But when packets gets dropped in between route then respective router informs the source that packets gets dropped due to greater PTUMD(in ipv6) then source will fragment the packet for delivery.
Packets are dropped for various reasons. If that happens due to congestion or corruption there's no source notification.
Only if the path MTU gets too small for a packet ICMPv6 returns a Packet too big notification to the source, so the source can reduce the packet size (possibly by fragmentation but protocols like TCP should adapt their segmentation size/reduce MSS instead).
My question is if ICMP synchronize with source for every issue for retransmisson then reliability of layer3 is possible?
No. ICMP would be required to unfailingly report dropped packets which it doesn't. (That would actually be impossible, when headers are corrupted during transport.)
But we say layer3 is unreliable?
Yes. There are only a few reasons where ICMP reports missing delivery. All other reasons aren't accounted for, so IP isn't reliable. Other L3 protocols may vary though.
The general approach is to keep L3 light-weight - only implement functionality that is absolutely required. It doesn't make sense to implement reliability when there are applications that don't need it. Therefore, reliability is implemented at a higher layer in the protocol stack where required - either in the transport layer/L4 or in the application layer/L7.
