GEE in R working correlation estimate not positive definite

Question

I'm trying to replicate the following code from SAS in R:

   proc genmod data=skinny ;
      class personid ;
      model  sample1_totalSpermCount_1 = samplePerson_1_byr 
            deepgen1935c totalchildren_1935 /  dist=normal link = identity;
      repeated  subject=personid / type=exch;
      where deepgen_1935>=3;
   run;

Here is the results in SAS:

Here is my code in R:

skinny.gee3 <- gee(sample1_totalSpermCount_1 ~ samplePerson_1_byr + deepgen1935.c + totalchildren_1935,
                   id = PersonID,
                   data = clean[clean$DeepGen_1935 >= 3, ],
                   family = gaussian,
                   corstr = "exchangeable")

I received an error, saying:

Warning message: In gee(sample1_totalSpermCount_1 ~ samplePerson_1_byr + deepgen1935.c + : Working correlation estimate not positive definite

And here is the results in R:

Coefficients:
                       Estimate   Naive S.E.    Naive z  Robust S.E.   Robust z
(Intercept)        1376.0792841 1659.6117221  0.8291574 1906.1089688  0.7219311
samplePerson_1_byr   -0.5671856    0.8434850 -0.6724312    0.9694925 -0.5850335
deepgen1935.c       -12.4639418    7.0401018 -1.7704207    4.3967397 -2.8348146
totalchildren_1935    0.6572002    0.6458183  1.0176240    0.4139229  1.5877359

I can see that the working correlation in SAS is different than in R. Is there an explanation why SAS's GEE run and R's GEE did not?

Thank you!