你好，非常感谢！
这个方法是可以实现，但如果直接在SQL语句中直接衍生出变量sum（不要在data步中衍生）,可以做到吗？

据我了解应该不可以，想要得到新的满足要求的sum变量，首先需要排序，而这个排序则需要通过sort过程来完成，SQL过程貌似做不到通过排序来利用逻辑运算符来满足要求。当然也有可能是我水平有限，了解还不够，也许能实现也不一定。

OK，非常感谢！
如果你什么时候实现了，记得上传上来以供学习。

办得到, 跟m的方法是一样的. 但也许分开写比较清楚.

proc sql;
create table xx as
select b.var1, a.var2, b.var3
from (select *, monotonic() as ord, (calculated ord-min(calculated ord)) as ord2  from a1 group by var1) as a right join (select *, monotonic() as  ord, (calculated ord-min(calculated ord)) as ord2 from a2 group by var1) as b
on  a.var1=b.var1 & a.ord2 =b.ord2;


---debug

proc sql;
create table  zz as
select *, monotonic() as ord, (calculated ord-min(calculated ord)) as ord2  from a1 group by var1;

proc sql;
create table  zz2 as
select *, monotonic() as ord, (calculated ord-min(calculated ord)) as ord2  from a2 group by var1;