libscuff API Documentation: Utility routines

libscuff provides some simple classes called HMatrix and HVector for working with numerical matrices and vectors.

This page documents a representative subset of the matrix and vector functions provided; for the full API,
consult the file include/scuff-em/libhmat.h that comes with the scuff-em distribution.

1. Creating from scratch

Create real-valued matrices and vectors of known dimensions:

  HVector *V = new HVector( 13);   // create a 13-component real-valued vector

HMatrix *M = new HMatrix( 5, 6); // create a 5x6 real-valued matrix


Create complex-valued matrices and vectors:

  HVector *V = new HVector( 13, LHM_COMPLEX);

HMatrix *M = new HMatrix( 5, 6, LHM_COMPLEX);


For symmetric or Hermitian matrices, you have the option of using packed storage; this roughly halves the RAM needed to store your matrix, but has the drawback of slowing down linear algebra operations.

  // real symmetric matrix, M_{ij} = M_{ji}
HMatrix *M = new HMatrix( 5, 5, LHM_REAL, LHM_SYMMETRIC );

// complex hermitian matrix, M_{ij} = M^*_{ji}
HMatrix *M = new HMatrix( 5, 5, LHM_COMPLEX, LHM_HERMITIAN );

// complex symmetric matrix, M_{ij} = M_{ji}
HMatrix *M = new HMatrix( 5, 5, LHM_COMPLEX, LHM_SYMMETRIC );


2. Importing from text or binary files

Text files:

Create a new vector or matrix by importing a list of numbers in an ASCII text file:

  HVector *V = new HVector("MyVector.dat", LHM_TEXT );
if (V->ErrMsg)
ErrExit(V->ErrMsg);

HVector *M = new HMatrix("MyMatrix.dat", LHM_TEXT );
if (M->ErrMsg)
ErrExit(M->ErrMsg);


Here MyVector.dat should be a file with a single number per line, while MyMatrix.dat may have multiple numbers per line. Complex numbers are allowed. Blank lines and comments (lines beginning with a pound sign #) are skipped. The dimension of the resulting HVector, and the number of rows of the resulting HMatrix, will be the number of non-blank non-comment lines in the file, while the number of columns of the HMatrix will be the largest number of numbers read from any one line.

Note that, if the file import operation fails, the constructor returns an HVector or HMatrix whose ErrMsg field points to a nonempty error message. (In this case, all other class fields in the object should be assumed to be invalid.)

If the operation is successful, then ErrMsg will be NULL on return.

HDF5 files:

Create a new vector or matrix by importing from an HDF5 binary data file:

   HMatrix *M1 = new HMatrix("MyFile.hdf5", LHM_HDF5, "M1");
HMatrix *M2 = new HMatrix("MyFile.hdf5", LHM_HDF5, "M2");
HVector *V  = new HVector("MyFile.hdf5", LHM_HDF5, "V");

if (M1->ErrMsg) ErrExit(M1->ErrMsg);
if (M2->ErrMsg) ErrExit(M2->ErrMsg);
if (V->ErrMsg)  ErrExit(V->ErrMsg);


Note that the third parameter to the constructor here is the label of the dataset to within the HDF5 file.

3. Exporting to text or binary files

Text files:

Write the contents of a vector or matrix to an ASCII text file:

  M->ExportToText("MyMatrix.dat");
V->ExportToText("MyVector.dat");


The argument to ExportToText supports printf-like semantics for inserting numbers, etc. into the file name:

  int N = 3;
M->ExportToText("Matrix_%i.dat", N);


HDF5 files:

There are two calling conventions for exporting matrices and vectors to HDF5 files.

If you want each matrix and vector to be exported to a separate HDF5 file, you can say simply

  M->ExportToHDF5("MyMatrix.hdf5","M");
V->ExportToHDF5("MyVector.hdf5","V");


Alternatively, you can write multiple matrices and vectors to a single HDF5 file.

  void *HC = HMatrix::OpenHDF5Context("MyFile.hdf5");
M->ExportToHDF5(HC, "M");
V->ExportToHDF5(HC, "V");
HMatrix::CloseHDF5Context(HC);


4. Simple manipulations

Note: All indices are zero-based.

Get or set single entries:

   HMatrix *M = new HMatrix(M, N, LHM_COMPLEX);
HVector *V = new HVector(N);

M->SetEntry( 3, 4, 5.6 );
M->SetEntry( 3, 4, cdouble(5.6+7.8) );

V->SetEntry( 7, 8.009 );

double D;
cdouble Z;

Z = M->GetEntry( 3, 4 );
D = M->GetEntryD( 3, 4 ); // discard any imaginary part

Z = V->GetEntry(7);
D = V->GetEntryD(8);


Get or set blocks of entries in an HMatrix:

  HMatrix *M = new HMatrix(10, 5, LHM_COMPLEX);

cdouble Col3;
M->GetEntriesD(":",3,Col3);

cdouble M2_14;
M->GetEntriesD(2,"14",M2_14);


Augment individual entries:

   M->AddEntry( 3, 4, 5.6 );  // the (3,4) entry gets increased by 5.6
V->AddEntry( 0, cdouble(0.0, 2.3) ); // the 0th entry gets increased by 2.3i


Replace a matrix with its conjugate or non-conjugate tranpose (no distinction for real-valued matrices):

   M->Adjoint();    // conjugate transpose
M->Transpose();  // non-conjugate transpose


5. Numerical linear algebra: LAPACK/BLAS

Multiply two matrices:

  int P, Q, R;
A=new HMatrix(P, Q);
B=new HMatrix(Q, R);
C=new HMatrix(P, R);
...
A->Multiply(B, C); // set C = A*B
A->Multiply(C, B, "--TransA T"); // set B = transpose(A) * C


Replace a matrix with its LU factorization:

   M->LUFactorize();


Solve a single linear system using an LU-factorized matrix:

   M=new HMatrix(N, N);
// insert code to fill in M
M->LUFactorize();

V=new HVector(N);
// insert code to fill in V
M->LUSolve(V);   // replaces V with M^{-1} * V


Of course, we only need to call LUFactorize once for a given matrix, after which we can make any number of calls to LUSolve() with different vectors.

We can also solve multiple simultaneous systems by passing an HMatrix instead of an HVector to
LUSolve:

   M=new HMatrix(N, N);
// insert code to fill in M
M->LUFactorize();

R=new HMatrix(N, M);
// insert code to fill in the M columns of R
M->LUSolve(R);   // replaces R with M^{-1} * R