cytnx::LinOp Class Reference

#include <LinOp.hpp>

Public Member Functions
	LinOp (const std::string &type, const cytnx_uint64 &nx, const int &dtype=Type.Double, const int &device=Device.cpu)
	Linear Operator class for iterative solvers.
template<class T >
void	set_elem (const cytnx_uint64 &i, const cytnx_uint64 &j, const T &elem, const bool check_exists=true)
Tensor::Tproxy	operator() (const cytnx_uint64 &i, const cytnx_uint64 &j)
void	set_device (const int &device)
void	set_dtype (const int &dtype)
int	device () const
int	dtype () const
cytnx_uint64	nx () const
void	_print ()
virtual Tensor	matvec (const Tensor &Tin)

Constructor & Destructor Documentation

LinOp()

cytnx::LinOp::LinOp ( const std::string &  type, const cytnx_uint64 &  nx, const int &  dtype = Type.Double, const int &  device = Device.cpu  )

inline

Linear Operator class for iterative solvers.

Parameters

type	the type of operator, currently it can only be "mv" (matvec) or "mv_elem" (matvec with pre-store element)
nx	the last dimension of operator, this should be the dimension of the input vector when "mv_elem" is used.
dtype	the Operator's dtype. Note that this should match the input/output Tensor's dtype.
device	the Operator's on device.

Note:

1. the device and dtype should be set. This should be the same as the input and output vectors. by default, we assume custom_f take input and output vector to be on CPU and Double type.

Details:

The LinOp class is a class that defines a custom Linear operation acting on a Tensor or UniTensor. To use, inherit this class and override the matvec function.
See the following examples for how to use them.

Example:

c++ API:

output>

python API:

from cytnx import *
 
 
# LinOp class provides a base class that defines the operation on a vector.
# This class or it's derived class are required 
# for using the cytnx's iterative solver such as Lanczos and Arnodi. 
 
#-----------------------------------------
# Suppose we want to define a custom linear operation 
# acting on input vector t (dim=4) that swap the first and last element 
# and add the 2nd and 3rd element with one. 
 
t = arange(4)
print(t)
 
# Method 1, write a custom function, and assign into the LinOp class
# -----------------------------------
# [Note] the function should have the signature 
#        Tensor f(const Tensor&) as in C++
def myfunc(v):
    out = v.clone();
    out[0], out[3] = v[3], v[0]; #swap
    out[1]+=1 #add 1
    out[2]+=1 #add 1
    return out;
 
 
 
lop = LinOp("mv",nx=4,dtype=Type.Double,device=Device.cpu,custom_f=myfunc)
print(lop.matvec(t)) 
 
 
# Method 2, write a custom class that inherit LinOp class.
# ------------------------------------
# [Note] Instead of writing a custom class, we overload the matvec() directly.
#        Inheritance is handy if there are additional parameters/arguments 
#        needed for the custom operation.
class MyOp(LinOp):
    AddConst = 1 # let's make it a class member.
 
    def __init__(self,typ,nx,aconst):
        
        LinOp.__init__(self,typ,nx,Type.Double,Device.cpu) 
        self.AddConst = aconst
 
    def matvec(self,v):
        out = v.clone()
        out[0],out[3] = v[3],v[0]
        out[1]+=self.AddConst #add the constant
        out[2]+=self.AddConst #add the constant
        return out
 
mylop = MyOp("mv",nx=4,aconst=3) # let's add 3 instead of one. 
print(mylop.matvec(t)) 
 
 
 
 
 
 
 
 

output>

Total elem: 4
type  : Double (Float64)
cytnx device: CPU
Shape : (4)
[0.00000e+00 1.00000e+00 2.00000e+00 3.00000e+00 ]




Total elem: 4
type  : Double (Float64)
cytnx device: CPU
Shape : (4)
[3.00000e+00 2.00000e+00 3.00000e+00 0.00000e+00 ]




Total elem: 4
type  : Double (Float64)
cytnx device: CPU
Shape : (4)
[3.00000e+00 4.00000e+00 5.00000e+00 0.00000e+00 ]

Member Function Documentation

_print()

void cytnx::LinOp::_print

(

)

device()

int cytnx::LinOp::device ( ) const

inline

dtype()

int cytnx::LinOp::dtype ( ) const

inline

matvec()

UniTensor cytnx::LinOp::matvec ( const Tensor &  Tin )

virtual

nx()

cytnx_uint64 cytnx::LinOp::nx ( ) const

inline

operator()()

Tensor::Tproxy cytnx::LinOp::operator() ( const cytnx_uint64 &  i, const cytnx_uint64 &  j  )

inline

set_device()

void cytnx::LinOp::set_device ( const int &  device )

inline

set_dtype()

void cytnx::LinOp::set_dtype ( const int &  dtype )

inline

set_elem()

template<class T >

void cytnx::LinOp::set_elem ( const cytnx_uint64 &  i, const cytnx_uint64 &  j, const T &  elem, const bool  check_exists = true  )

inline

---

The documentation for this class was generated from the following files:

• include/LinOp.hpp
• src/LinOp.cpp