One of the most fascinating discoveries of recent years has been that of the neurophysiological mechanisms of elementary arithmetic in the macaque monkey. Following earlier work by Thompson et al (1970) and Sawamura et al (2002), Andreas Nieder and Earl Miller, at the Massachusetts Institute of Technology and then at the University of Tübingen, recorded the activity of hundreds of neurons in awake monkeys trained to perform a delayed comparison task of the numbers of two sets of objects. In the prefrontal and intraparietal cortex, they discovered populations of neurons whose rate of discharge varied with the number of objects presented. Some neurons are preferentially activated by a single object, others by two, three, four or five objects (Nieder, 2005), and even up to thirty objects (Nieder & Merten, 2007).
The detailed profile of responses from these neurons indicates a Weberian coding of numbering, coinciding precisely with that inferred from psychophysical studies in humans. In fact, each neuron shows a tuning curve around its preferred numerosity. On a linear scale, the width of the curve increases linearly with the preferred numerosity, which corresponds quantitatively to Weber's law. But the most compact description of neuronal responses is a constant tuning curve, with fixed variability and a Gaussian shape, when the numerosity is represented on a logarithmic scale. This representation is called "log-Gaussian". It implies that, at the level of the population of neurons, the number parameter is represented by a sparse group of neurons according to a partially distributed code that represents approximate numerosity rather than exact cardinality.
