Climate over the past million years has been dominated by glaciation c
ycles with periods near 23,000, 41,000, and 100,000 years. In a linear
version of the Milankovitch theory, the two shorter cycles can be exp
lained as responses to insolation cycles driven by precession and obli
quity. But the 100,000-year radiation cycle (arising from eccentricity
variation) is much too small in amplitude and too late in phase to pr
oduce the corresponding climate cycle by direct forcing. We present ph
ase observations showing that the geographic progression of local resp
onses over the 100,000-year cycle is similar to the progression in the
other two cycles, implying that a similar set of internal climatic me
chanisms operates in all three. But the phase sequence in the 100,000-
year cycle requires a source of climatic inertia having a time constan
t (similar to 15,000 years) much larger than the other cycles (similar
to 5,000 years). Our conceptual model identifies massive northern hem
isphere ice sheets as this larger inertial source. When these ice shee
ts, forced by precession and obliquity, exceed a critical size, they c
ease responding as linear Milankovitch slaves and drive atmospheric an
d oceanic responses that mimic the externally forced responses. In our
model, the coupled system acts as a nonlinear amplifier that is parti
cularly sensitive to eccentricity-driven modulations in the 23,000-yea
r sea level cycle. During an interval when sea level is forced upward
from a major low stand by a Milankovitch response acting either alone
or in combination with an internally driven, higher-frequency process,
ice sheets grounded on continental shelves become unstable, mass wast
ing accelerates, and the resulting deglaciation sets the phase of one
wave in the train of 100,000-year oscillations. Whether a glacier or i
ce sheet influences the climate depends very much on the scale....The
interesting aspect is that an effect on the local climate can still ma
ke an ice mass grow larger and larger, thereby gradually increasing it
s radius of influence.