For n≥1 we have
an+1−an=(an−an−1)+(an−1−an−2)+⋯+(a1−a0)=an−a0,
which yields an+1=2an−a0. The sequence can therefore be written in the form
a0=2012, a1=2012+d, a2=2012+2d, …, an=2012+2n−1d, …
For n≥3 we therefore have an=4⋅(503+2n−3d). Since 503+2n−3d≡3(mod4) holds for n≥5, we see that an can never be a perfect square for n≥5. The only numbers in the sequence that can possibly be perfect squares are therefore
a0=2012, a1=2012+d, a2=2012+2d, a3=4(503+d) or a4=4(503+2d)
with 1≤d≤43. Since 2012=a0<a4≤2012+8⋅43=2356, the only perfect squares that can occur in the sequences must lie between 2012 and 2356, i.e.
452462472482=2025=2012+13,=2116=2012+104=2012+2⋅52=2012+4⋅26=2012+8⋅13,=2209=2012+197 or=2304=2012+292=2012+2⋅146=2012+4⋅73.
Since we must have d≤43, the only sequences of the required type containing perfect squares are those with d=13 (which contains 2025 and 2116) and with d=26 (which also contains 2116). □