我的C ++代码在MS C ++编译器上运行完美，但在g ++编译器上给了我NaN。为什么？

Question

I'm simulating a biological model involving several (27) stiff ordinary differential equations using C++. My program runs perfectly under MS C++ 2010 expression compiler but fails under the g++ compiler (NetBeans 6.8, Ubuntu 10.04 LTS). The problem is that some of the variables become NaN . The following are the values of the variable Vm after each step of the program under the g++ compiler:

-59.4 -59.3993 -59.6081 100.081 34.6378 -50392.8 nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan nan

And here is the output of the same code without any change under the MS C++ compiler:

-59.4   -59.3993    -59.3986    -59.3979    -59.3972    -59.3966    -59.3959    -59.3952    -59.3946    -59.3939    -59.3933    -59.3926    -59.392 -59.3914    -59.3907    -59.3901    -59.3895    -59.3889    -59.3883    -59.3877    -59.3871    -59.3865    -59.3859    -59.3853    -59.3847    -59.3841    -59.3836    -59.383 -59.3824    -59.3819    -59.3813    -59.3808    -59.3802    -59.3797    -59.3791    -59.3786    -59.3781    -59.3775    -59.377 -

I've only used the "cmath" and "fstream" libraries.

Where is the problem? The code in both scenarios is exactly the same.

Edit 1:

OK guys, here is the whole code:

#include <iostream>
#include <fstream>
#include <cmath>
using namespace std;
void FCN(void);

const int TMAX = 10000; //[ms] simulation time
const int TSTEP = 1;
const int MXSTEP = TMAX / TSTEP;
int ISTEPPRINT = MXSTEP / 5000; //time step for storing on disc

int ISTEP = 0;

const double R = 8341.0;
const double temp = 293.0;
const double F = 96487.0;
const double RT_F = R*temp / F;
const double z_K = 1;
const double z_Na = 1;
const double z_Ca = 2;
const double z_Cl = -1;
const double N_Av = 6.022e23;


double Ca_o = 2.0;
double Na_o = 140.0;
double Cl_o = 129.0;
double K_o = 5;

double NE = 0;
double NO = 0;

double cGMP = 0; //[mM]    [cGMP]i
double cGMPprime = 0; //var
double IP3 = 0; //[mM]    [IP3]i
double IP3d = 0; //var
double IP3prime = 0; //var
double DAG = 0; //[mM]
double DAGprime = 0; //var

double Ca_u = 0.66;
double Ca_r = 0.57;
double Ca_i = 68e-6;
double Na_i = 8.4;
double K_i = 140;
double Cl_i = 59.4;
double V_m = -59.4;
double V_mprime;
double Na_iprime;
double K_iprime;
double Cl_iprime;
double Ca_iprime;
double vol_i = 1;
double I_Natotm;
double I_Ktotm;
double I_Cltotm;
double I_Catotm; //[pA]    total membrane Ca current

//Reversal potentials
double E_Ca; //[mV]
double E_Na; //[mV]
double E_K; //[mV]
double E_Cl;

//Membrane capacitance and area
double C_m = 25.0;
double A_m = C_m / 1e6;

//Voltage dependent calcium current I_CaL
double I_CaL;
double P_CaL = 1.88e-5;
double d_L;
double d_Lprime;
double d_Lbar;
double tau_d_L;
double f_L;
double f_Lbar;
double tau_f_L;
double f_Lprime;

//Delayed rectifier current I_K
double I_K;
double g_K = 1.35;
double p_K;
double p_Kbar;
double V_1_2 = -11.0;
double k = 15.0;
double tau_p_K;
double p_Kprime;
double q_1 = 1;
double q_2 = 1;
double q_bar;
double q_1prime;
double q_2prime;


double Pmin_NSC = 0.4344;
double Po_NSC;
double PNa_NSC = (5.11e-7);
double PCa_NSC = (5.11e-7)*4.54;
double PK_NSC = (5.11e-7)*1.06; //
double d_NSCmin = 0.0244;
double K_NSC = 3.0e-3;
double INa_NSC;
double ICa_NSC;
double IK_NSC;
double I_NSC;

