diff --git a/common/lib/share/mccode-complex-lib.h b/common/lib/share/mccode-complex-lib.h index c99e3e9b7b..94ece28389 100644 --- a/common/lib/share/mccode-complex-lib.h +++ b/common/lib/share/mccode-complex-lib.h @@ -30,25 +30,25 @@ // Note: leave two spaces between double and cdouble because double->float // in generate._convert_type is not clever enough. typedef complex double cdouble; -inline cdouble cplx(const double x, const double y) { return x + y*I; } -inline cdouble cneg(const cdouble a) { return -a; } -inline cdouble radd(const double a, const cdouble b) { return a+b; } -inline cdouble rmul(const double a, const cdouble b) { return a*b; } -inline cdouble cadd(const cdouble a, const cdouble b) { return a+b; } -inline cdouble csub(const cdouble a, const cdouble b) { return a-b; } -inline cdouble cmul(const cdouble a, const cdouble b) { return a*b; } -inline cdouble cdiv(const cdouble a, const cdouble b) { return a/b; } +static inline cdouble cplx(const double x, const double y) { return x + y*I; } +static inline cdouble cneg(const cdouble a) { return -a; } +static inline cdouble radd(const double a, const cdouble b) { return a+b; } +static inline cdouble rmul(const double a, const cdouble b) { return a*b; } +static inline cdouble cadd(const cdouble a, const cdouble b) { return a+b; } +static inline cdouble csub(const cdouble a, const cdouble b) { return a-b; } +static inline cdouble cmul(const cdouble a, const cdouble b) { return a*b; } +static inline cdouble cdiv(const cdouble a, const cdouble b) { return a/b; } #else // MSVC typedef _Dcomplex cdouble; -inline cdouble cplx(const double x, const double y) { cdouble z = _Cbuild(x,y); return z; } -inline cdouble cneg(const cdouble a) { return cplx( -creal(a) , -cimag(a) ); } -inline cdouble radd(const double a, const cdouble b) { return cplx(a+creal(b),cimag(b)); } -inline cdouble rmul(const double a, const cdouble b) { return cplx(a*creal(b),a*cimag(b)); } -inline cdouble cadd(const cdouble a, const cdouble b) { return cplx(creal(a)+creal(b),cimag(a)+cimag(b)); } -inline cdouble csub(const cdouble a, const cdouble b) { return cplx(creal(a)-creal(b),cimag(a)-cimag(b)); } -inline cdouble cmul(const cdouble a, const cdouble b) { return cplx(creal(a)*creal(b) - cimag(a)*cimag(b), cimag(a)*creal(b) + creal(a)*cimag(b)); } -inline cdouble cdiv(const cdouble a, const cdouble b) { return cplx((creal(a)*creal(b)+cimag(a)*cimag(b))/(creal(b)*creal(b)+cimag(b)*cimag(b)),(cimag(a)*creal(b)-creal(a)*cimag(b))/(creal(b)*creal(b)+cimag(b)*cimag(b))); } +static inline cdouble cplx(const double x, const double y) { cdouble z = _Cbuild(x,y); return z; } +static inline cdouble cneg(const cdouble a) { return cplx( -creal(a) , -cimag(a) ); } +static inline cdouble radd(const double a, const cdouble b) { return cplx(a+creal(b),cimag(b)); } +static inline cdouble rmul(const double a, const cdouble b) { return cplx(a*creal(b),a*cimag(b)); } +static inline cdouble cadd(const cdouble a, const cdouble b) { return cplx(creal(a)+creal(b),cimag(a)+cimag(b)); } +static inline cdouble csub(const cdouble a, const cdouble b) { return cplx(creal(a)-creal(b),cimag(a)-cimag(b)); } +static inline cdouble cmul(const cdouble a, const cdouble b) { return cplx(creal(a)*creal(b) - cimag(a)*cimag(b), cimag(a)*creal(b) + creal(a)*cimag(b)); } +static inline cdouble cdiv(const cdouble a, const cdouble b) { return cplx((creal(a)*creal(b)+cimag(a)*cimag(b))/(creal(b)*creal(b)+cimag(b)*cimag(b)),(cimag(a)*creal(b)-creal(a)*cimag(b))/(creal(b)*creal(b)+cimag(b)*cimag(b))); } #endif #endif diff --git a/mcstas-comps/contrib/Phonon_BvK_PG.comp b/mcstas-comps/contrib/Phonon_BvK_PG.comp index fcb9a2bfdc..c0d88d3963 100644 --- a/mcstas-comps/contrib/Phonon_BvK_PG.comp +++ b/mcstas-comps/contrib/Phonon_BvK_PG.comp @@ -12,11 +12,11 @@ * * A sample for phonon scattering based on cross section expressions from the Boothroyd textbook * based upon the algorithm from component Phonon_simple (with expressions from Squires, Ch. 3) -* Using the Born-von Karman force constnts and the normal mode formalism +* Using the Born-von Karman force constants and the normal mode formalism * * * %D -* Single-cylinder or slap shaped shape. +* Single cylinder or slab shaped sample. * Absorption included. * No multiple scattering. * No incoherent scattering emitted, but incoherent is present in attenuation @@ -36,8 +36,8 @@ * INPUT PARAMETERS: * radius: [m] Outer radius of sample in (x,z) plane * yheight: [m] Height of sample in y direction -* xwidth: [m] Horiz. dimension of sample if slap -* zdepth: [m] Depth of sample if slap +* xwidth: [m] Horiz. dimension of sample if slab +* zdepth: [m] Depth of sample if slab * * sigma_abs: [barns] Absorption cross section at 2200 m/s per atom * sigma_inc: [barns] Incoherent scattering cross section per atom @@ -52,7 +52,7 @@ * target_y: [m] position of target to focus at. Vertical coordinate * target_z: [m] position of target to focus at. Straight ahead. * target_index: [1] relative index of component to focus at, e.g. next is +1 -* mode_input: [1] order of the phonon mode to be scattered (unphysical). In the range 0 through 11 . If mode_input == 12, then the mode is choosen at random +* mode_input: [1] order of the phonon mode to be scattered (unphysical). In the range 0 through 11 . If mode_input == 12, then the mode is chosen at random * verbose_input: [1] toggles verbose mode (verbose>1) or eigenvector mode (verbose==1) * dispersion: [1] 0: do nothing special; 1: calculate and output the dispersion; 2: calculate and output the dispersion while varying v_f * e_steps_low: [1] Amount of possible intersections beneath the elastic line @@ -64,10 +64,10 @@ * OUTPUT PARAMETERS: * V_rho: [AA^-3] Unit cell density * V_my_s: [m^-1] Attenuation factor due to incoherent scattering -* V_my_a_v: [m^-1] Attenuation factor due to absorbtion +* V_my_a_v: [m^-1] Attenuation factor due to absorption * * %L -* The test/example instrument Test_Phonon.instr. +* The test/example instrument Test_Phonon_BvK_PG.instr. * * %E ******************************************************************************/ @@ -75,8 +75,6 @@ DEFINE COMPONENT Phonon_BvK_PG SETTING PARAMETERS (hh=0,kk=0,ll=0,radius=0, xwidth=0, yheight=0, zdepth=0, sigma_abs=0,sigma_inc=0,DW=1,T, focus_r=0,focus_xw=0,focus_yh=0,focus_aw=0,focus_ah=0,target_x=0, target_y=0, target_z=0, int target_index=0, int mode_input, int e_steps_low, int e_steps_high, int verbose_input=0, int dispersion=0) -NOACC -// The component is currently "NOACC" only. SHARE %{ @@ -84,24 +82,22 @@ SHARE // James Avery eigenvalue solver %include "eigen" - #ifndef PHONON_SIMPLE - #define PHONON_SIMPLE $Revision$ // KL comment 040125: what is this used for? - #pragma acc routine + #ifndef PHONON_BVK_PG + #define PHONON_BVK_PG 0.9 // ---------------- THINGS THAT COULD BE GENERAL FOR McStas -------------- #define T2E (1/11.605) // Kelvin to meV converter - #define pi 3.14159265358979323846264338327950288419716939937510L - #define sqrt3 1.73205080756887729352744634150587236694280525381038L + #define sqrt3 1.73205080756887729352744634150587236694280525381038 /* double, not long double: no x87 on the GPU */ #define THz2meV 4.13566 // THz to meV converter #define Da2kg 1.66054e-27 //Dalton to kg converter #define dyn2N 1e-3 // dyn/cm to N/m converter - #define hbar 1.0546E-34 - #define mn (1.0087*Da2kg) + #define BVK_MEV2J 1.6022E-22 // meV to J converter + #pragma acc routine seq double - nbose (double omega, double T) /* Give it another name ?? */ + bvk_nbose (double omega, double T) /* Give it another name ?? */ { double nb; nb = (omega > 0) ? 1 + 1 / (exp (omega / (T * T2E)) - 1) : 1 / (exp (-omega / (T * T2E)) - 1); @@ -109,34 +105,36 @@ SHARE } #undef T2E - /* Routine types from Numerical Recipies book */ + /* Routine types from Numerical Recipes book */ #define UNUSED (-1.11e30) #define MAXRIDD 60 void - fatalerror_cpu (char* s) { + bvk_fatalerror_cpu (char* s) { fprintf (stderr, "%s \n", s); - exit (1); + exit (-1); } + #pragma acc routine seq void nrerror (const char* error_text) { - printf ("Numerical Recipes run-time error ...\n"); - printf ("%s\n", error_text); - printf ("... now exiting to system.\n"); - exit (1); + #ifndef OPENACC + fprintf (stderr, "Numerical Recipes run-time error ...\n"); + fprintf (stderr, "%s\n", error_text); + fprintf (stderr, "... now exiting to system.