... bulge

The data are shared with the MEGA collaboration.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... filters

Each night $\sim$20(11) images are taken in the R(I) filter, which are then averaged to get the single image per night that we use in the following analysis.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...Ansari et al. 1997)

An alternative method based on image subtraction is currently used by the MEGA collaboration (Tomaney & Crotts 1996).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... flux

With the notable exception of geometrical alignment, throughout the analysis we substitute for the pixel value the sum of the fluxes taken in a square of $5\times 5$ pixels around the central pixel, i.e., the corresponding superpixel value.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... constant)

Note that this same test can be applied to resolved-source microlensing, provided that "background'' is replaced by "baseline'', i.e. the combined flux from the true source plus any unresolved blended companions. That is, blending breaks achromaticity only for the ratio of the total fluxes in two bands, not differences in these fluxes from baseline.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... bands

For this analysis we do not use the available g' band data.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... astrometric

The INT pixel size is $0\hbox{$.\!\!^{\prime\prime}$ }33$ versus $0\hbox{$.\!\!^{\prime\prime}$ }50$ for MDM pixels.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... fits

We allow here a larger margin because we are aware that probing such an effect can be difficult, especially when dealing with variations showing both a long time width and a red colour, these being possibly due to red variable stars that do not necessarily show a strictly periodic and regular behaviour.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...

We note that, in principle, for a cluster of pixels with Q>100, we can have L₁<100.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... days

We note that these short timescales are consistent with what we expect from the Monte Carlo simulations discussed in Paper I.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.