//KATP current I_KATP
double I_KATP; //[pA]    background K current
double g_KATP = 0.067; //[nS]    max. background K current conductance

//Inward rectifier current I_K_i
double I_K_i; //[pA]
double g_maxK_i; //[nS]    max. slope conductance
const double G_K_i = 0; //        inward rectifier constant
const double n_K_i = 0.5; //        inward rectifier constant

//Calcium-activated potassium current I_KCa
double I_KCa;
double i1_KCa;
double P_BKCa = 3.9e-13;
double N_BKCa = 6.6e6;
double P_KCa;
double p_obar;
double V_1_2_KCa;
double p_f;
double p_s;
double p_fprime;
double p_sprime;
double tau_pf = 0.84;
double tau_ps = 35.9;
double dV_1_2_KCa_NO = 46.3;
double R_NO;
double dV_1_2_KCa_cGMP = 76;
double R_cGMP;

double k_leak = 1;
double R_00;
double R_01 = 0.9955;
double R_10 = 0.0033;
double R_11 = 4.0e-6;
double R_01prime;
double R_10prime;
double R_11prime;
const double Kr1 = 2500.0;
const double Kr2 = 1.05;
const double K_r1 = 0.0076;
const double K_r2 = 0.084;
double I_up;
const double I_upbar = 3.34 * (k_leak + 1);
const double K_mup = 0.001;
double I_tr;
const double vol_u = 0.07;
double tau_tr = 1000.0;
double I_rel;
const double vol_r = 0.007;

const double tau_rel = 0.0333; //[ms]
const double R_leak = 1.07e-5 * (k_leak); ////equal to R_10^2 during concentration clamp
//        time constant of SR release
double Ca_uprime; //        dCa_u/dt
double Ca_rprime; //        dCa_r/dt

double S_CM;
const double K_d = 2.6e-4;
const double S_CMbar = 0.1;
double CaCM;
const double K_dB = 5.298e-4;
const double B_Fbar = 0.1;
const double vol_Ca = 0.7;
const double CSQNbar = 15;
const double K_CSQN = 0.8;

double I_PMCA;
double I_PMCAbar = 5.37;
double K_mPMCA = 170e-6;

double I_NaK; ////[pA]    Na/K pump
double I_NaKbar = 2.3083;
const double K_mK = 1.6;
const double K_mNa = 22;
const double Q_10_NaK = 1.87;

double I_NCX;
const double gamma2 = 0.45; //
double g_NCX = 0.000487; //[nS]
double d_NCX = 0.0003; //
double Fi_F; //
double Fi_R; //

double I_NaKCl_Cl; //[pA]
double L_cotr = 1.79e-8;

double I_M = 0; //[pA]
double I_MCa = 0;
double I_MNa = 0; //[pA]    Na component
double I_MK = 0;

double I_SOC; //[pA]
double I_SOCCa;
double I_SOCNa; //[pA]    Na component
const double g_SOCCa = 0.0083; //[nS]
const double g_SOCNa = 0.0575; //[nS]
const double H_SOC = 1;
const double K_SOC = 0.0001;
const double tau_SOC = 100;
double P_SOCbar;
double P_SOC = 0;
double P_SOCprime;

//Chloride currents
double I_Cl;
double g_Cl = 0.23;
double alpha_Cl;
double P_Cl;

//Stimulation current
double I_stim = 0; //[pA]

//IP3 receptor
double I_IP3;
double I_IP3bar = 2880e-6; //[1/ms]
double K_IP3 = 0.12e-3;
double K_actIP3 = 0.17e-3;
double K_inhIP3 = 0.1e-3; //[mM]
double h_IP3;
double k_onIP3 = 1.4;
double h_IP3prime;

double R_TG = 2e4;
double K_1G = 0.01;
double K_2G = 0.2;
double k_rG = 1.75e-7;
double k_pG = 0.1e-3;
double k_eG = 6e-6;
double ksi_G = 0.85;
double G_TG = 1e5;
double k_degG = 1.25e-3;
double k_aG = 0.17e-3;
double k_dG = 1.5e-3;
double PIP2_T = 5e7;
double r_rG = 0.015e-3;
double K_cG = 0.4e-3;
double alpha_G = 2.781e-8;
double vol_IP3 = vol_i;
double gamma_G = N_Av*vol_IP3 * 1e-15;
double RS_G = R_TG*ksi_G;
double RS_PG = 0;
double PIP2; //
double r_hG;
double G;
double delta_G; //
double RS_Gprime;
double RS_PGprime;
double Gprime;
double PIP2prime;
double rho_rG;

