The input language to an SMT solver is first-order logic (with theories) and as such has no notion of computational operations such as loops.
You can
either use a loop invariant to encode an arbitrary loop iteration (and the pre- and post-state of the loop) and prove your relevant properties with respect to that arbitrary iteration, which is what deductive program verifiers such as Boogie, Dafny or Viper do
or, if the number of iterations is statically known, you unroll the loop and basically use single static assignment form to encode the different unrollings
For your loop, the latter would look as follows (not using proper SMT syntax here because I'm lazy):
declare x0, a0 // initial values
declare a1, x1 // values after first unrolling
x0 > 10 && x0 < 100 ==> a1 == a0 + x0 && x1 == x0 + 1
declare a2, x2 // values after second unrolling
x1 > 10 && x1 < 100 ==> a2 == a1 + x1 && x2 == x1 + 1
...
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…