When calculating a*a-b*b the form (a-b)*(a+b) is usually more numerically stable.

[u/mrichter/AliRoot.git] / ITS / AliITSNeuralTrack.cxx

diff --git a/ITS/AliITSNeuralTrack.cxx b/ITS/AliITSNeuralTrack.cxx
index 348dc0947680e5913d8cd01b499909d6eb552832..4f3f06066282c3284ca5cfa24029b31335a3a707 100644 (file)
--- a/ITS/AliITSNeuralTrack.cxx
+++ b/ITS/AliITSNeuralTrack.cxx
@@ -423,8 +423,8 @@ Int_t AliITSNeuralTrack::PropagateTo(Double_t rk)
        Double_t bk = ArgB(rk), bkm1 = ArgB(rkm1);
        Double_t ck = ArgC(rk), ckm1 = ArgC(rkm1);
        
-       Double_t f02 = ck / TMath::Sqrt(1. - aAk * aAk) - ckm1 / TMath::Sqrt(1. - aAkm1 * aAkm1);
-       Double_t f04 = bk / TMath::Sqrt(1. - aAk * aAk) - bkm1 / TMath::Sqrt(1. - aAkm1 * aAkm1);
+       Double_t f02 = ck / TMath::Sqrt((1.-aAk)*(1.+aAk)) - ckm1 / TMath::Sqrt((1.-aAkm1)*(1.+aAkm1));
+       Double_t f04 = bk / TMath::Sqrt((1.-aAk)*(1.+aAk)) - bkm1 / TMath::Sqrt((1.-aAkm1)*(1.+aAkm1));
        Double_t f12 = fTanL * fDt * (1. / rk - 1. / rkm1);
        Double_t f13 = rk - rkm1;