Nonlinear optimisation techniques for accelerator performance improvement on-line: recent trials and experiment for the CERN antiproton accumulator

Abstract

The use of function minimisation techniques for optimum design according to given performance criteria is well-known. Given a well-defined criterion and a means of evaluating it precisely, the problem reduces to choosing the best optimisation procedure to suit the problem. Direct search techniques which do not generally rely on the computation of derivatives of the error function are ideal for on-line improvement of the global accelerator performance since the error function is not known analytically, e.g. the number of antiprotons stored in the antiproton accumulator ring on a pulse-to-pulse basis as a function of all the antiproton production and stochastic cooling system parameters. The user-friendliness of the NODAL interpreter at the man-machine interaction level, its capability to easily control and manipulate equipment as well as its capability to synchronise with respect to time events on a cycle-to-cycle basis makes it suitable for an on-line accelerator performance optimisation type of application. A modular procedure, based on the Simplex technique [1] has been implemented recently which allows function minimisation depending on the error function definition module. This enables an easy manipulation of variables and synchronization with machine events. For the antiproton accumulator (AA), while the circulating beam current transformer lacks the resolution to measure the exact number of antiprotons stored on a pulse-to-pulse basis, there are a large number of electrons produced in the production process [2] and a signal emanating from these can be adapted to provide the performance criterion and appropriate parameters used as function variables in the optimisation process. First trials based on optimisation of injection of antiprotons in the AA look promising, but further work is necessary in the direct definition of the error functions.