\n"); + exit (-1); + #endif } - #pragma acc routine + #pragma acc routine seq void - fatalerror (char* s) { + bvk_fatalerror (char* s) { #ifndef OPENACC - fatalerror_cpu (s); + bvk_fatalerror_cpu (s); #endif } - #pragma acc routine - + #pragma acc routine seq void bubblesort (int size, double* inputarray, int* index) // this function modifies the inputarray by bubble sorting, and also modifies the index array as the @@ -161,22 +159,14 @@ SHARE // ---------------- THINGS THAT ARE SPECIFIC FOR THE PG COMPONENT ------------ - #define SMALL_NUMBER 1E-6 - #define V_HIGH 8000 // Highest velocity to search, corresponding to 320 meV, far above the to of the PG phonon band (202 meV along main axes) - #define MATRIX_TEST 0 // Change to 1 to write matrices to disk - #define TIME_EIGENSYSTEMS 1 - - #define DIM 12 // dimensionality of the phonon eigenvectors (4 atoms in the unit cell times 3 dimensions) - #define DV 0.001 // Velocity change used for numerical derivative [m/s] CHECK if this should be smaller - #include #include #include - #include - extern FILE *matrix_test_file, *parms_test_file; - extern int matrix_test_count; - int mode, verbose, shape; + #define V_HIGH 8000 // Highest velocity to search, corresponding to 320 meV, far above the top of the PG phonon band (202 meV along main axes) + + #define DIM 12 // dimensionality of the phonon eigenvectors (4 atoms in the unit cell times 3 dimensions) + #define BVK_DV 0.001 // Velocity change used for numerical derivative [m/s] CHECK if this should be smaller #define a_latt 2.461 //PG lattice constant in AA; Nicklow1972 gives 2.45 (WHICH PAGE?) #define c_latt 6.708 // PG lattice constant in AA; Nicklow1972 gives 6.70 @@ -200,9 +190,6 @@ SHARE double Delta[4 * 3] = { 0, 0, 0, a_latt / (2 * sqrt3), a_latt / 2, 0, 0, 0, c_latt / 2, -a_latt / (2 * sqrt3), a_latt / 2, c_latt / 2 }; // Lattice vectors from paper fig. 7 - double avec[3] = { a_latt * sqrt3 / 2, a_latt * 0.5, 0 }; // THIS IS NEVER USED ??!! - double bvec[3] = { a_latt * (-sqrt3 / 2), a_latt * 0.5, 0 }; - double cvec[3] = { 0, 0, c_latt }; // reciprocal lattice vectors double astar = 4 * PI / (sqrt3 * a_latt); // length of reciprocal lattice vector a* @@ -211,34 +198,20 @@ SHARE // Rotation matrices // m(12*i+j) = M(i,j) double Rot120[3][3] = { { -0.5, sqrt3 / 2, 0 }, { -sqrt3 / 2, -0.5, 0 }, { 0, 0, 1 } }; - double Rot120_flat[9] = { -0.5, sqrt3 / 2, 0, -sqrt3 / 2, -0.5, 0, 0, 0, 1 }; double Rot60[3][3] = { { 0.5, sqrt3 / 2, 0 }, { -sqrt3 / 2, 0.5, 0 }, { 0, 0, 1 } }; double Rot120rev[3][3] = { { -0.5, -sqrt3 / 2, 0 }, { sqrt3 / 2, -0.5, 0 }, { 0, 0, 1 } }; // rotate -120 double Rot60rev[3][3] = { { 0.5, -sqrt3 / 2, 0 }, { sqrt3 / 2, 0.5, 0 }, { 0, 0, 1 } }; // rotate -60 double Rot180[3][3] = { { -1, 0, 0 }, { 0, -1, 0 }, { 0, 0, 1 } }; // rotate 180 or -180 - // double Rot5[3][3] = {{0.5 , sqrt3/2 , 0} , {-sqrt3/2 , 0.5 , 0} , {0 , 0 , 1}}; //rotate 60 - // double Rot5rev[3][3] = {{0.5 , -sqrt3/2 , 0} , {sqrt3/2 , 0.5 , 0} , {0 , 0 , 1}}; //rotate -60 // 1st neighbour - - // double r_j1[3][3] = {{-a_latt/sqrt3 , 0 , 0} , {a_latt*0.5/sqrt3 , -a_latt*0.5 , 0} , {a_latt*0.5/sqrt3 , a_latt*0.5 , 0}}; // This holds from atoms A and - // D double r_j2[3][3] = {{a_latt/sqrt3 , 0 , 0} , {-a_latt*0.5/sqrt3 , a_latt*0.5 , 0} , {-a_latt*0.5/sqrt3 , -a_latt*0.5 , 0}}; // This holds from atoms B - // and C - - double Phi_nn1[3][3] = { { K_l1 * dyn2N, 0, 0 }, { 0, K_t1* dyn2N, 0 }, { 0, 0, K_t1* dyn2N } }; - double Phi1[3][3] = { { K_l1 * dyn2N, 0, 0 }, { 0, K_t1* dyn2N, 0 }, { 0, 0, K_t1* dyn2N } }; // Phi1 = Phi_nn1 + double Phi_nn1[3][3] = { { K_l1 * dyn2N, 0, 0 }, { 0, K_t1 * dyn2N, 0 }, { 0, 0, K_t1 * dyn2N } }; + double Phi1[3][3] = { { K_l1 * dyn2N, 0, 0 }, { 0, K_t1 * dyn2N, 0 }, { 0, 0, K_t1 * dyn2N } }; // Phi1 = Phi_nn1 double Phi2[3][3]; double Phi3[3][3]; // 2nd neighbour - // double r_j11[9][3] = {{a_latt*(-1/sqrt3) , 0 , 0} , {a_latt*0.5/sqrt3 , a_latt*(-0.5) , 0} , {a_latt*0.5/sqrt3 , a_latt*0.5 , 0} , {0 , a_latt , 0} , - // {-a_latt*sqrt3/2 , a_latt/2 , 0} , {-a_latt*sqrt3/2 , -a_latt/2 , 0} , {0 , -a_latt , 0} , {a_latt*sqrt3/2 , -a_latt/2 , 0} , {a_latt*sqrt3/2 , a_latt/2 , - // 0}};//r_j1 + r_add double r_j21[9][3] = {{-a_latt*(-1/sqrt3) , -a_latt*0 , 0} , {-a_latt*0.5/sqrt3 , -a_latt*(-0.5) , 0} , {-a_latt*0.5/sqrt3 , - // -a_latt*0.5 , 0} , {0 , a_latt , 0} , {-a_latt*sqrt3/2 , a_latt/2 , 0} , {-a_latt*sqrt3/2 , -a_latt/2 , 0} , {0 , -a_latt , 0} , {a_latt*sqrt3/2 , - // -a_latt/2 , 0} , {a_latt*sqrt3/2 , a_latt/2 , 0}};//r_j2 + r_add - - double Phi_nn2[3][3] = { { K_t2 * dyn2N, 0, 0 }, { 0, K_l2* dyn2N, 0 }, { 0, 0, K_t2* dyn2N } }; - double Phi4[3][3] = { { K_t2 * dyn2N, 0, 0 }, { 0, K_l2* dyn2N, 0 }, { 0, 0, K_t2* dyn2N } }; + double Phi_nn2[3][3] = { { K_t2 * dyn2N, 0, 0 }, { 0, K_l2 * dyn2N, 0 }, { 0, 0, K_t2 * dyn2N } }; + double Phi4[3][3] = { { K_t2 * dyn2N, 0, 0 }, { 0, K_l2 * dyn2N, 0 }, { 0, 0, K_t2 * dyn2N } }; double Phi5[3][3]; double Phi6[3][3]; double Phi7[3][3]; @@ -246,9 +219,6 @@ SHARE double Phi9[3][3]; // 3rd neighbour - // double r_j12[12][3] = {{a_latt*(-1/sqrt3) , 0 , 0} , {a_latt*0.5/sqrt3 , a_latt*(-0.5) , 0} , {a_latt*0.5/sqrt3 , a_latt*0.5 , 0} , {0 , a_latt , 0} , - // {-a_latt*sqrt3/2 , a_latt/2 , 0} , {-a_latt*sqrt3/2 , -a_latt/2 , 0} , {0 , -a_latt , 0} , {a_latt*sqrt3/2 , -a_latt/2 , 0} , {a_latt*sqrt3/2 , a_latt/2 , - // 0} , {2*a_latt/sqrt3 , 0 , 0} , {-a_latt/sqrt3 , a_latt , 0} , {-a_latt/sqrt3 , -a_latt , 0}};//r_j11 + r_add2 double r_j2[12][3] = { { a_latt / sqrt3, 0, 0 }, { -a_latt * 0.5 / sqrt3, a_latt * 0.5, 0 }, { -a_latt * 0.5 / sqrt3, -a_latt * 0.5, 0 }, @@ -262,14 +232,9 @@ SHARE { a_latt / sqrt3, -a_latt, 0 }, { a_latt / sqrt3, a_latt, 0 } }; // r_j21 - r_add2 - double Phi_nn3[3][3] = { { K_l3 * dyn2N, 0, 0 }, { 0, K_t3* dyn2N, 0 }, { 0, 0, K_t3* dyn2N } }; + double Phi_nn3[3][3] = { { K_l3 * dyn2N, 0, 0 }, { 0, K_t3 * dyn2N, 0 }, { 0, 0, K_t3 * dyn2N } }; - // c2 = 2/sqrt(6); - // c1 = 1/sqrt(6); - // c0 = 1/sqrt(2); - // c3 = 1/sqrt(3); - - double Phi10[3][3] = { { K_l3 * dyn2N, 0, 0 }, { 0, K_t3* dyn2N, 0 }, { 0, 0, K_t3* dyn2N } }; // Phi_nn3 + double Phi10[3][3] = { { K_l3 * dyn2N, 0, 0 }, { 0, K_t3 * dyn2N, 0 }, { 0, 0, K_t3 * dyn2N } }; // Phi_nn3 double Phi11[3][3]; double Phi12[3][3]; @@ -289,22 +254,22 @@ SHARE { 0, 0, c_latt / 2 }, { 0, 0, -c_latt / 2 } }; // r_j12 + r_add3 Sublattice B and D do not have this coupling - double Phi_nn4[3][3] = { { K_t4 * dyn2N, 0, 0 }, { 0, K_t4* dyn2N, 0 }, { 0, 0, K_l4* dyn2N } }; + double Phi_nn4[3][3] = { { K_t4 * dyn2N, 0, 0 }, { 0, K_t4 * dyn2N, 0 }, { 0, 0, K_l4 * dyn2N } }; + + double Phi13[3][3] = { { K_t4 * dyn2N, 0, 0 }, { 0, K_t4 * dyn2N, 0 }, { 0, 0, K_l4 * dyn2N } }; // Phi_nn4; + double Phi14[3][3] = { { K_t4 * dyn2N, 0, 0 }, { 0, K_t4 * dyn2N, 0 }, { 0, 0, K_l4 * dyn2N } }; // Phi_nn4; - double Phi13[3][3] = { { K_t4 * dyn2N, 0, 0 }, { 0, K_t4* dyn2N, 0 }, { 0, 0, K_l4* dyn2N } }; // Phi_nn4; - double Phi14[3][3] = { { K_t4 * dyn2N, 0, 0 }, { 0, K_t4* dyn2N, 0 }, { 0, 0, K_l4* dyn2N } }; // Phi_nn4; + // Globals read in TRACE / bvk_omega_q; copied to the device at the end of INITIALIZE + #pragma acc declare create(M, Delta, astar, cstar, r_j13) + #pragma acc declare create(Phi1, Phi2, Phi3, Phi4, Phi5, Phi6, Phi7, Phi8, Phi9, Phi10, Phi11, Phi12, Phi13, Phi14) - // ----------------- END SPECIFIC FOR THE PG COPONENT -------------------------- + // ----------------- END SPECIFIC FOR THE PG COMPONENT -------------------------- /* TODO: Det er meget langsomt at bruge pointere. Erstat med flad memory. */ // m(12*i+j) = M(i,j) // 3*3 matrices multiplication void matrixmultiplication (double matrix1[3][3], double matrix2[3][3], double result[3][3]) { - // multiplies matrix1 by matrix2 and places the result in result - // double matrix1[3][3]={{*(array1), *(array1+1), *(array1+2)}, {*(array1+3), *(array1+4), *(array1+5)}, {*(array1+6), *(array1+7), *(array1+8)}}; - // double matrix2[3][3]={{*(array2), *(array2+1), *(array2+2)}, {*(array2+3), *(array2+4), *(array2+5)}, {*(array2+6), *(array2+7), *(array2+8)}}; - result[0][0] = matrix2[0][0] * matrix1[0][0] + matrix2[1][0] * matrix1[0][1] + matrix2[2][0] * matrix1[0][2]; result[0][1] = matrix2[0][1] * matrix1[0][0] + matrix2[1][1] * matrix1[0][1] + matrix2[2][1] * matrix1[0][2]; result[0][2] = matrix2[0][2] * matrix1[0][0] + matrix2[1][2] * matrix1[0][1] + matrix2[2][2] * matrix1[0][2]; @@ -314,77 +279,32 @@ SHARE result[2][0] = matrix2[0][0] * matrix1[2][0] + matrix2[1][0] * matrix1[2][1] + matrix2[2][0] * matrix1[2][2]; result[2][1] = matrix2[0][1] * matrix1[2][0] + matrix2[1][1] * matrix1[2][1] + matrix2[2][1] * matrix1[2][2]; result[2][2] = matrix2[0][2] * matrix1[2][0] + matrix2[1][2] * matrix1[2][1] + matrix2[2][2] * matrix1[2][2]; - return; } - // TODO: (James Avery) Benchmark against pointer versions: matrix3mupltiplication or matrixmultiplication - // m(12*i+j) = M(i,j) - void - matrix3x3_multiply3 (const double A[9], const double B[9], const double C[9], double result[9]) { - // R_{i,j} = \sum_{k,l=1}^3 A_{i,k}*B_{k,l}*C_{l,j} - for (int i = 0; i < 3; i++) - for (int j = 0; j < 3; j++) { - double ij_sum = 0; - for (int k = 0; k < 3; k++) - for (int l = 0; l < 3; l++) - ij_sum += A[i * 3 + k] * B[k * 3 + l] * C[l * 3 + j]; - result[i * 3 + j] = ij_sum; - } - } - /* TODO: (James Avery) It is slow to use points. Replace with flat memory. */ // m(12*i+j) = M(i,j) // Counter argument: (Kim Lefmann) pointers are now only used in the initialization. In the TRACE code, memory is flat void matrix3multiplication (double matrix1[3][3], double matrix2[3][3], double matrix3[3][3], double result[3][3]) { // Performs the multiplication (matrix1*matrix2)*matrix3 and placed the result in result - // #define TEST_MULTIPLY double temp[3][3]; matrixmultiplication (matrix1, matrix2, temp); matrixmultiplication (temp, matrix3, result); - #ifdef TEST_MULTIPLY - printf ("Matrix3multiply called with \n"); - printf ("matrix1 = ("); - for (int i = 0; i < 3; i++) { - printf ("( %g %g %g), ", matrix1[i][0], matrix1[i][1], matrix1[i][2]); - } - printf (")\n Matrix2 = ("); - for (int i = 0; i < 3; i++) { - printf ("( %g %g %g), ", matrix2[i][0], matrix2[i][1], matrix2[i][2]); - } - printf (")\n Matrix3 = ("); - for (int i = 0; i < 3; i++) { - printf ("( %g %g %g), ", matrix3[i][0], matrix3[i][1], matrix3[i][2]); - } - printf (")\n Temp = ("); - for (int i = 0; i < 3; i++) { - printf ("( %g %g %g), ", temp[i][0], temp[i][1], temp[i][2]); - } - printf (")\n Result = ("); - for (int i = 0; i < 3; i++) { - printf ("( %g %g %g), ", result[i][0], result[i][1], result[i][2]); - } - printf (")\n"); - #endif // TEST_MULTIPLY - return; } - /* TODO: (James Avery) Replace by standard complex.h operation */ + /* Returns e^{i q.r_j} using the wrapper-safe cplx() constructor instead of the I literal. */ + #pragma acc routine seq cdouble complexexp_qdotr (double* q, double* rj) { - // double qrjtest double qrj = *(q) * *(rj) + *(q + 1) * *(rj + 1) + *(q + 2) * *(rj + 2); - // printf(" q= (%g %g %g), r = (%g %g %g) qrj = %g \n",q[0],q[1],q[2],rj[0],rj[1],rj[2],qrj); - // return (cexp(qrj)); // KL comment 030125 WHY does this not run? return cplx (cos (qrj), sin (qrj)); } void initialize_omega_q () { // Make the matrix algebra that finalizes the force constant set-up - matrix3multiplication (Rot120, Phi_nn1, Rot120rev, Phi2); // Phi2=Rot120*Phi_nn1*Rot120^(-1) matrix3multiplication (Rot120rev, Phi_nn1, Rot120, Phi3); // Phi3=Rot120^{-1}*Phi_nn1*Rot120 matrix3multiplication (Rot60, Phi_nn2, Rot60rev, Phi5); // Phi5=Rot60*Phi_nn2*Rot60^{-1} @@ -394,19 +314,36 @@ SHARE matrix3multiplication (Rot60rev, Phi_nn2, Rot60, Phi9); // Phi9=Rot60^{-1}*Phi_nn2*Rot60; matrix3multiplication (Rot120, Phi_nn3, Rot120rev, Phi11); // Phi11=Rot120*Phi_nn3*Rot120^(-1); matrix3multiplication (Rot120rev, Phi_nn3, Rot120, Phi12); // Phi12=Rot120^{-1}*Phi_nn3*Rot120; - return; } //---------------------------- START CENTRAL FUNCTION OMEGA-Q ---------------------------- + #pragma acc routine seq double - omega_q (cdouble* parms, cdouble* eigenvectorgood) { + bvk_omega_q (cdouble* parms, cdouble* eigenvectorgood) { + + // _______________ DETERMINE Q ____________________________ + double vi, vf, vv_x, vv_y, vv_z, vi_x, vi_y, vi_z; - double q, qx, qy, qz, Jq, hw_phonon, hw_neutron; + double qx, qy, qz, hw_phonon, hw_neutron; double h, k, l, hk_inplane, hk_outofplane; int disp; + int mode = (int)creal (parms[13]); + int verbose = (int)creal (parms[14]); + #ifndef OPENACC + for (int ii = 0; ii < 8; ii++) + if (isnan (creal (parms[ii])) || isnan (cimag (parms[ii]))) { + fprintf (stderr, "omega_q: Array parms contains a nan at position %d.\n", ii); + fprintf (stderr, "parms = ["); + for (int i = 0; i < 8; i++) + fprintf (stderr, "%g + i%g%s", creal (parms[i]), cimag (parms[i]), i + 1 < 8 ? ", " : "]\n"); + exit (-1); + } + #endif + + /* Extract real-valued parameters from the cdouble parms array. */ vf = creal (parms[0]); vi = creal (parms[1]); vv_x = creal (parms[2]); @@ -415,7 +352,6 @@ SHARE vi_x = creal (parms[5]); vi_y = creal (parms[6]); vi_z = creal (parms[7]); - h = creal (parms[9]); k = creal (parms[10]); l = creal (parms[11]); @@ -425,266 +361,219 @@ SHARE qy = V2K * (vi_y - vf * vv_y); qz = V2K * (vi_z - vf * vv_z); - if (disp == 1) { + if (disp == 1) /* calculate dispersion directly */ + { printf ("Dispersion calculation: "); qx = h * astar; - qy = k * astar; + qy = k * astar; // correct only for qy=0 ! qz = l * cstar; } double qvec[3] = { qx, qy, qz }; - /* ===================== Phi_diag ===================== */ - - cdouble Phi_diag[9]; - - for (int i = 0; i < 3; i++) - for (int j = 0; j < 3; j++) { - cdouble sum = cplx (Phi1[i][j] + Phi2[i][j] + Phi3[i][j] + Phi10[i][j] + Phi11[i][j] + Phi12[i][j], 0.0); - - cdouble e; - - e = (*complexexp_qdotr) (&qvec[0], &r_j13[3][0]); - sum = cadd (sum, rmul (Phi4[i][j], csub (cplx (1.0, 0.0), e))); - - e = (*complexexp_qdotr) (&qvec[0], &r_j13[4][0]); - sum = cadd (sum, rmul (Phi5[i][j], csub (cplx (1.0, 0.0), e))); - - e = (*complexexp_qdotr) (&qvec[0], &r_j13[5][0]); - sum = cadd (sum, rmul (Phi6[i][j], csub (cplx (1.0, 0.0), e))); - - e = (*complexexp_qdotr) (&qvec[0], &r_j13[6][0]); - sum = cadd (sum, rmul (Phi7[i][j], csub (cplx (1.0, 0.0), e))); - - e = (*complexexp_qdotr) (&qvec[0], &r_j13[7][0]); - sum = cadd (sum, rmul (Phi8[i][j], csub (cplx (1.0, 0.0), e))); + if (verbose >= 4) + printf ("Enter omega_q; q = (%g, %g, %g) \n", qx, qy, qz); - e = (*complexexp_qdotr) (&qvec[0], &r_j13[8][0]); - sum = cadd (sum, rmul (Phi9[i][j], csub (cplx (1.0, 0.0), e))); + // ________________ END DETERMINE Q ________________- - Phi_diag[i * 3 + j] = sum; + cdouble Phi_diag[3 * 3]; + int i; + int j; + for (i = 0; i < 3; i++) { + for (j = 0; j < 3; j++) { + /* Each term: Phi_n[i][j] * (1 - e^{iq.r}) or just Phi_n[i][j] (purely real, for diagonal self-terms). + The real-valued self-terms are accumulated separately and added via radd() at the end. */ + double real_sum = Phi1[i][j] + Phi2[i][j] + Phi3[i][j] + Phi10[i][j] + Phi11[i][j] + Phi12[i][j]; + cdouble csum = rmul (Phi4[i][j], radd (1.0, cneg (complexexp_qdotr (&qvec[0], &r_j13[3][0])))); + csum = cadd (csum, rmul (Phi5[i][j], radd (1.0, cneg (complexexp_qdotr (&qvec[0], &r_j13[4][0]))))); + csum = cadd (csum, rmul (Phi6[i][j], radd (1.0, cneg (complexexp_qdotr (&qvec[0], &r_j13[5][0]))))); + csum = cadd (csum, rmul (Phi7[i][j], radd (1.0, cneg (complexexp_qdotr (&qvec[0], &r_j13[6][0]))))); + csum = cadd (csum, rmul (Phi8[i][j], radd (1.0, cneg (complexexp_qdotr (&qvec[0], &r_j13[7][0]))))); + csum = cadd (csum, rmul (Phi9[i][j], radd (1.0, cneg (complexexp_qdotr (&qvec[0], &r_j13[8][0]))))); + Phi_diag[i * 3 + j] = radd (real_sum, csum); } + } - /* ===================== Phi_diag1 ===================== */ - - cdouble Phi_diag1[9]; - - for (int i = 0; i < 3; i++) - for (int j = 0; j < 3; j++) + cdouble Phi_diag1[3 * 3]; + for (i = 0; i < 3; i++) { + for (j = 0; j < 3; j++) { Phi_diag1[i * 3 + j] = cplx (Phi13[i][j] + Phi14[i][j], 0.0); - - /* ===================== Phi_offdiag ===================== */ - - cdouble Phi_offdiag1[9]; - cdouble Phi_offdiag2[9]; - - for (int i = 0; i < 3; i++) - for (int j = 0; j < 3; j++) { - cdouble s = cplx (0.0, 0.0); - - s = cadd (s, cneg (rmul (Phi1[i][j], (*complexexp_qdotr) (&qvec[0], &r_j13[0][0])))); - - s = cadd (s, cneg (rmul (Phi2[i][j], (*complexexp_qdotr) (&qvec[0], &r_j13[1][0])))); - - s = cadd (s, cneg (rmul (Phi3[i][j], (*complexexp_qdotr) (&qvec[0], &r_j13[2][0])))); - - s = cadd (s, cneg (rmul (Phi10[i][j], (*complexexp_qdotr) (&qvec[0], &r_j13[9][0])))); - - s = cadd (s, cneg (rmul (Phi11[i][j], (*complexexp_qdotr) (&qvec[0], &r_j13[10][0])))); - - s = cadd (s, cneg (rmul (Phi12[i][j], (*complexexp_qdotr) (&qvec[0], &r_j13[11][0])))); - - Phi_offdiag1[i * 3 + j] = s; - Phi_offdiag2[i * 3 + j] = conj (s); } + } - /* ===================== Phi_6Dim ===================== */ - - cdouble Phi_6Dim[36]; + cdouble Phi_offdiag1[3 * 3]; + cdouble Phi_offdiag2[3 * 3]; + for (i = 0; i < 3; i++) { + for (j = 0; j < 3; j++) { + cdouble od = rmul (-Phi1[i][j], complexexp_qdotr (&qvec[0], &r_j13[0][0])); + od = cadd (od, rmul (-Phi2[i][j], complexexp_qdotr (&qvec[0], &r_j13[1][0]))); + od = cadd (od, rmul (-Phi3[i][j], complexexp_qdotr (&qvec[0], &r_j13[2][0]))); + od = cadd (od, rmul (-Phi10[i][j], complexexp_qdotr (&qvec[0], &r_j13[9][0]))); + od = cadd (od, rmul (-Phi11[i][j], complexexp_qdotr (&qvec[0], &r_j13[10][0]))); + od = cadd (od, rmul (-Phi12[i][j], complexexp_qdotr (&qvec[0], &r_j13[11][0]))); + Phi_offdiag1[i * 3 + j] = od; + Phi_offdiag2[i * 3 + j] = conj (od); + } + } - for (int i = 0; i < 3; i++) - for (int j = 0; j < 3; j++) { + cdouble Phi_6Dim[6 * 6]; + for (i = 0; i < 3; i++) { + for (j = 0; j < 3; j++) { Phi_6Dim[i * 6 + j] = cadd (Phi_diag[i * 3 + j], Phi_diag1[i * 3 + j]); Phi_6Dim[i * 6 + j + 3] = Phi_offdiag1[i * 3 + j]; Phi_6Dim[(i + 3) * 6 + j] = Phi_offdiag2[i * 3 + j]; - Phi_6Dim[(i + 3) * 6 + j + 3] = Phi_diag[i * 3 + j]; + Phi_6Dim[(i + 3) * 6 + j + 3] = Phi_diag[i * 3 + j]; // Phi_diag1 does not appear for the B,C sublattices } + } - /* ===================== Phi_AC ===================== */ - - cdouble Phi_AC[9]; - - for (int i = 0; i < 3; i++) - for (int j = 0; j < 3; j++) { - cdouble s = cplx (0.0, 0.0); - - s = cadd (s, cneg (rmul (Phi13[i][j], (*complexexp_qdotr) (&qvec[0], &r_j13[12][0])))); - - s = cadd (s, cneg (rmul (Phi14[i][j], (*complexexp_qdotr) (&qvec[0], &r_j13[13][0])))); - - Phi_AC[i * 3 + j] = s; + cdouble Phi_AC[3 * 3]; + for (i = 0; i < 3; i++) { + for (j = 0; j < 3; j++) { + Phi_AC[i * 3 + j] + = cadd (rmul (-Phi13[i][j], complexexp_qdotr (&qvec[0], &r_j13[12][0])), rmul (-Phi14[i][j], complexexp_qdotr (&qvec[0], &r_j13[13][0]))); } + } - /* ===================== Phi_6Dim_offdiag ===================== */ - - cdouble Phi_6Dim_offdiag[36]; - - for (int i = 0; i < 3; i++) - for (int j = 0; j < 3; j++) { + cdouble Phi_6Dim_offdiag[6 * 6]; + for (i = 0; i < 3; i++) { + for (j = 0; j < 3; j++) { Phi_6Dim_offdiag[i * 6 + j] = Phi_AC[i * 3 + j]; Phi_6Dim_offdiag[i * 6 + j + 3] = cplx (0.0, 0.0); Phi_6Dim_offdiag[(i + 3) * 6 + j] = cplx (0.0, 0.0); Phi_6Dim_offdiag[(i + 3) * 6 + j + 3] = cplx (0.0, 0.0); } + } - /* ===================== Phi_12Dim ===================== */ - - cdouble Phi_12Dim[144]; - - for (int i = 0; i < 6; i++) - for (int j = 0; j < 6; j++) { + cdouble Phi_12Dim[12 * 12]; + for (i = 0; i < 6; i++) { + for (j = 0; j < 6; j++) { Phi_12Dim[i * DIM + j] = Phi_6Dim[i * 6 + j]; Phi_12Dim[i * DIM + j + 6] = Phi_6Dim_offdiag[i * 6 + j]; Phi_12Dim[(i + 6) * DIM + j] = conj (Phi_6Dim_offdiag[i * 6 + j]); Phi_12Dim[(i + 6) * DIM + j + 6] = Phi_6Dim[i * 6 + j]; } - - /* ===================== Build Hermitian matrix ===================== */ + } double eigenvalue[DIM]; cdouble Q[DIM * DIM]; cdouble matrix[DIM * DIM]; int index[DIM]; - for (int i = 0; i < DIM; i++) - for (int j = i; j < DIM; j++) { + for (i = 0; i < DIM; i++) // transforms from 2D matrix to 1D pseudo-matrix + for (j = i; j < DIM; j++) { matrix[i * DIM + j] = Phi_12Dim[i * DIM + j]; - matrix[j * DIM + i] = conj (matrix[i * DIM + j]); - } + matrix[j * DIM + i] = conj (matrix[i * DIM + j]); // Forces the pseudo-matrix Hermitian + } // m(12*i+j) = M(i,j) - eigensystem_hermitian (matrix, DIM, eigenvalue, Q); + eigensystem_hermitian (matrix, DIM, eigenvalue, Q); /* Produces unitary matrix Q: rows hold the conjugate eigenvectors */ - for (int i = 0; i < DIM; i++) + for (i = 0; i < DIM; i++) index[i] = i; + if (verbose >= 6) { + printf ("Before bubblesort\n"); + for (i = 0; i < DIM; i++) { + printf ("i = %i eigenvalue[i] = %g index[i] %i \n", i, eigenvalue[i], index[i]); + } + } + + // Sort eigenvalues from lowest to highest bubblesort (DIM, eigenvalue, index); + if (verbose >= 5) { + printf ("After bubblesort\n"); + for (i = 0; i < DIM; i++) { + printf ("i = %i eigenvalue[i] = %g index[i] %i \n", i, eigenvalue[i], index[i]); + } + } double eigenvaluegood[DIM]; - for (int j = 0; j < DIM; j++) - eigenvectorgood[j] = Q[index[mode] * DIM + j]; + for (j = 0; j < DIM; j++) { + eigenvectorgood[j] = Q[index[mode] * DIM + j]; /* Rows of Q hold the complex conjugate eigenvectors */ + } + + if (verbose >= 7) { + printf ("Q = ["); + for (i = 0; i < DIM; i++) { + printf ("\n ("); + for (j = 0; j < DIM; j++) + printf (" %g +i %g, ", creal (Q[i * DIM + j]), cimag (Q[i * DIM + j])); + printf (")"); + } + printf ("]\n"); + } + + for (i = 0; i < DIM; i++) + eigenvaluegood[i] = sqrt (fmax (eigenvalue[index[i]], 0)) / sqrt (M) / (2 * PI * 1E12) * THz2meV; + + if (verbose >= 5) { + printf ("Diagonalization done, eigenvalues (energies) are: \n ("); + for (i = 0; i < DIM; i++) + printf (" %g, ", eigenvaluegood[i]); + printf (" )\n"); + printf ("mode = %i , eigenvectorgood[mode] = (", mode); + for (j = 0; j < DIM; j++) + printf (" %g +i %g, ", creal (eigenvectorgood[j]), cimag (eigenvectorgood[j])); + printf (" )\n"); + } + + // ___________________ RETURN RESULTS TO CALLING FUNCTION __________________ - for (int i = 0; i < DIM; i++) - eigenvaluegood[i] = sqrt (eigenvalue[index[i]]) / sqrt (M) / (2 * PI * 1E12) * THz2meV; + hw_neutron = fabs (VS2E * (vi * vi - vf * vf)); // neutron energy transfer + hw_phonon = eigenvaluegood[mode]; // phonon energy - hw_neutron = fabs (VS2E * (vi * vi - vf * vf)); - hw_phonon = eigenvaluegood[mode]; + if (verbose >= 4) { + printf ("Return from omega_q \n"); + printf ("hw_neutron = %g (vi = %g, vf = %g)\n", hw_neutron, vi, vf); + printf ("hw_phonon = %g = eigenvaluegood[%d] = %g = eigenvalues[%d] = %g\n", hw_phonon, mode, eigenvaluegood[mode], index[mode], eigenvalue[index[mode]]); + } parms[8] = cplx (hw_phonon, 0.0); + if (disp == 1) { + printf ("mode = %d, (h,k,l) = (%g,%g,%g), hw_q = %g", mode, h, k, l, hw_phonon); + printf (" polarization: ( "); + for (j = 0; j < DIM; j++) { + printf ("%g + i %g ,", creal (eigenvectorgood[j]), cimag (eigenvectorgood[j])); + } + printf (")\n"); + } + if (disp == 2) { + hk_inplane = qx / astar; + hk_outofplane = qy / astar; + k = 2 / sqrt (3) * hk_outofplane; + h = hk_inplane - hk_outofplane / sqrt (3); + l = qz / cstar; + printf ("mode = %d, vf = %g, (h,k,l) = (%g,%g,%g), hw_q = %g", mode, vf, h, k, l, hw_phonon); + printf (" polarization: (%g + i %g , %g + i %g , %g + i %g) \n", creal (eigenvectorgood[0]), cimag (eigenvectorgood[0]), creal (eigenvectorgood[1]), + cimag (eigenvectorgood[1]), creal (eigenvectorgood[2]), cimag (eigenvectorgood[2])); + } + return (hw_phonon - hw_neutron); } + //--------------------- END CENTRAL FUNCTION OMEGA-Q ------------------------ // ------------------START OF THE GENERAL ROOT-FINDING CODE-------------------- - double last_fh, last_x2; + #pragma acc routine seq double - zridd (double (*func) (cdouble*, cdouble*), double x1, double x2, cdouble* parms, cdouble* vector, double xacc) { + bvk_zridd (double x1, double x2, cdouble* parms, cdouble* vector, double xacc, double* last_fh, double* last_x2) { int j; double ans, fh, fl, fm, fnew, s, xh, xl, xm, xnew; - /* ---- f(x1) ---- */ - parms[0] = cplx (x1, 0.0); - if (xacc > 0) { - fl = (*func) (parms, vector); + fl = bvk_omega_q (parms, vector); } else { - fl = last_fh; + fl = *last_fh; xacc = -xacc; - - if (fabs (x1 - last_x2) > xacc) - printf ("*** error in zridd *** x1: %g last_x2: %g \n", x1, last_x2); + if (fabs (x1 - *last_x2) > xacc) + printf ("*** error in zridd *** x1: %g last_x2: %g \n", x1, *last_x2); } - - /* ---- f(x2) ---- */ - parms[0] = cplx (x2, 0.0); - last_fh = fh = (*func) (parms, vector); - last_x2 = x2; - - if (fl * fh >= 0) { - if (fl == 0) - return x1; - if (fh == 0) - return x2; - return UNUSED; - } - - xl = x1; - xh = x2; - ans = UNUSED; - - for (j = 1; j < MAXRIDD; j++) { - xm = 0.5 * (xl + xh); - - parms[0] = cplx (xm, 0.0); - fm = (*func) (parms, vector); - - s = sqrt (fm * fm - fl * fh); - if (s == 0.0) - return ans; - - xnew = xm + (xm - xl) * ((fl >= fh ? 