//cGMP formation
double k1sGC = 2e3; //[1/mM/ms]
double k_1sGC = 15e-3; //[1/ms]
double k2sGC = 0.64e-5; //[1/ms]
double k_2sGC = 0.1e-6; //[1/ms]
double k3sGC = 4.2; //[1/mM/ms]
double kDsGC = 0.4e-3;
double kDact_deactsGC = 0.1e-3; //[1/ms]
double V_cGMP = 0; //
double V_cGMPprime;
double V_cGMPmax = 0.1 * 1.26e-6; //[mM/ms]
double V_cGMPbar;
double B5sGC = k2sGC / k3sGC;
double A0sGC = ((k_1sGC + k2sGC) * kDsGC + k_1sGC*k_2sGC) / (k1sGC*k3sGC);
double A1sGC = ((k1sGC + k3sGC) * kDsGC + (k2sGC + k_2sGC) * k1sGC) / (k1sGC*k3sGC);
double kpde_cGMP = 0.0695e-3; //[1/ms]
double tausGC;

const int N = 27;
double Y[N], Y1[N], YPRIME[N];
int N1 = 33;

double T = 0;

int main(void) {

    ofstream fileY, fileY1, fileT;
    // initial conditions SMC
    //ICaL
    d_Lbar = 1.0 / (1 + exp(-(V_m) / 8.3));
    d_L = d_Lbar;
    f_Lbar = 1.0 / (1 + exp((V_m + 42.0) / 9.1));
    f_L = f_Lbar;
    //IKCa
    R_NO = NO / (NO + 200e-6);
    R_cGMP = pow(cGMP, 2) / (pow(cGMP, 2) + pow(0.55e-3, 2));
    V_1_2_KCa = -41.7 * log10(Ca_i) - 128.2 - dV_1_2_KCa_NO * R_NO - dV_1_2_KCa_cGMP*R_cGMP;
    p_obar = 1 / (1 + exp(-(V_m - V_1_2_KCa) / 18.25));
    p_f = p_obar;
    p_s = p_obar;
    //I_K
    p_Kbar = 1 / (1 + exp(-(V_m - V_1_2) / k));
    p_K = p_Kbar;
    q_bar = 1.0 / (1 + exp((V_m + 40) / 14));
    q_1 = q_bar;
    q_2 = q_bar;
    //IP3 receptor
    h_IP3 = K_inhIP3 / (Ca_i + K_inhIP3);
    //norepinephrine receptor
    PIP2 = PIP2_T - (1 + k_degG / r_rG) * gamma_G*IP3;
    r_hG = k_degG * gamma_G * IP3 / PIP2;
    G = (K_cG + Ca_i) / (alpha_G * Ca_i) * r_hG;
    delta_G = k_dG * G / (k_aG * (G_TG - G));

    Y[0] = V_m;
    Y[1] = d_L;
    Y[2] = f_L;

    Y[3] = p_K;

    Y[4] = q_1;

    Y[5] = p_f;

    Y[6] = p_s;

    Y[7] = R_01;
    Y[8] = R_10;
    Y[9] = R_11;
    Y[10] = Ca_u;
    Y[11] = Ca_r;
    Y[12] = Ca_i;
    Y[13] = Na_i;
    Y[14] = K_i;
    Y[15] = q_2;

