The same algo gives tl 4 in java ... In C++ I use vector of vectors for the capacity matrix as well as in java and the input is with the Scanner class . How can I optimize ? Thanks in advance.

Did you read the FAQ (Frequently Asked Questions) ?

I have TLE4 in pascal! why?

I don't know. Using C++ and good old Dinic you can get 0.015ms. Actually, you can perform greedy initialization in linear time (of matrix size) and get AC with most suboptimal flow algorithms.