Measurements of synchrotron radiation emitted by 30-MeV runaway electrons in the TEXTOR-94 tokamak show that the runaway population decays after switching on neutral beam injection (NBI). The decay starts only with a significant delay, which decreases with increasing NBI heating power. This delay provides direct evidence of the energy dependence of runaway confinement, which is expected if magnetic modes govern the loss of runaways. Application of the theory by Mynick and Strachan [Phys. Fluids

more »

... 24, 695 (1981)] yields estimates for the "mode width" ͑d͒ of magnetic perturbations: d , 0.5 cm in Ohmic discharges, increasing to d 4.4 cm for 0.6 MW NBI. PACS numbers: 52.55.Fa One of the outstanding issues in thermonuclear fusion research remains the anomalous conduction of heat by the electrons in the plasma. In a tokamak, the hot plasma is confined in a toroidal geometry by means of magnetic fields. The field topology is such that field lines lie on nested toroidal surfaces. Transport in the direction perpendicular to the surfaces is reduced by many orders of magnitude by the presence of the field. However, the measured heat fluxes carried by the electrons exceed the theoretically achievable minimum by 1-2 orders of magnitude. This anomaly is generally ascribed to turbulence, which may be of electrostatic or of magnetic nature, or both. There has been extensive research into transport caused by electrostatic fluctuations. Recently, means have been found to greatly reduce the heat loss caused by these [1]. Magnetic turbulence is more difficult to diagnose, since perturbing fields of the orderB͞B 10 25 can already contribute significantly to the heat flux carried by the electrons. The only direct measurements ofB in the core of a tokamak plasma, using the cross-polarization scattering of microwaves, did show the presence ofB at transport relevant levels in Tore Supra [2, 3] . Electrons with energy much higher than the thermal energy, in principle, can provide a probe to study magnetic turbulence, since diffusion due to electrostatic turbulence scales with y 21 , whereas the magnetically induced diffusion scales as y, where y is the electron velocity. Moreover, since in a plasma the mean free path of an electron scales as y 4 , collisional transport is negligibly small for high energy electrons. The absence of collisions is also the reason why in tokamak plasmas of sufficiently low density a small fraction of the electrons (so-called runaway electrons) undergo a free fall acceleration and can reach energies in the MeV range, in a background plasma with a temperature of ϳ1 keV. In several studies, runaway electrons have been used to assess magnetic turbulence. One principal difficulty is that runaway electrons in the 1-MeV energy range cannot be diagnosed until they leave the plasma and hit the wall and produce x rays. Thus, in [4] experimental techniques have been used to probe magnetic turbulence in the edge of the plasma. Direct observation of runaway electrons in the center of the plasma column has been performed at the TEXTOR-94 experiment, making use of the synchrotron emission. This method diagnoses runaway electrons in a much higher energy range, typically 25 30 MeV. The high energy poses another problem. The orbits of electrons of such high energy are shifted with respect to the magnetic field topology by a few cm, which strongly reduces their sensitivity to magnetic perturbations with a radial correlation length smaller than this orbit shift [5] . In fact, this orbit shift is the reason why high energy runaway electrons are often observed to have much better confinement than then thermal electrons (e.g., [6, 7] ). So, the question is, How can the observable high energy runaway electrons be used to probe magnetic turbulence with a scale much smaller than their orbit shift? In this Letter, an analysis is presented based on a novel observation: when auxiliary heating is applied to a plasma with a preexisting runaway population, this population is observed to decrease. Clearly, the runaway confinement is deteriorated by the power input, which is expected since it is well known that confinement of heat and thermal particles also degrades with increasing heating power. However, the decline of the runaway population does not start instantaneously after the heating is switched on, but is significantly delayed. It is shown that this delay is due to loss of runaway confinement at a lower energy, which only later appears as reduced influx of observable runaway electrons at high energy. In this picture, there is a certain energy below which the runaway electrons are lost. The delay time in the radiation signals is the time the electrons at this energy need to be accelerated to the energy where they become observable. Thus, the delay time can be related to a 3606 0031-9007͞00͞84(16)͞3606(4)$15.00