The origin of the major ice-sheet variations during the last 2.7 million years remains a mystery. Neither the dominant 41 000-year cycles in ?¹⁸O and ice-volume during the late Pliocene and early Pleistocene nor the late-Pleistocene variations near 100 000 years is a linear (''Milankovitch'') response to summer insolation forcing. Both result from non-linear behavior within the climate system. Greenhouse gases (primarily CO₂) are a plausible source of this non-linearity, but confusion has persisted over whether the gases force ice volume or are a positive feedback. During the last several hundred thousand years, CO₂ and ice volume (marine ?¹⁸O) have varied in phase both at the 41 000-year obliquity cycle and within the ~100 000-year eccentricity band. This timing argues against greenhouse-gas forcing of a slow ice response and instead favors ice control of a fast CO₂ response. Because the effect of CO₂ on temperature is logarithmic, the temperature/CO₂ feedback on ice volume is also logarithmic.

In the schematic model proposed here, ice sheets were forced by insolation changes at the precession and obliquity cycles prior to 0.9 million years ago and responded in a linear way, but CO₂ feedback amplified (roughly doubled) the ice response at 41 000 years. After 0.9 million years ago, as polar climates continued to cool, ablation weakened. CO₂ feedback continued to amplify ice-sheet growth at 41 000-year intervals, but weaker ablation permitted ice to survive subsequent insolation maxima of low intensity. These longer-lived ice sheets persisted until peaks in northern summer insolation paced abrupt deglaciations every 100 000±15 000 years. Most ice melting during deglaciations was achieved by the same CO₂/temperature feedback that had built the ice sheets, but now acting in the opposite direction. Several processes have the northern geographic origin, as well as the requisite orbital tempo and phasing, to have been the mechanisms by which ice sheets controlled CO₂ and drove their own feedback.