1.0 : -1.0) * fm / s); - - if (fabs (xnew - ans) <= xacc) - return ans; - - ans = xnew; - - parms[0] = cplx (ans, 0.0); - fnew = (*func) (parms, vector); - - if (fnew == 0.0) - return ans; - - if (fabs (fm) * SIGN (fnew) != fm) { - xl = xm; - fl = fm; - xh = ans; - fh = fnew; - } else if (fabs (fl) * SIGN (fnew) != fl) { - xh = ans; - fh = fnew; - } else if (fabs (fh) * SIGN (fnew) != fh) { - xl = ans; - fl = fnew; - } else { - fatalerror ("never get here in zridd"); - } - - if (fabs (xh - xl) <= xacc) - return ans; - } - - fatalerror ("zridd exceeded maximum iterations"); - return 0.0; - } - - #ifdef OPENACC - #pragma acc routine - double - zridd_gpu (double x1, double x2, cdouble* parms, cdouble* vector, double xacc) { - int j; - double ans, fh, fl, fm, fnew, s, xh, xl, xm, xnew; - - parms[0] = x1; - // fprintf(stderr,"fl=omega_q(%g+i%g)\n",creal(parms[0]),cimag(parms[0])); - fl = omega_q (parms, vector); - parms[0] = x2; - // fprintf(stderr,"fh=omega_q(%g+i%g)\n",creal(parms[0]),cimag(parms[0])); - fh = omega_q (parms, vector); + *last_fh = fh = bvk_omega_q (parms, vector); + *last_x2 = x2; if (fl * fh >= 0) { if (fl == 0) return x1; @@ -697,9 +586,8 @@ SHARE ans = UNUSED; for (j = 1; j < MAXRIDD; j++) { xm = 0.5 * (xl + xh); - parms[0] = xm; - // fprintf(stderr,"fm=omega_q(%g+i%g)\n",creal(parms[0]),cimag(parms[0])); - fm = omega_q (parms, vector); + parms[0] = cplx (xm, 0.0); + fm = bvk_omega_q (parms, vector); s = sqrt (fm * fm - fl * fh); if (s == 0.0) return ans; @@ -707,9 +595,8 @@ SHARE if (fabs (xnew - ans) <= xacc) return ans; ans = xnew; - parms[0] = ans; - // fprintf(stderr,"fnew=omega_q(%g+i%g)\n",creal(parms[0]),cimag(parms[0])); - fnew = omega_q (parms, vector); + parms[0] = cplx (ans, 0.0); + fnew = bvk_omega_q (parms, vector); if (fnew == 0.0) return ans; if (fabs (fm) * SIGN (fnew) != fm) { @@ -724,121 +611,99 @@ SHARE xl = ans; fl = fnew; } else - fatalerror ("never get here in zridd"); + bvk_fatalerror ("never get here in zridd"); if (fabs (xh - xl) <= xacc) return ans; } - fatalerror ("zridd exceeded maximum iterations"); + bvk_fatalerror ("zridd exceeded maximum iterations"); } - return 0.0; /* Never get here */ + return UNUSED; /* Never get here on the CPU; on the GPU, drop this bracket */ } - #endif #define ROOTACC 1e-8 - #define PG_MAX_ROOTS 20 /* size of vf_list in TRACE */ + #define BVK_MAX_ROOTS 20 /* size of vf_list in TRACE */ - int - findroots (double brack_low, double brack_mid, double brack_high, double* list, int* index, double (*f) (cdouble*, cdouble*), cdouble* parms, cdouble* vector, - int low_steps, int high_steps) { + #pragma acc routine seq + void + bvk_addroot (double* list, int* index, double root) { + if (*index < BVK_MAX_ROOTS) + list[(*index)++] = root; + #ifndef OPENACC + else { + static int warned = 0; // ponytail: benign race under threads, only guards a one-time warning + if (!warned++) + fprintf (stderr, "Phonon_BvK_PG: more than %d roots found, extra roots dropped (intensity biased low)\n", BVK_MAX_ROOTS); + } + #endif + } + + #pragma acc routine seq + void + bvk_findroots (double brack_low, double brack_mid, double brack_high, double* list, int* index, cdouble* parms, cdouble* vector, int low_steps, int high_steps) { double root, range; + double last_fh = 0, last_x2 = 0; int i; + int verbose = (int)creal (parms[14]); if (verbose == 2) - printf ("Enter findroots() \n"); - - /* ---- Energy loss side ---- */ + printf ("Enter bvk_findroots() \n"); + // First, find roots in energy loss side, if any range = brack_mid - brack_low; - for (i = 0; i < (low_steps - 1); i++) { if (i == 0) - root = zridd (f, brack_low + range * i / low_steps, brack_low + range * (i + 1) / low_steps, parms, vector, ROOTACC); + root = bvk_zridd (brack_low + range * i / low_steps, brack_low + range * (i + 1) / low_steps, parms, vector, ROOTACC, &last_fh, &last_x2); else - root = zridd (f, brack_low + range * i / low_steps, brack_low + range * (i + 1) / low_steps, parms, vector, -ROOTACC); - + root = bvk_zridd (brack_low + range * i / low_steps, brack_low + range * (i + 1) / low_steps, parms, vector, -ROOTACC, &last_fh, &last_x2); if (root != UNUSED) { - if (*index < PG_MAX_ROOTS) - list[(*index)++] = root; + bvk_addroot (list, index, root); if (verbose >= 4) printf ("--- findroots returned a root on the energy loss side: vf = %g \n", root); } } range = (brack_mid - brack_low) / (double)low_steps; - for (i = 0; i < low_steps; i++) { if (low_steps == 1) - root = zridd (f, brack_low + range * (low_steps - 1 + i / (double)low_steps), brack_low + range * (low_steps - 1 + (i + 1) / (double)low_steps), parms, - vector, ROOTACC); + root = bvk_zridd (brack_low + range * (low_steps - 1 + i / (double)low_steps), brack_low + range * (low_steps - 1 + (i + 1) / (double)low_steps), + parms, vector, ROOTACC, &last_fh, &last_x2); else - root = zridd (f, brack_low + range * (low_steps - 1 + i / (double)low_steps), brack_low + range * (low_steps - 1 + (i + 1) / (double)low_steps), parms, - vector, -ROOTACC); - + root = bvk_zridd (brack_low + range * (low_steps - 1 + i / (double)low_steps), brack_low + range * (low_steps - 1 + (i + 1) / (double)low_steps), + parms, vector, -ROOTACC, &last_fh, &last_x2); if (root != UNUSED) { - if (*index < PG_MAX_ROOTS) - list[(*index)++] = root; + bvk_addroot (list, index, root); if (verbose >= 4) printf ("--- findroots returned a root on the energy loss side: vf = %g \n", root); } } - /* ---- Energy gain side ---- */ - + // Second, find roots in energy gain side, there is always some range = (brack_high - brack_mid) / (double)high_steps; - for (i = 0; i < high_steps; i++) { if (i == 0) - root = zridd (f, brack_mid + range * i / (double)high_steps, brack_mid + range * (i + 1) / (double)high_steps, parms, vector, ROOTACC); + root = bvk_zridd (brack_mid + range * i / (double)high_steps, brack_mid + range * (i + 1) / (double)high_steps, parms, vector, ROOTACC, &last_fh, + &last_x2); else - root = zridd (f, brack_mid + range * i / (double)high_steps, brack_mid + range * (i + 1) / (double)high_steps, parms, vector, -ROOTACC); - + root = bvk_zridd (brack_mid + range * i / (double)high_steps, brack_mid + range * (i + 1) / (double)high_steps, parms, vector, -ROOTACC, &last_fh, + &last_x2); if (root != UNUSED) { - if (*index < PG_MAX_ROOTS) - list[(*index)++] = root; + bvk_addroot (list, index, root); if (verbose >= 4) printf ("*** findroots returned a root on the energy gain side: vf = %g \n", root); } } range = brack_high - brack_mid; - for (i = 1; i < high_steps; i++) { - root = zridd (f, brack_mid + range * i / (double)high_steps, brack_mid + range * (i + 1) / (double)high_steps, parms, vector, -ROOTACC); - + root = bvk_zridd (brack_mid + range * i / (double)high_steps, brack_mid + range * (i + 1) / (double)high_steps, parms, vector, -ROOTACC, &last_fh, + &last_x2); if (root != UNUSED) { - if (*index < PG_MAX_ROOTS) - list[(*index)++] = root; + bvk_addroot (list, index, root); if (verbose >= 4) printf ("*** findroots returned a root on the energy gain side: vf = %g \n", root); } } - - return *index; - } - - #ifdef OPENACC - #pragma acc routine - int - findroots_gpu (double brack_low, double brack_mid, double brack_high, double* list, int* index, double* parms, cdouble* vector, int low_steps, int high_steps) { - double root, range = brack_mid - brack_low; - int i; - - for (i = 0; i < low_steps; i++) { - root = zridd_gpu (brack_low + range * i / (int)low_steps, brack_low + range * (i + 1) / (int)low_steps, (cdouble*)parms, (cdouble*)vector, ROOTACC); - if (root != UNUSED) { - if (*index < PG_MAX_ROOTS) - list[(*index)++] = root; - } - } - for (i = 0; i < high_steps; i++) { - root = zridd_gpu (brack_mid + range * i / (int)high_steps, brack_high + range * (i + 1) / (int)high_steps, (cdouble*)parms, (cdouble*)vector, ROOTACC); - if (root != UNUSED) { - if (*index < PG_MAX_ROOTS) - list[(*index)++] = root; - } - } } - #endif #undef UNUSED #undef MAXRIDD @@ -850,27 +715,20 @@ DECLARE double V_rho; double V_my_s; double V_my_a_v; - cdouble** Matrix; - int i, j; - double q[3]; /* Scattering vector */ - double qx; - double qy; - double qz; - double q_x; - double q_y; - double q_z; - cdouble eigenvectormode[DIM]; - cdouble p_call[15]; /* Parameter list to transfer to omega_q; elsewhere known as parms */ + int shape; + int verbose; %} INITIALIZE %{ - int i, j; - if (xwidth && yheight && zdepth) shape = 1; /* box */ else if (radius > 0 && yheight) shape = 0; /* cylinder */ + else { + fprintf (stderr, "Phonon_BvK_PG: %s: give either radius and yheight, or xwidth, yheight and zdepth.\n", NAME_CURRENT_COMP); + exit (-1); + } V_rho = 1 / (a_latt * a_latt * c_latt * sqrt3 / 2); // (unit: Å^-3) V_my_s = (V_rho * 100 * sigma_inc); @@ -886,140 +744,151 @@ INITIALIZE coords_get (ToTarget, &target_x, &target_y, &target_z); } if (!(target_x || target_y || target_z)) { - printf ("Phonon_BkV_PG: %s: The target is not defined. Using direct beam (Z-axis).\n", NAME_CURRENT_COMP); + printf ("Phonon_BvK_PG: %s: The target is not defined. Using direct beam (Z-axis).