The cause of ice ages, Koppen decided, is to be found in cool summers, not brutal winters. If summers are too cool to melt all the snow that falls on a given area, more incoming sunlight is bounced back by the reflective surface, exacerbating the cooling effect and encouraging yet more snow to fall. The consequence would tend to be self-perpetuating. As snow accumulated into an ice sheet, the region would grow cooler, prompting more ice to accumulate. As the glaciologist Gwen Schultz has noted: "It is not necessarily the amount of snow that causes ice sheets but the fact that snow, however little, lasts." It is thought that an ice age could start from a single unseasonal summer. The leftover snow reflects heat and exacerbates the chilling effect. "The process is self-enlarging, unstoppable, and once the ice is really growing it moves," says McPhee. You have advancing glaciers and an ice age.

In the 1950s, because of imperfect dating technology, scientists were unable to correlate Milankovitch's carefully worked-out cycles with the supposed dates of ice ages as then perceived, and so Milankovitch and his calculations increasingly fell out of favor. He died in 1958, unable to prove that his cycles were correct. By this time, write John and Mary Gribbin, "you would have been hard pressed to find a geologist or meteorologist who regarded the model as being anything more than an historical curiosity." Not until the 1970s and the refinement of a potassium-argon method for dating ancient seafloor sediments were his theories finally vindicated.