    Y[16] = P_SOC;
    Y[17] = Cl_i;
    Y[18] = h_IP3;
    Y[19] = RS_G;
    Y[20] = RS_PG;
    Y[21] = G;
    Y[22] = IP3;
    Y[23] = PIP2;
    Y[24] = DAG;
    Y[25] = V_cGMP;
    Y[26] = cGMP;

    ISTEP = -1;
    T = 0.0 - TSTEP;

    fileY.open("Y.txt");
    fileY1.open("Y1.txt");
    fileT.open("T.txt");

    for (;;) {
       ISTEP = ISTEP + 1;
       T = T + TSTEP;

       //Norepinephrine
       if (T > 10000) NE = 1e-3; //NE [mM] beginning of stimulation
       if (T > 70000) NE = 0; //end of stimulation

       //Nitric oxide
       //IF (T>30000)  NO = 1D-3 //NO [mM]
       //IF (T>70000)  NO = 0

       //Extracellular potassium
       //IF (T>10000)  K_o = 30
       //IF (T>70000)  K_o = 5

       //Current
       //IF (T>10000) I_stim = 5 //I_stim [pA] current injection
       //IF (T>40000) I_stim = -5
       //IF (T>70000) I_stim = 0

       // For the time being, I just interested in Y[0] values (which is V_m actually)
       fileY << Y[0];
       fileY << "\t";

       if ((ISTEP % ISTEPPRINT) == 0) {

           //            for (int i=0; i< N; i++) {
           //            fileY << Y[i];
           //            fileY << "\t";
           //            }
           //            fileY << endl;


           //            for (int i=0; i< N1; i++) {
           //            fileY1 << Y1[i];
           //            fileY1 << "\t";
           //            }
           //            fileY1 << endl;
           //
           //
           //
           //            fileT << T;
           //            fileT << "\t";
       }

       FCN();
       for (int i = 0; i < N; i++) {
           Y[i] = Y[i] + TSTEP * YPRIME[i];
       }
       //     disp(Yconcat(1))
       if (ISTEP == MXSTEP)
           break;


    }
    cout << "It is done!" << endl;
    cout << Y[0] << endl;

    fileY.close();
    fileY1.close();
    fileT.close();
    return 0;
}

void FCN(void) {
    V_m = Y[0];
    d_L = Y[1];
    f_L = Y[2];
    p_K = Y[3];
    q_1 = Y[4];
    p_f = Y[5];
    p_s = Y[6];
    R_01 = Y[7];
    R_10 = Y[8];
    R_11 = Y[9];
    Ca_u = Y[10];
    Ca_r = Y[11];
    Ca_i = Y[12];
    Na_i = Y[13];
    K_i = Y[14];
    q_2 = Y[15];
    P_SOC = Y[16];
    Cl_i = Y[17];
    h_IP3 = Y[18];
    RS_G = Y[19];
    RS_PG = Y[20];
    G = Y[21];
    IP3 = Y[22];
    PIP2 = Y[23];
    DAG = Y[24];
    V_cGMP = Y[25];
    cGMP = Y[26];

    //-------------------------------------- Model equations  ---------------------------------------------

    //cGMP formation
    V_cGMPbar = V_cGMPmax * (B5sGC * NO + pow(NO, 2)) / (A0sGC + A1sGC * NO + pow(NO, 2));
    if ((V_cGMPbar - V_cGMP) >= 0) {
       tausGC = 1 / (k3sGC * NO + kDact_deactsGC); //kDact_deactsGC different from original kDsGC to uncouple Km from time constants
    } else {
       tausGC = 1 / (kDact_deactsGC + k_2sGC);
    }

    V_cGMPprime = (V_cGMPbar - V_cGMP) / tausGC;
    cGMPprime = V_cGMP - kpde_cGMP * cGMP * cGMP / (1e-3 + cGMP);