\n", NAME_CURRENT_COMP); target_z = 1; } initialize_omega_q (); verbose = verbose_input; - // m(12*i+j) = M(i,j) - // OK for: Rot120 - for (i = 0; i < 3; i++) - for (j = 0; j < 3; j++) { - printf ("i,j: %i, %i ROT120: %g ROT120FLAT; %g \n", i, j, Rot120[i][j], Rot120_flat[3 * i + j]); - } - %} + #pragma acc update device(M, Delta, astar, cstar, r_j13) + #pragma acc update device(Phi1, Phi2, Phi3, Phi4, Phi5, Phi6, Phi7, Phi8, Phi9, Phi10, Phi11, Phi12, Phi13, Phi14) +%} TRACE %{ - double t0, t1, t2, t3; - double v_i, v_f; - double vx_i, vy_i, vz_i; - double dt0, dt; - double l_full; - double l_i, l_o; - double my_a_i; - double my_a_f; - double solid_angle; - double aim_x = 0, aim_y = 0, aim_z = 1; - double omega; - double qsquare; - double bose_factor; - int nf, index; - double vf_list[PG_MAX_ROOTS]; - double J_factor, J0; - double f1, f2; - double p_att, p_MC, p_ph1, p_ph2, p_ph3, p_J, p_tot; - cdouble Gn; - cdouble q_dot_e; - double q_dot_r; - double qh, qk, ql, qhkx, qhky, qh_Bragg, qk_Bragg, ql_Bragg, qh_res, qk_res, qh_add, qk_add; - double dist_00, dist_01, dist_10, dist_11, dist_min; - int i, j, intersect; + double t0, t1, t2, t3; /* Entry/exit times for shape */ + double v_i, v_f; /* Neutron velocities: initial, final */ + double vx_i, vy_i, vz_i; /* Neutron initial velocity vector */ + double dt0, dt; /* Flight times through sample */ + double l_full; /* Flight path length for non-scattered neutron */ + double l_i, l_o; /* Flight path length in/out for scattered neutron */ + double my_a_i; /* Initial attenuation factor */ + double my_a_f; /* Final attenuation factor */ + double solid_angle; /* Solid angle of target as seen from scattering point */ + double aim_x = 0, aim_y = 0, aim_z = 1; /* Position of target relative to scattering point */ + double omega; /* Energy transfer */ + double qsquare; /* Square of the scattering vector */ + double bose_factor; /* Calculated value of the Bose factor */ + int nf, index; /* Number of allowed final velocities */ + double vf_list[BVK_MAX_ROOTS]; /* List of allowed final velocities */ + double J_factor, J0; /* Jacobian from delta fnc.s in cross section; elastic J-factor */ + double f1, f2; /* Probed values of omega_q minus omega */ + double p_att, p_MC, p_ph1, p_ph2, p_ph3, p_J, p_tot; /* Temporary multipliers */ + cdouble Gn; /* Phonon structure factor */ + cdouble q_dot_e; /* q dot e */ + double q_dot_r; /* q dot r */ + double qh, qk, ql, qhkx, qhky, qh_Bragg, qk_Bragg, ql_Bragg, qh_res, qk_res, qh_add, qk_add; /* components of q */ + double dist_00, dist_01, dist_10, dist_11, dist_min; /* used to calculate nearest Bragg point */ + int i, j, intersect, mode; + double q[3], q_x, q_y, q_z; /* Scattering vector */ + cdouble eigenvectormode[DIM]; + cdouble p_call[15]; /* Parameters for bvk_omega_q */ if (shape == 0) intersect = cylinder_intersect (&t0, &t3, x, y, z, vx, vy, vz, radius, yheight); - else if (shape == 1) + else intersect = box_intersect (&t0, &t3, x, y, z, vx, vy, vz, xwidth, yheight, zdepth); if (intersect) { if (t0 < 0) - ABSORB; + ABSORB; /* Neutron came from the sample or begins inside */ if (verbose >= 2) - printf ("neutron entered Phonon_BvK_PG\n"); - - dt0 = t3 - t0; + printf ("neutron entered Phonon_BvK_PG \n"); + /* Neutron enters at t=t0. */ + dt0 = t3 - t0; /* Time in sample */ v_i = sqrt (vx * vx + vy * vy + vz * vz); - l_full = v_i * dt0; - dt = rand01 () * dt0; - l_i = v_i * dt; - + l_full = v_i * dt0; /* Length of path through sample if not scattered */ + dt = rand01 () * dt0; /* Time of scattering (relative to t0) */ + l_i = v_i * dt; /* Penetration in sample at scattering */ vx_i = vx; vy_i = vy; vz_i = vz; + PROP_DT (dt + t0); /* Point of scattering */ - PROP_DT (dt + t0); - - aim_x = target_x - x; + aim_x = target_x - x; /* Vector pointing at target (e.g. analyzer) */ aim_y = target_y - y; aim_z = target_z - z; - if (focus_aw && focus_ah) + if (focus_aw && focus_ah) { randvec_target_rect_angular (&vx, &vy, &vz, &solid_angle, aim_x, aim_y, aim_z, focus_aw, focus_ah, ROT_A_CURRENT_COMP); - else if (focus_xw && focus_yh) + } else if (focus_xw && focus_yh) { randvec_target_rect (&vx, &vy, &vz, &solid_angle, aim_x, aim_y, aim_z, focus_xw, focus_yh, ROT_A_CURRENT_COMP); - else + } else { randvec_target_sphere (&vx, &vy, &vz, &solid_angle, aim_x, aim_y, aim_z, focus_r); - + } NORM (vx, vy, vz); - nf = 0; - - if (mode_input == 12) - mode = round (12 * rand01 () - 0.5); - else - mode = mode_input; - - if ((mode < 0) || (mode > 11)) { + if (mode_input == 12) { + mode = (int)floor (12 * rand01 ()); // Select the phonon mode + if (mode > 11) + mode = 11; /* rand01() may return 1.0 */ + } else + mode = mode_input; // Use the given mode + if ((mode < 0) || (mode > 11)) // Undefined mode specification + { printf ("mode = %d ", mode); nrerror ("illegal value of mode"); + ABSORB; /* reached only on the GPU, where nrerror does not exit */ } - p_call[0] = cplx (-1, 0); - p_call[1] = cplx (v_i, 0); - p_call[2] = cplx (vx, 0); - p_call[3] = cplx (vy, 0); - p_call[4] = cplx (vz, 0); - p_call[5] = cplx (vx_i, 0); - p_call[6] = cplx (vy_i, 0); - p_call[7] = cplx (vz_i, 0); - p_call[9] = cplx (hh, 0); - p_call[10] = cplx (kk, 0); - p_call[11] = cplx (ll, 0); - p_call[12] = cplx ((double)dispersion, 0); + /* Load the parameter vector. All values are real-valued physics quantities; + they are stored as cdouble so that omega_q and the root-finder can mutate + p_call[0] (v_f) as a simple scalar without breaking the MSVC struct layout. */ + p_call[0] = cplx (-1.0, 0.0); // in the end this will be v_f + p_call[1] = cplx (v_i, 0.0); + p_call[2] = cplx (vx, 0.0); + p_call[3] = cplx (vy, 0.0); + p_call[4] = cplx (vz, 0.0); + p_call[5] = cplx (vx_i, 0.0); + p_call[6] = cplx (vy_i, 0.0); + p_call[7] = cplx (vz_i, 0.0); + p_call[9] = cplx (hh, 0.0); + p_call[10] = cplx (kk, 0.0); + p_call[11] = cplx (ll, 0.0); + p_call[12] = cplx ((double)dispersion, 0.0); + p_call[13] = cplx ((double)mode, 0.0); + p_call[14] = cplx ((double)verbose, 0.0); if (dispersion == 1) { - f1 = omega_q (p_call, eigenvectormode); - p_call[12] = cplx (0, 0); + f1 = bvk_omega_q (p_call, eigenvectormode); + p_call[12] = cplx (0.0, 0.0); } - #ifndef OPENACC - findroots (0, v_i, v_i + V_HIGH, vf_list, &nf, omega_q, p_call, eigenvectormode, e_steps_low, e_steps_high); - #else - findroots_gpu (0, v_i, v_i + V_HIGH, vf_list, &nf, p_call, eigenvectormode, e_steps_low, e_steps_high); - #endif + if (verbose >= 2) + printf ("Call findroots \n"); + bvk_findroots (0, v_i, v_i + V_HIGH, vf_list, &nf, p_call, eigenvectormode, e_steps_low, e_steps_high); + if (verbose >= 2) + printf ("Findroots returned %d roots \n", nf); + + if (verbose >= 3) { + printf ("Findroots done, mode %i , last phonon energy %g \n", mode, creal (p_call[8])); + } - if (nf == 0) - ABSORB; /* no final velocity satisfies energy and momentum conservation */ - index = (int)floor (rand01 () * nf); + // ________________ ALL FROM HERE IS CALCULATION OF INTENSITY - SHOULD BE REVISITED ___________________ + + if (!nf) + ABSORB; /* no allowed final velocity, the weight would be zero */ + index = (int)floor (rand01 () * nf); // Select random solution if (index >= nf) index = nf - 1; /* rand01() may return 1.0 */ v_f = vf_list[index]; - p_call[0] = cplx (v_f, 0); - - f1 = omega_q (p_call, eigenvectormode); - - p_call[0] = cplx (v_f - DV, 0); - f1 = omega_q (p_call, eigenvectormode); - - p_call[0] = cplx (v_f + DV, 0); - f2 = omega_q (p_call, eigenvectormode); - - J_factor = 1.6022E-22 * 1E-10 * fabs (f2 - f1) / (2 * DV * V2K); - - omega = VS2E * (v_i * v_i - v_f * v_f); - + p_call[0] = cplx (v_f, 0.0); // transfer choice of v_f to omega_q + + f1 = bvk_omega_q (p_call, eigenvectormode); + if (verbose >= 2) { + printf ("Chosen solution is %i; v_f = %g, hw = %g, energy match is %g; eigenvector is: (", index, v_f, creal (p_call[8]), f1); + for (j = 0; j < DIM; j++) + printf (" %g +i %g, ", creal (eigenvectormode[j]), cimag (eigenvectormode[j])); + printf (" )\n"); + printf ("--------------------------------------------\n"); + } + p_call[0] = cplx (v_f - BVK_DV, 0.0); + f1 = bvk_omega_q (p_call, eigenvectormode); + p_call[0] = cplx (v_f + BVK_DV, 0.0); + f2 = bvk_omega_q (p_call, eigenvectormode); + J_factor = BVK_MEV2J * 1E-10 * fabs (f2 - f1) / (2 * BVK_DV * V2K); // unit: meV Å converted to SI units + omega = VS2E * (v_i * v_i - v_f * v_f); // unit: meV vx *= v_f; vy *= v_f; vz *= v_f; @@ -1029,13 +898,13 @@ TRACE q_z = q[2] = V2K * (vz_i - vz); ql = q_z / cstar; - qhkx = q_x / astar; - qhky = q_y / astar; - qk = 2 * qhky / sqrt3; - qh = qhkx - qhky / sqrt3; + qhkx = q_x / astar; // These are the Cartesian coordinates + qhky = q_y / astar; // These are the Cartesian coordinates + qk = 2 * qhky / sqrt3; // transform to hexagonal coordinates + qh = qhkx - qhky / sqrt3; // transform to hexagonal coordinates + // Now find out which Bragg peak is closest in hexagonal plane ql_Bragg = round (ql); - qh_Bragg = floor (qh); qh_res = qh - qh_Bragg; qk_Bragg = floor (qk); @@ -1052,14 +921,12 @@ TRACE qh_add = 0; qk_add = 1; } - dist_10 = (qh_res - 1) * (qh_res - 1) + qk_res * qk_res + (qh_res - 1) * qk_res; if (dist_10 < dist_min) { dist_min = dist_10; qh_add = 1; qk_add = 0; } - dist_11 = (qh_res - 1) * (qh_res - 1) + (qk_res - 1) * (qk_res - 1) + (qh_res - 1) * (qk_res - 1); if (dist_11 < dist_min) { dist_min = dist_11; @@ -1067,82 +934,95 @@ TRACE qk_add = 1; } - qh_Bragg += qh_add; + qh_Bragg += qh_add; // Potentially correct to the right BZ qk_Bragg += qk_add; + if (verbose >= 