If Chess is a draw game then why does white lose here?

The problem is that there is a limited horizon. Even a good move within the search depth can turn out to be the losing one in the long run. 0/0/0 doesn’t mean error-free at all.

Yes, the game was lost, so there must be at least one mistake, whatever the engine tells.

You can use strong cloud servers I think f4 is mistake

#23
Yes, 36 f4 may be the culprit as said above.

#20
47 b4 I do not think white could save the game at that point.

47.b4?? 47.Kf2!

@em48 #25
Still after 47. Kf2 a draw?

@ujcn yes

#25
Yes, it seems 47 Kf2 can hold the draw, the point being that black can win pawn d6, but white then wins pawn g5 with Rd3-f3-f5xg5
For that reason I think black should have taken the pawn right away with 43...Rxd6
So IMHO 36 f4? loses 43...Rc5? draws, 47 b4? loses again

I've found the solution to your question:

47. b4?? was the first serious mistake
47... e4 48. Rd1 Rd8 49. bxa5 bxa5 50. Rf1 Rxd6 51. Rf5 Rd3+ 52. Kg2 Rd5 black is better

48. Rd4?? is the decisive error losing the game
after
48... Ke5 49. d7 Rd8 50. Rd1 axb4 51. Kf2 Kf4 52. Ke2 b3 53. Rd2 b2 l and it's GGs.

It was better to play
47. Kf2 a draw, probably the best move but
47. d7! is an interesting computer alternative... Rd8 48. Kf2 Rxd7 49. Rh3 and it's equal again.

I think @tpr is correct, in that each position in chess must be either a win for one side or a draw. The issue is that no current system can evaluate far enough to give an answer for all positions, currently this can be done for all positions with 7 pieces or less, and software like FinalGen and Freezer can compute some 8 and 9 piece positions to completion and give a definite result (White/Black wins in N, or draw). It is estimated that computing a 32 piece tablebase (to solve chess from the opening position) with modern computing hardware would take longer than the current lifespan of the universe. This is why a computer cannot give an absolute result for all positions, otherwise all evaluations would be #N, 0 or #-N.

The best computers can do now is a "confidence" value. This can change and sometimes be completely wrong (see fortresses). The horizon effect can lead to evaluations which gradually favour one side, even though the other isn't making any "bad" moves.... all that is happening is that a won endgame is coming closer and closer but until that arrives, the computer is making ab "educated guess".