    //Norepinephrine receptor
    RS_Gprime = (k_rG * ksi_G * R_TG - (k_rG + k_pG * NE / (K_1G + NE)) * RS_G - k_rG * RS_PG);
    RS_PGprime = NE * (k_pG * RS_G / (K_1G + NE) - k_eG * RS_PG / (K_2G + NE));
    rho_rG = NE * RS_G / (ksi_G * R_TG * (K_1G + NE));
    Gprime = k_aG * (delta_G + rho_rG)*(G_TG - G) - k_dG*G;
    r_hG = alpha_G * Ca_i / (K_cG + Ca_i) * G;
    IP3prime = r_hG / gamma_G * PIP2 - k_degG*IP3;
    PIP2prime = -(r_hG + r_rG) * PIP2 - r_rG * gamma_G * IP3 + r_rG*PIP2_T;
    DAGprime = r_hG / gamma_G * PIP2 - k_degG*DAG;

    //Reversal potentials
    E_Ca = RT_F / z_Ca * log(Ca_o / Ca_i);
    E_Na = RT_F * log(Na_o / Na_i);
    E_K = RT_F * log(K_o / K_i);
    E_Cl = RT_F / z_Cl * log(Cl_o / Cl_i);

    //Voltage dependent calcium current I_CaL
    tau_d_L = 2.5 * exp(-pow((V_m + 40) / 30, 2)) + 1.15;
    d_Lbar = 1.0 / (1 + exp(-(V_m) / 8.3));
    d_Lprime = (d_Lbar - d_L) / tau_d_L;

    f_Lbar = 1.0 / (1 + exp((V_m + 42.0) / 9.1));
    tau_f_L = 65 * exp(-pow((V_m + 35) / 25, 2)) + 45;
    f_Lprime = (f_Lbar - f_L) / tau_f_L;

    if (V_m == 0) {
       I_CaL = d_L * f_L * P_CaL * A_m * 1e6 * z_Ca * F * (Ca_i - Ca_o); //[pA]
    } else {
       I_CaL = d_L * f_L * P_CaL * A_m * 1e6 * V_m * pow(z_Ca * F, 2) / (R * temp)*(Ca_o - Ca_i * exp(V_m * z_Ca / (RT_F))) / (1 - exp(V_m * z_Ca / (RT_F))); //[pA]
    }

    //Delayed rectifier current I_K
    p_Kbar = 1 / (1 + exp(-(V_m - V_1_2) / k));
    tau_p_K = 61.49 * exp(-0.0268 * V_m);
    p_Kprime = (p_Kbar - p_K) / tau_p_K;

    q_bar = 1.0 / (1 + exp((V_m + 40) / 14));
    q_1prime = (q_bar - q_1) / 371;
    q_2prime = (q_bar - q_2) / 2884;

    I_K = 1 * g_K * p_K * (0.45 * q_1 + 0.55 * q_2) * (V_m - E_K);

    //Alpha-adrenoceptor-activated nonselective cation channel NSC
    Po_NSC = Pmin_NSC + (1 - Pmin_NSC) / (1 + exp(-(V_m - 47.12) / 24.24));
    if (V_m == 0) {
       INa_NSC = 1 * (DAG / (DAG + K_NSC) + d_NSCmin) * Po_NSC * PNa_NSC * A_m * 1e6 * F * (Na_i - Na_o);
       ICa_NSC = 1 * (0 * DAG / (DAG + K_NSC) + d_NSCmin) * Po_NSC * PCa_NSC * A_m * 1e6 * z_Ca * F * (Ca_i - Ca_o);
       IK_NSC = 1 * (DAG / (DAG + K_NSC) + d_NSCmin) * Po_NSC * PK_NSC * A_m * 1e6 * F * (K_i - K_o);
    } else {
       INa_NSC = 1 * (DAG / (DAG + K_NSC) + d_NSCmin) * Po_NSC * PNa_NSC * A_m * 1e6 * V_m * pow(F, 2) / (R * temp)*(Na_o - Na_i * exp(V_m / RT_F)) / (1 - exp(V_m / RT_F));
       ICa_NSC = 1 * (0 * DAG / (DAG + K_NSC) + d_NSCmin) * Po_NSC * PCa_NSC * A_m * 1e6 * V_m * pow(z_Ca * F, 2) / (R * temp)*(Ca_o - Ca_i * exp(V_m * z_Ca / RT_F)) / (1 - exp(V_m * z_Ca / RT_F));
       IK_NSC = 1 * (DAG / (DAG + K_NSC) + d_NSCmin) * Po_NSC * PK_NSC * A_m * 1e6 * V_m * pow(F, 2) / (R * temp)*(K_o - K_i * exp(V_m / RT_F)) / (1 - exp(V_m / RT_F));
    }
    I_NSC = ICa_NSC + INa_NSC + IK_NSC;