3) { + printf ("q transforms in r.l.u to (h, k, l) = (%g, %g, %g). Nearest Bragg point: (%g, %g, %g)\n", qh, qk, ql, qh_Bragg, qk_Bragg, ql_Bragg); + printf ("--------------------------------------------\n"); + } + + if (dispersion == 1) + printf ("Dispersion found: (h,k,l) = (%g, %g, %g), hw = %g \n", q[0] / astar, q[1] / astar, q[2] / cstar, omega); + qsquare = 1; - Gn = cplx (0, 0); + Gn = cplx (0.0, 0.0); - double q_Bragg[3]; + double q_Bragg[3]; // Bragg point in Å-1; Cartesian coordinates q_Bragg[0] = qh_Bragg * astar + qk_Bragg * astar / 2.0; q_Bragg[1] = qk_Bragg * astar * 2.0 / sqrt3; q_Bragg[2] = ql_Bragg * cstar; - for (i = 0; i < 4; i++) { - q_dot_e = cplx (0, 0); + for (i = 0; i < 4; i++) // Calculate phonon structure factor; loop over atoms in unit cell + { + q_dot_e = cplx (0.0, 0.0); q_dot_r = 0; - - for (j = 0; j < 3; j++) { + for (j = 0; j < 3; j++) // loop over x,y,z + { q_dot_e = cadd (q_dot_e, rmul (q[j], eigenvectormode[3 * i + j])); q_dot_r += q_Bragg[j] * Delta[3 * i + j]; } - - cdouble phase = cexp (cplx (0, q_dot_r)); - Gn = cadd (Gn, cmul (cplx (b_length, 0), cmul (q_dot_e, phase))); + if (verbose >= 7) { + cdouble contrib = rmul (b_length, cmul (q_dot_e, cexp (cplx (0.0, q_dot_r)))); + printf ("i = %i , q dot r = %g , q dot e = (%g + i %g) Fn contrib = (%g + i %g)", i, q_dot_r, creal (q_dot_e), cimag (q_dot_e), creal (contrib), + cimag (contrib)); + } + /* Gn += b_length * q_dot_e * e^{i q_Bragg . Delta_i} */ + Gn = cadd (Gn, rmul (b_length, cmul (q_dot_e, cexp (cplx (0.0, q_dot_r))))); } + if (verbose >= 7) + printf ("\n"); if (shape == 0) intersect = cylinder_intersect (&t0, &t3, x, y, z, vx, vy, vz, radius, yheight); - else if (shape == 1) + else intersect = box_intersect (&t0, &t3, x, y, z, vx, vy, vz, xwidth, yheight, zdepth); - if (!intersect) { - printf ("FATAL ERROR\n"); - exit (1); - } + if (!intersect) + ABSORB; /* numerical edge case, e.g. near-tangent path */ + + if (verbose >= 2) + printf ("neutron left Phonon_BvK_PG, p_init = %g,", p); dt = t3; l_o = v_f * dt; - my_a_i = V_my_a_v / v_i; my_a_f = V_my_a_v / v_f; + bose_factor = bvk_nbose (omega, T); + J0 = HBAR * HBAR * v_f * V2K * 1E10 / MNEUTRON; // The J-factor for a flat mode - bose_factor = nbose (omega, T); - - J0 = hbar * hbar * v_f * V2K * 1E10 / mn; - - p_att = exp (-(V_my_s * (l_i + l_o) + my_a_i * l_i + my_a_f * l_o)); - p_MC = nf * solid_angle * l_full; - + p_att = exp (-(V_my_s * (l_i + l_o) + my_a_i * l_i + my_a_f * l_o)); /* Absorption factor (units: 1) */ + p_MC = nf * solid_angle * l_full; /* Focusing factors; assume random choice of n_f possibilities (units: m) */ if (mode_input == 12) - p_MC *= 12; - - p_ph1 = (v_f / v_i) * DW * bose_factor * V_rho * 1E30; - p_ph2 = hbar * hbar / (2 * (fabs (omega) * 1.6022E-22)); - - cdouble G2 = cmul (Gn, conj (Gn)); - p_ph3 = creal (G2) * 1E-10 / M; - + p_MC *= 12; // The factor 12 is due to MC choice between the 12 modes + p_ph1 = (v_f / v_i) * DW * bose_factor * V_rho * 1E30; /* Cross section factor 1 (units: m^-3) */ + p_ph2 = HBAR * HBAR / (2 * (fabs (omega) * BVK_MEV2J)); /* Jacobian of delta functions in cross section; and constants. Units: J s^2 */ + p_ph3 = creal (cmul (Gn, conj (Gn))) * 1E-10 / M; /* Structure factor |Gn|^2. (units: kg^-1) */ p_J = J0 / J_factor; - p_tot = p_att * p_MC * p_ph1 * p_ph2 * p_ph3 * p_J; + if (!isfinite (p_tot)) + ABSORB; /* e.g. omega == 0 */ p *= p_tot; - } + + if (verbose >= 2) { + cdouble GnGnconj = cmul (Gn, conj (Gn)); + printf ("nf = %d, solid_angle = %g, l_full = %g \n", nf, solid_angle, l_full); + printf ("(p in SI units): p_att = %g, p_MC= %g, p_ph1 = %g, p_ph2 = %g, p_ph3 = %g, p_J = %g, bose_factor = %g, V_rho = %g, J_factor/J0 = %g, G_factor = " + "%g +i %g, conj(G) = %g + I %g, |G|^2 = %g + I %g, ", + p_att, p_MC, p_ph1, p_ph2, p_ph3, p_J, bose_factor, V_rho, J_factor / J0, creal (Gn), cimag (Gn), creal (conj (Gn)), cimag (conj (Gn)), + creal (GnGnconj), cimag (GnGnconj)); + printf (" p_tot = %g, p_final = %g\n", p_tot, p); + printf ("q is (h,k,l) = (%g, %g, %g) \n", q_x / astar, q_y / astar, q_z / cstar); + } + } /* else transmit: Neutron did not hit the sample */ %} MCDISPLAY %{ - if (radius > 0 && yheight) { /* cylinder */ + if (shape == 0) cylinder (0, 0, 0, radius, yheight, 0, 0, 1, 0); - } else if (xwidth && yheight) { /* box/rectangle */ + else box (0, 0, 0, xwidth, yheight, zdepth, 0, 0, 1, 0); - } - // circle("xz", 0, yheight/2.0, 0, radius); - // circle("xz", 0, -yheight/2.0, 0, radius); - // line(-radius, -yheight/2.0, 0, -radius, +yheight/2.0, 0); - // line(+radius, -yheight/2.0, 0, +radius, +yheight/2.0, 0); - // line(0, -yheight/2.0, -radius, 0, +yheight/2.0, -radius); - // line(0, -yheight/2.0, +radius, 0, +yheight/2.0, +radius); %} END diff --git a/mcstas-comps/examples/Tests_samples/Test_Phonon_BvK_PG/Test_Phonon_BvK_PG.instr b/mcstas-comps/examples/Tests_samples/Test_Phonon_BvK_PG/Test_Phonon_BvK_PG.instr new file mode 100644 index 0000000000..ebf596a3bf --- /dev/null +++ b/mcstas-comps/examples/Tests_samples/Test_Phonon_BvK_PG/Test_Phonon_BvK_PG.instr @@ -0,0 +1,263 @@ +/******************************************************************************* +* Instrument: Test_Phonon_BvK_PG +* +* %I +* Written by: Kim Lefmann +* Date: 20.07.2025 +* Origin: NBI, KU +* %INSTRUMENT_SITE: Generic +* +* Simple thermal triple axis spectrometer for test of Phonon_BvK_PG component +* Modified from older version by Dorothy Wang, Kim Lefmann +* +* %D +* Simple TAS to test the Phonon_BvK_PG component +* +* %Example: -y Detector: e_mon_beforeana_zoom_I=6.90143e-16 +* +* %P +* E: [meV] Energy transfer +* Ef: [meV] Final energy +* Dlambda: [AA] width of wavelength band +* h: [rlu] q-value in hexagonal plane +* l: [rlu] q-value out of hexagonal plane +* dA3: [deg] A3 offset +* Temp: [K] Sample temperature +* width: [m] Beam and sample size scale +* coll: [arcmin] Collimation before the sample +* phononmode: [1] Phonon mode 0..11, or 12 for random choice +* E_steps_high: [1] Root-search intervals above the elastic line +* E_steps_low: [1] Root-search intervals below the elastic line +* Verbose: [1] Verbosity level of the sample component +* DISP: [1] Dispersion output mode of the sample component +* +* %E +*******************************************************************************/ +DEFINE INSTRUMENT Test_Phonon_BvK_PG(E=4.5,Ef=25,Dlambda=0.05,h=0,l=3.5,dA3=90, Temp=2, width = 0.001, coll = 20, int phononmode=0,int E_steps_high=100, int E_steps_low=100, int Verbose=0, int DISP=0) + +DECLARE +%{ + double Ei; + double thetaM; + double twothetaS; + double thetaA; + double A3; + double QM; + double alpha; + double lambda_i; + double SMALL__NUMBER; +%} + +INITIALIZE +%{ + // Set monochromator/analyzer Q-value for PG + QM = 1.8734; + SMALL__NUMBER = 1e-6; + + double a = 2.461; + double c = 6.708; + // PG lattice constants, in AA, same as in Phonon_BvK_PG.comp + + double astar = 4 * PI / (sqrt (3) * a); + double cstar = 2 * PI / c; + // PG reciprocal lattice vectors, in 1/AA + + // calculate Ei + Ei = Ef + E; + + // calculate ki, kf, lambda_i, q + double ki = V2K * SE2V * sqrt (Ei); + double kf = V2K * SE2V * sqrt (Ef); + lambda_i = 2 * PI / ki; + double q = sqrt (h * h * astar * astar + l * l * cstar * cstar); + + // calculate 2thetaM and 2thetaA + thetaM = asin (QM / (2 * ki)) * RAD2DEG; + thetaA = asin (QM / (2 * kf)) * RAD2DEG; + + // calculate scattering angle and sample rotation + twothetaS = acos ((q * q - ki * ki - kf * kf) / (-2 * ki * kf)) * RAD2DEG; + alpha = acos ((kf * kf - ki * ki - q * q) / (-2 * ki * q)); + A3 = (asin (l * cstar / (q + SMALL__NUMBER)) - alpha) * RAD2DEG; + + // Printout calculations + printf ("a = %g c = %g astar = %g cstar = %g \n", a, c, astar, cstar); + printf ("ki = %g kf = %g q = %g lambda_i = %g\n", ki, kf, q, lambda_i); + printf ("thetaA = %g thetaM = %g\n", thetaA, thetaM); + printf ("twothetaS = %g \n", twothetaS); + printf ("A3 = %g \n", A3); + printf ("alpha = %g \n", alpha); +%} + +TRACE + +COMPONENT origin = Progress_bar() +AT (0, 0, 0) RELATIVE ABSOLUTE + +COMPONENT source = Source_Maxwell_3( + xwidth=width, + yheight=width, + Lmin=lambda_i-Dlambda/2, + Lmax=lambda_i+Dlambda/2, + dist=7.5, + focus_yh=width, + focus_xw=width, + T1=300, T2=300, T3=300, + I1=1E15, I2=1E15, I3=1E15) +AT (0, 0, 0) RELATIVE PREVIOUS + +COMPONENT l_monitor_before_mono = L_monitor( + nL=200, + filename="L_before_mono", + Lmin=0, + Lmax=4, + xwidth=0.16, + yheight=0.25, + restore_neutron=1) +AT (0, 0, 7) RELATIVE PREVIOUS + +COMPONENT monochromator_flat = Monochromator_flat( + mosaicv=30, + mosaich=30, + zwidth = width, + yheight = width, + Q=QM) +AT (0, 0, 7.5) RELATIVE