam is brighter than the proton beam. A big differ-ence in brightness leads to beam-beam effects, wherethe bright antiproton beam shifts the proton beam inphase space where it may be less stable. We inten-tionally “blow up” the antiproton emittances in theTevatron before collisions in order to better match the proton emittance. Brighter protons would there-fore increase instantaneous luminosity, reduce lossesdue to beam-beam effects, and allow brighter antipro-tons which increases luminosity again.We have achieved brighter protons by scraping theproton halo with collimators in the Main Injector be-fore they are accelerated and injected into the Teva-tron. That is, we start with higher intensity beam,and scrape to nominal intensity but smaller emittance.As shown in Fig. 7, this has improved initial lumi-nosities by about 3-4%. Smaller proton emittancesalso lead to improved transfer and acceleration effi-ciencies, and the improved dynamic aperture of themachine has reduced quenching, where beam falls outof the machine catastrophically.
FIG. 7: Effect of proton scraping on initial luminosity asa function of number of antiprotons used for the collidershot.FIG. 8: Proton brightness in the Tevatron as a functionof time, showing the effect of operational and machine im-provements.
The proton brightness is shown in Fig. 8 as a func-tion of time (referenced by store number). Along with
Proceedings of the DPF-2009 Conference, Detroit, MI, July 27-31, 2009 scraping protons in the Main Injector, an effect wasobserved dependent on the number of turns worth ofprotons loaded into the Booster. Although 11 turnsled to higher intensity than 10, the emittance was alsosignificantly larger, such that the overall brightnesswas higher for 10 turns. Also marked is the periodfollowing the realignment of a section of beampipe inthe CDF collision hall which was an aperture restric-tion. This allowed us to go to even higher protonbrightness without quenching. At high antiproton in-tensities, where beam-beam effects are an issue, thefraction of protons surviving acceleration in the Teva-tron went from 95-97% to 97-98% with the removal ofthe aperture restriction.
C. Consistency and reliability
Following the guidance of the model discussed inSec. II, we have a target stash size for collider shot se-tups. Using that consistent stash size, Recycler cool-ing and Tevatron tunes are also consistent from storeto store.Tevatron stability has also been improved by auto-matically setting the proton tune based on antiprotonintensity, an orbit stabilization program, and monitor-ing lattice stability. Beam-beam effects have been re-duced by controlling the antiproton/proton emittanceratio, both by “blowing up” antiprotons and scrapingprotons. The realignment of the beampipe near CDFto remove the aperture restriction has also improvedstability.Another area where we have improved reliabilityis recovering from a collision hall access. After theexperiments make an access, there is an overhead ofapproximately 2hrs before we are ready for collidershot setup. Since the low-beta quadrupoles inside thecollision halls must be turned off for access, follow-ing an access, they are turned back on and the Teva-tron is ramped and brought through a “dry squeeze”,where the low-beta quads are ramped without beamin the Tevatron. The same process may be repeatedwith proton beam, referred to as a “wet squeeze”, inorder to check and correct orbits. If corrections areneeded, another wet squeeze is performed. Figure 9shows that the initial luminosity of a store followingan access is generally about 3% higher if a wet squeezeis performed. Whether or not this slight increase ininitial luminosity is worth the time it takes to per-form the wet squeeze(s), the fact that going throughthe process makes it less likely to develop problemsduring shot setup makes it worth the effort.
V. RESULTS
Figure 10 Shows the weekly integrated luminosityfrom Oct 2007 to present. Three distinct periods can
FIG. 9: Effect of a wet squeeze after a collision hall accesson initial luminosity as a function of number of antiprotonsused for the collider shot. be observed: first where luminosities were below 50pb − / wk, second where the highest integrated lumi-nosities we in the range 50-60 pb − / wk, and finallywhere 50-60 pb − / wk is our average. The second pe-riod begins when we implemented the model describedin Sec. II around the end of April 2008 (note thatthe machine uptime early in that period was low, lessthan 100 hrs/wk through the second week in May).The third period begins after the shutdown in Octo-ber 2008 when the aperture restriction near CDF wasremoved. Most of the other improvements mentionedabove were also implemented during this third period.Figure 11 shows the effective use of available an-tiprotons in terms of integrated luminosity. Lookingat the period of April 2008 and later, we see thatantiprotons are being used more efficiently since themodel has been employed. VI. CONCLUSIONS
We have a model for optimizing integrated luminos-ity at the Fermilab Tevatron which is used to deter-mine the target number of antiprotons for terminatinga collider store and putting in a new one. The imple-mentation of this model has led to an improvement ofapproximately 35% in integrated luminosity.Operational changes have increased overall antipro-ton production, including optimizing pbar transfers,speeding up the time needed to prepare for and exe-cute a transfer, and a new method for leaving behinda fraction of the antiprotons in the Recycler when ex-tracting for a Tevatron store to allow more antiprotonsin the Recycler than we want to use for the collidershot.Other recent operational improvements include de-creasing collider shot setup time, reducing beam-beam roceedings of the DPF-2009 Conference, Detroit, MI, July 27-31, 2009 − even with no further improve-ments. However, we continue to work on operationalimprovements similar to what has been shown hereand hope to continue to surpass expectations. Acknowledgments
Many thanks to my partner Run Coordinator, ConsGattuso, who had the inspiration for many of theseimprovements and many more to come. [1] C. Gattuso, M.E. Convery and M.J. Syphers, “Opti-mization of Integrated Luminosity in the Tevatron,”Proceedings of PAC09, Vancouver, Canada, May 4-8,2009.[2] B. Drendel, J.P. Morgan and D. VanderMeulen,“Operating Procedure Changes to ImproveAntiproton Production at the Fermilab Tevatron Collider,” Proceedings of PAC09, Vancouver, Canada,May 4-8, 2009.[3] J.P. Morgan, D. Vander Meulen and B. Drendel, “Im-provements to Antiproton Accumulator to RecyclerTransfers at the Fermilab Tevatron Collider,” Proceed-ings of PAC09, Vancouver, Canada, May 4-8, 2009.
Proceedings of the DPF-2009 Conference, Detroit, MI, July 27-31, 2009
FIG. 10: Weekly integrated luminosity over the period from the 2007 shutdown to the 2009 shutdown.FIG. 11: Ratio of monthly integrated luminosity to an-tiprotons delivered to the Recycler. roceedings of the DPF-2009 Conference, Detroit, MI, July 27-31, 20099