    //KATP current I_KATP
    I_KATP = g_KATP * (V_m - E_K);

    //Inward rectifier current I_K_i
    g_maxK_i = G_K_i * pow(K_o, n_K_i);
    I_K_i = g_maxK_i * (V_m - E_K) / (1 + exp((V_m - E_K) / 28.89));

    //Calcium-activated potassium current I_KCa
    if (V_m == 0) {
       i1_KCa = 1e6 * P_BKCa * F * (K_i - K_o); //[pA]
    } else {
       i1_KCa = 1e6 * P_BKCa * V_m * F / RT_F * (K_o - K_i * exp(V_m / RT_F)) / (1 - exp(V_m / RT_F)); //[pA]
    } //Mistry and Garland 1998

    R_NO = NO / (NO + 200e-6);
    R_cGMP = pow(cGMP, 2) / (pow(cGMP, 2) + pow(1.5e-3, 2));
    V_1_2_KCa = -41.7 * log10(Ca_i) - 128.2 - dV_1_2_KCa_NO * R_NO - dV_1_2_KCa_cGMP*R_cGMP;
    p_obar = 1 / (1 + exp(-(V_m - V_1_2_KCa) / 18.25));
    p_fprime = (p_obar - p_f) / tau_pf;
    p_sprime = (p_obar - p_s) / tau_ps;

    P_KCa = 0.17 * p_f + 0.83 * p_s;

    I_KCa = A_m * N_BKCa * i1_KCa * P_KCa;

    //Store operated non-selective cation channel
    P_SOCbar = 1 / (1 + pow(Ca_u / K_SOC, H_SOC));
    P_SOCprime = (P_SOCbar - P_SOC) / tau_SOC;

    I_SOCCa = 1 * P_SOC * g_SOCCa * (V_m - E_Ca);
    I_SOCNa = 1 * P_SOC * g_SOCNa * (V_m - E_Na);
    I_SOC = I_SOCCa + I_SOCNa;

    //Chloride currents
    alpha_Cl = pow(cGMP, 3.3) / (pow(cGMP, 3.3) + pow(6.4e-3, 3.3));
    P_Cl = pow(Ca_i, 2) / (pow(Ca_i, 2) + pow(365e-6, 2)) * 0.0132 + pow(Ca_i, 2) / (pow(Ca_i, 2) + pow(400e-6 * (1 - alpha_Cl * 0.9), 2)) * alpha_Cl;
    I_Cl = P_Cl * g_Cl * C_m * (V_m - E_Cl);

    //IP3 receptor current
    h_IP3prime = k_onIP3 * (K_inhIP3 - (Ca_i + K_inhIP3) * h_IP3);
    I_IP3 = I_IP3bar * pow(IP3 / (IP3 + K_IP3) * Ca_i / (Ca_i + K_actIP3) * h_IP3, 3)*(Ca_u - Ca_i) * z_Ca * F*vol_Ca;

    //Calcium-induced calcium release
    R_00 = 1 - R_01 - R_10 - R_11;
    R_10prime = Kr1 * pow(Ca_i, 2) * R_00 - (K_r1 + Kr2 * Ca_i) * R_10 + K_r2*R_11;
    R_11prime = Kr2 * Ca_i * R_10 - (K_r1 + K_r2) * R_11 + Kr1 * pow(Ca_i, 2) * R_01;
    R_01prime = Kr2 * Ca_i * R_00 + K_r1 * R_11 - (K_r2 + Kr1 * pow(Ca_i, 2)) * R_01;
    I_up = I_upbar * Ca_i / (Ca_i + K_mup);
    I_tr = (Ca_u - Ca_r) * (2 * F * vol_u) / tau_tr;