source +ROTATED (0, thetaM, 0) RELATIVE source + +COMPONENT arm1 = Arm() +AT (0, 0, 0) RELATIVE PREVIOUS +ROTATED (0, thetaM, 0) RELATIVE PREVIOUS + +COMPONENT collimator_linear1 = Collimator_linear( + xwidth=0.1, + yheight=0.2, + length=0.2, + divergence=coll) +AT (0, 0, 0.6) RELATIVE arm1 + +COMPONENT l_monitor_before_sample_big = L_monitor( + nL=300, + filename="L_before_sample", + Lmin=lambda_i - Dlambda, + Lmax=lambda_i + Dlambda, + xwidth = 0.1, + yheight = 0.1, + restore_neutron=1) +AT (0, 0, 0.9) RELATIVE arm1 + +COMPONENT e_monitor_before_sample_big = E_monitor( + nE=100, + filename="E_before_sample", + Emin=Ei-3, + Emax=Ei+3, + xwidth = 0.1, + yheight = 0.1, + restore_neutron=1) +AT (0, 0, 0.01) RELATIVE PREVIOUS + +COMPONENT PSD_monitor_before_sample = PSD_monitor( + nx = 200, + ny = 200, + xwidth = width/2, + yheight = width/2, + filename="PSD_before_sample", + restore_neutron=1) +AT (0, 0, 0.998) RELATIVE arm1 + +COMPONENT arm2 = Arm() +AT (0, 0, 1) RELATIVE arm1 +ROTATED (0, -A3+dA3, 0) RELATIVE arm1 + +SPLIT 10 +COMPONENT phonon_bvk_pg = Phonon_BvK_PG( +// radius=width/2, + xwidth=width/2, zdepth=width/20, + yheight=width/2, + sigma_abs=0, + sigma_inc=0, + DW=1, + T=Temp, + mode_input=phononmode, + target_index=3, + focus_xw=width, + focus_yh=width, + e_steps_low = E_steps_low, + e_steps_high = E_steps_high, + verbose_input=Verbose, + hh = h, kk = 0, ll = l, dispersion = DISP +) +AT (0, 0, 0) RELATIVE PREVIOUS +ROTATED (0, 0, 0) RELATIVE PREVIOUS // To use the (0,k,l) scattering plane, rotate (0, 0, -60); for (h,k,0) rotate (90, 0, 0) + +COMPONENT arm3 = Arm() +AT (0, 0, 0) RELATIVE PREVIOUS +ROTATED (0, -twothetaS, 0) RELATIVE arm1 + +COMPONENT PSD_monitor_after_sample = PSD_monitor( + nx=200, + ny=200, + filename="PSD_after_sample", + restore_neutron=1) +AT (0, 0, 0.9999) RELATIVE arm3 + +COMPONENT slit1 = Slit( + xwidth=width, + yheight=width) +AT (0, 0, 1) RELATIVE arm3 + +COMPONENT e_monitor_beforeana = E_monitor( + nE=750, + filename="E_before_ana", + xwidth=2*width, + yheight=2*width, + Emin=0, + Emax=75) +AT (0, 0, 0.001) RELATIVE PREVIOUS + +COMPONENT e_mon_beforeana_zoom = E_monitor( + nE=750, + filename="E_before_ana_zoom", + xwidth=2*width, + yheight=2*width, + Emin=Ef*0.9, + Emax=Ef*1.1) +AT (0, 0, 0.001) RELATIVE PREVIOUS +COMPONENT analyzer = Monochromator_flat( + zwidth=4*width, + yheight=4*width, + mosaicv=30, + mosaich=30, + Q=QM + ) +AT (0, 0, 0.1) RELATIVE PREVIOUS +ROTATED (0, thetaA, 0) RELATIVE PREVIOUS + +COMPONENT arm4 = Arm() +AT (0, 0, 0) RELATIVE PREVIOUS +ROTATED (0, thetaA, 0) RELATIVE PREVIOUS + +COMPONENT slit2 = Slit( + xwidth=width, + yheight=width) +AT (0, 0, 1) RELATIVE PREVIOUS + +COMPONENT Emon_after_analyzer = E_monitor( + nE=750, + filename="E_after_analyzer", + xwidth=0.05, + yheight=0.10, + Emin=0, + Emax=250) +AT (0, 0, 0.001) RELATIVE PREVIOUS + +COMPONENT Emon_1storder = E_monitor( + nE=50, + filename="E_1st_order", + xwidth=width, + yheight=width, + Emin=Ef*0.9, + Emax=Ef*1.1) +AT (0, 0, 0.001) RELATIVE PREVIOUS + +END diff --git a/mcstas-comps/share/eigen.c b/mcstas-comps/share/eigen.c new file mode 100644 index 0000000000..32d00e3fde --- /dev/null +++ b/mcstas-comps/share/eigen.c @@ -0,0 +1,517 @@ +/************************************************************ + * McStas Eigensolver library * + * Contributed by James Emil Avery, 202x * + * Department of Computer Science, University of Copenhagen * + ************************************************************/ + +/* Original file "eigen.c": */ + +#ifndef EIGEN_LIB +#define EIGEN_LIB + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* TODO: Tridiagonal input/outpout in (3,m) storage. */ +/* TODO: Symmetric tridiagonal input/outpout in (2,m) storage. */ +/* TODO: Add pivot to handle ill-conditioned matrices */ +// #include "eigen.h" + +#define G PRINTF_G +#define ROWS 0 +#define COLS 1 + +/* Fixed-size scratch: abort on the CPU if n exceeds EIGEN_MAXN (see eigen.h) */ +#ifndef OPENACC +#define EIGEN_CHECK_N(n) do { if ((n) > EIGEN_MAXN) { \ + fprintf(stderr, "eigen: matrix size n=%d exceeds EIGEN_MAXN=%d\n", (int)(n), EIGEN_MAXN); \ + exit(-1); } } while (0) +#else +#define EIGEN_CHECK_N(n) +#endif + +/* TOOD: Generic matrix functions to separate file */ +#pragma acc routine seq +INLINE real_t vector_norm2(const cdouble*x, int n){ + real_t sum = 0; + for(int i=0;i numerical_zero) // Only process if subdiagonal element is not already zero. + { + a[0] = D[k]; a[1] = L[k]; // a = T[k:k+2,k] is the vector of nonzeros in kth subdiagonal column. + + real_t anorm = sqrt(a[0]*a[0] + a[1]*a[1]); + + // // Udrullet + // // reflection_vector(a,anorm,v); + v[0] = D[k]; v[1] = L[k]; + real_t alpha = -copysign(anorm,a[0]); // Koster ingenting + v[0] -= alpha; + + real_t vnorm = sqrt(v[0]*v[0]+v[1]*v[1]); + real_t norm_inv = 1/vnorm; /* Normalize */ + v[0] *= norm_inv; v[1] *= norm_inv; + + Vout[2*k] = v[0]; Vout[2*k+1] = v[1]; + + // // Udrullet + // // apply_reflection(T({k,k+2},{k,k+3}),v); + // // if(k+1=0;k--){ + // We start by targeting the (n,n)-eigenvalue, and gradually + // deflate, working on smaller and smaller submatrices. + real_t d = D[k]; // d = T(k,k) + real_t shift = d; + + // The Gershgorin disk radius is defined by just the row-sums of + // absolute off-diagonal elements, since T is symmetric. As T is + // tridiagonal, only T(k,k-1),T(k,k), and T(k,k+1) are nonzero. + // Thus, the k'th Gershgorin radius is just |T(k,k-1)| + + // |T(k,k+1)| = |T(k,k-1)| + |T(k+1,k)| = |L[k-1]|+|L[k]|. + int i=0; + + real_t GR = (k>0?fabs(L[k-1]):0)+fabs(L[k]); // Initial Gershgorin radius + int not_done = 2; + while(not_done > 0){ + i++; + T_QTQ(k+1, D,L, D,L, V, shift); + if(Q!=0) apply_real_reflections(V,k,Qhat,n); + + GR = (k>0?fabs(L[k-1]):0)+(k+10){ + real_pair l = eigvalsh2x2(D[k-1],L[k-1], /* T[(k-1):k, (k-1):k] */ + L[k-1],D[k] ); + real_t l0 = l.value[0], l1 = l.value[1]; + + shift = fabs(l0-d) < fabs(l1-d)? l0 : l1; // Pick closest eigenvalue + } else + shift = D[k]; + + if(GR <= gershgorin_tolerance) not_done--; + if(i>max_iterations && GR > gershgorin_tolerance){ +#ifndef OPENACC + fprintf(stderr,"Cannot converge eigenvalue %d to tolerance " G + " in %d iterations using machine precision " G " (d=" G ", shift=" G ", GR=" G ")\n" + "D[k] = " G ", L[k-1] = " G ", L[k] = " G "\n", + k,gershgorin_tolerance,i, + machine_precision,d,shift,GR, + D[k], (k>0)?L[k-1]:0, (k+1 + +// PW: Non-float64 logic suppressed. + #define real_t double + #define scalar cdouble + #define machine_precision DBL_EPSILON + #define PRINTF_G "%g" + +#ifndef _MSC_EXTENSIONS +#define INLINE inline __attribute__((always_inline)) +#else +#define INLINE inline +#endif + +typedef struct { + real_t value[2]; +} real_pair; + +/* Largest matrix dimension n. Scratch arrays are fixed-size stack arrays of this + size (no heap allocation per call, so the solver can run on an OpenACC device). */ +#ifndef EIGEN_MAXN +#define EIGEN_MAXN 12 +#endif + + +/* End of "types.h" */ +/* ------------------------ */ +/* Original file "eigen.h": */ + +void print_vector(const char *name, cdouble *a, int l); +void print_matrix(const char *name, cdouble *A, int m, int n); + +#pragma acc routine seq +real_t vector_norm(const cdouble *x, int n); +#pragma acc routine seq +void extract_region(cdouble *S, int N, + int i0, int j0, int m, int n, + cdouble *D); +#pragma acc routine seq +double max_norm(cdouble* A, int m, int n); +#pragma acc routine seq +void identity_matrix(cdouble *Q, int n); +#pragma acc routine seq +void real_identity_matrix(double *Q, int n); +#pragma acc routine seq +void matrix_inplace_multiply(cdouble *A, cdouble *B, + int m, int n, int q); + +#pragma acc routine seq +void reflection_vector(/*in*/cdouble *a, double anorm, + /*out*/cdouble *v, cdouble *sigma, int n); +#pragma acc routine seq +void apply_reflection(/*in/out*/cdouble *A, const cdouble *v, + int m, int n, cdouble sigma, int transpose); + +#pragma acc routine seq +void reflect_region(/*in/out*/cdouble *A, int N, + int i0, int j0, int m, int n, + const cdouble *v, cdouble sigma, int cols); + +#pragma acc routine seq +void apply_real_reflections(double *V, int n, double *Q, int m); + + +#pragma acc routine seq +void QHQ(/*in/out*/cdouble *A, int n, cdouble *Q); +#pragma acc routine seq +void T_QTQ(const int n, + const real_t *Din, const real_t *Lin, + real_t *Dout, real_t *Lout, real_t *Vout, + real_t shift); + +#pragma acc routine seq +real_pair eigvalsh2x2(real_t a, real_t b, real_t c, real_t d); + +#pragma acc routine seq +real_pair eigensystem_hermitian(const cdouble *A, int n, + real_t *lambdas, cdouble *Q); + +/* End of eigen.h */ +#endif