    I_rel = (pow(R_10, 2) + R_leak) * (Ca_r - Ca_i) * (2 * F * vol_r) / tau_rel;
    Ca_uprime = (I_up - I_tr - I_IP3) / (2 * F * vol_u);
    Ca_rprime = (I_tr - I_rel) / (2 * F * vol_r) / (1 + CSQNbar * K_CSQN / pow((K_CSQN + Ca_r), 2));

    //Ca buffering and cytosolic material balance
    S_CM = S_CMbar * K_d / (K_d + Ca_i);
    CaCM = S_CMbar - S_CM;
    I_PMCA = I_PMCAbar * Ca_i / (Ca_i + K_mPMCA);

    //NaK pump
    I_NaK = pow(Q_10_NaK, ((temp - 309.15) / 10)) * C_m * I_NaKbar * ((pow(K_o, 1.1)) / (pow(K_o, 1.1) + (pow(K_mK, 1.1)))
           *(pow(Na_i, 1.7)) / ((pow(Na_i, 1.7))+(pow(K_mNa, 1.7)))) *(V_m + 150) / (V_m + 200);

    Fi_F = exp(gamma2 * V_m * F / (R * temp));
    Fi_R = exp((gamma2 - 1) * V_m * F / (R * temp));
    I_NCX = 1 * (1 + 0.55 * cGMP / (cGMP + (45e-3))) * g_NCX * (pow(Na_i, 3) * Ca_o * Fi_F - pow(Na_o, 3) * Ca_i * Fi_R) / (1 + d_NCX * (pow(Na_o, 3) * Ca_i + pow(Na_i, 3) * Ca_o));

    I_NaKCl_Cl = (1 + 7 / 2 * cGMP / (cGMP + 6.4e-3))*(-A_m * L_cotr * R * temp * z_Cl * F * log(Na_o / Na_i * K_o / K_i * pow(Cl_o / Cl_i, 2)));

    I_Catotm = I_SOCCa + I_CaL - 2 * I_NCX + I_PMCA + ICa_NSC + I_MCa;
    Ca_iprime = -(I_Catotm + I_up - I_rel - I_IP3) / (2 * F * vol_Ca) / (1 + S_CMbar * K_d / (pow(K_d + Ca_i, 2)) + B_Fbar * K_dB / (pow(K_dB + Ca_i, 2)));

    I_Natotm = -0.5 * I_NaKCl_Cl + I_SOCNa + 3 * I_NaK + 3 * I_NCX + INa_NSC + I_MNa;
    Na_iprime = -(I_Natotm) / (F * vol_i);

    I_Ktotm = -0.5 * I_NaKCl_Cl + I_K + I_KCa + I_K_i + IK_NSC + I_KATP - 2 * I_NaK + I_MK;
    K_iprime = -(I_Ktotm) / (F * vol_i);

    I_Cltotm = I_NaKCl_Cl + I_Cl;
    Cl_iprime = -(I_Cltotm) / (z_Cl * F * vol_i);

    //Transmembrane potential
    V_mprime = -1 / C_m * (-I_stim + I_Cl + I_SOC + I_CaL + I_K + I_KCa + I_K_i + I_M + I_NCX + I_NaK + I_PMCA + I_NSC + I_KATP);

    //YPRIME = zeros(1, N);
    YPRIME[0] = V_mprime;
    YPRIME[1] = d_Lprime;
    YPRIME[2] = f_Lprime;
    YPRIME[3] = p_Kprime;
    YPRIME[4] = q_1prime;
    YPRIME[5] = p_fprime;
    YPRIME[6] = p_sprime;
    YPRIME[7] = R_01prime;
    YPRIME[8] = R_10prime;
    YPRIME[9] = R_11prime;
    YPRIME[10] = Ca_uprime;
    YPRIME[11] = Ca_rprime;
    YPRIME[12] = Ca_iprime;
    YPRIME[13] = Na_iprime;
    YPRIME[14] = K_iprime;
    YPRIME[15] = q_2prime;
    YPRIME[16] = P_SOCprime;
    YPRIME[17] = Cl_iprime;
    YPRIME[18] = h_IP3prime;
    YPRIME[19] = RS_Gprime;
    YPRIME[20] = RS_PGprime;
    YPRIME[21] = Gprime;
    YPRIME[22] = IP3prime;
    YPRIME[23] = PIP2prime;
    YPRIME[24] = DAGprime;
    YPRIME[25] = V_cGMPprime;
    YPRIME[26] = cGMPprime;

    //Non state variables
    Y1[0] = I_CaL;
    Y1[1] = I_K;
    Y1[2] = I_K_i;
    Y1[3] = I_NSC;
    Y1[4] = I_KCa;
    Y1[5] = I_up;
    Y1[6] = I_rel;
    Y1[7] = I_PMCA;
    Y1[8] = I_NCX;
    Y1[9] = I_NaK;
    Y1[10] = INa_NSC;
    Y1[11] = ICa_NSC;
    Y1[12] = IK_NSC;
    Y1[13] = I_SOCCa;
    Y1[14] = I_SOCNa;
    Y1[15] = I_Cl;
    Y1[16] = I_NaKCl_Cl;
    Y1[17] = I_IP3;
    Y1[18] = I_tr;
    Y1[19] = NE;
    Y1[20] = I_KATP;
    Y1[21] = I_stim;
    Y1[22] = V_cGMPbar;
    Y1[23] = NO;
    Y1[29] = I_Catotm;
    Y1[30] = I_Natotm;
    Y1[31] = I_Cltotm;
    Y1[32] = I_Ktotm;
}

Answer 1

Windows and Linux set different floating point defaults. <fenv.h> may help you debug this by allowing the code to throw instead of silently nanning you. There are Windows-specific APIs that control the config of the FPU, and I think that by default they are not set for IEEE-754 compliance.

On Windows you'll want to call _controlfp if you want to operate in IEEE-754 standard mode.

More details on this page .

Answer 2

I suspect that your code is giving you an answer with your MS C compiler, but I am skeptical that it is running perfectly. NaNs (Not A Number)are the result of computing functions out of their ranges. Since you do not mention how you are solving your stiff system, I have no idea whether you are taking the log of a negative number, or some other squirrelly arithmetic, but I am sure that you are doing the same arithmetic in your windows code. Just because the code attempts to muddle through does not mean that it is running correctly.

I would suggest that you start looking at where the two programs start diverging, and you will probably find a questionable operation where the MS compiler returning 0 or Inf in a case where g++ is returning NaN.

Answer 3

EDIT: For sure these four lines are going off the end of the Y1 array and apparently in g++ they're walking on top of YPRIME. When I changed the declaration of Y1 to Y1[33] to make room I got -59.1481 as the result with g++.

Y1[29] = I_Catotm;
Y1[30] = I_Natotm;
Y1[31] = I_Cltotm;
Y1[32] = I_Ktotm;

Original answer:

If you have the ability to run both versions in a debugger, your best bet is to run both versions up to the point where they still agree (59.3993) and then single-step both debuggers side-by-side, watching each statement's output to see where they diverge. Once you see where the results are different it should be a lot clearer what the code is doing differently.

I once used this approach to find a bug in a major refactoring/optimization of an internal math library.

Answer 4

The solution to your problem is not easy. The variable you are recording ( Y[0] ) evolves at line 425: Y[i] = Y[i] + TSTEP * YPRIME[i] . The derivation YPRIME[i] is calculated in FCN() . Since we are looking only at Y[0] , we are interested in the value of V_mprime which is calculated in line 636 as -1 / C_m * (-I_stim + I_Cl + I_SOC + I_CaL + I_K + I_KCa + I_K_i + I_M + I_NCX + I_NaK + I_PMCA + I_NSC + I_KATP) .

Now, the divergence from MSVC may start in any of the variables in the expression above :-) I suggest you put a debugging output statement just above the line 636, and print all of the involved variables. Then run the program at both platforms and report the two outputs so we can focus on the differences.