slimgroup / waveform Goto Github PK

View Code? Open in Web Editor NEW

13.0 9.0 7.0 457 KB

Framework for large-scale non-linear inverse problems

License: MIT License

MATLAB 85.02% C 14.52% Makefile 0.42% M 0.04%

waveform's Introduction

WAVEFORM - a softWAre enVironmEnt For nOnlinear inveRse probleMs

Waveform is a flexible and modular approach to solving PDE-constrained inverse problems that aims to

accurately reflect the underlying mathematics of the problem
produce readable code that reflects the underlying mathematics
scales easily from small 2D problems to large-scale 3D problems with minimal code modifications
allow users to integrate their own preconditioners, stencils, linear solvers, etc. easily in to the entire inversion framework

For more details, as well as the full design of the software, see (https://arxiv.org/abs/1703.09268). All functions are written by Curt Da Silva, unless indicated otherwise in the function documentation.

Requirements

Waveform requires Matlab (tested on R2015b), the Parallel Toolbox (for using parallelism), and the following packages:

SPOT: A linear-operator toolbox for Matlab, available at https://github.com/curtd/spot
- Note if you're using the original SPOT toolbox from https://github.com/mpf/spot, you'll need to modify the first line of the file
```
  [SPOT directory]/@opSpot/opSpot.m
```
by replacing
```
classdef opSpot
```
with
```
classdef (HandleCompatible) opSpot
```
pSPOT: Parallel extensions to SPOT, available at https://github.com/slimgroup/pSPOT

Optional packages, for running certain examples

Curvelab, for the Curvelet transform, available at http://www.curvelet.org/

Installation

At the terminal, type

git clone [email protected]:slimgroup/WAVEFORM.git
matlab

In the Matlab console, once you have the prerequisite packages installed in your path, type

addpath(genpath('WAVEFORM'));
compile_waveform_mex;

Examples

The examples from the attendant paper, as well as some other scripts demonstrating how to use the framework, can be found in the /examples/ subdirectory.

Citing

If you find this software useful in your own work, please cite

"A Unified 2D/3D Large Scale Software Environment for Nonlinear Inverse Problems", Curt Da Silva and Felix J. Herrmann, 2017.

waveform's People

Contributors

Stargazers

Watchers

Forkers

yuanyuxin0077 dongqianbei bimo1212 hongtao-argmin xushang zhouzhuhuang

waveform's Issues

Issue with returning Hessian

Hi!

I am trying to obtain Hessian by [f,g,h] = obj(m) where obj = misfit_setup(m0,Q,D,model_r,opts). However, the computation is failed due to incompatibility of the size of matrix multiplication. I tried to debug the code, and I found out the error lies in the PDEfunc.m line 266 or 268 from the function handler 'T' in

    if strcmp(func,HESS) 
        [H,comp_grid,T,DT_adj] = discrete_helmholtz(v,model,freq(k),params);        
    else
        [H,comp_grid,T] = discrete_helmholtz(v,model,freq(k),params);
    end

This leads to error in line 368 for the Hessian computation.

case HESS_GN
            dm = to_comp*vec(input);
            dU = H\(-T(U)*dm); % Error HERE!
            dU = H'\(-Pr'*Pr*dU);
            output = output + sum_srcs(T(U)'*dU);

Below are the errors for your reference. Kindly seek your assistance in this matter.

Great thanks!

Unable to perform assignment because the indices on the left side are not compatible with the size of the
right side.

Error in opSpot/applyMultiply (line 96)
                    y(:,i) = op.multiply(x(:,i), mode);

Error in  *  (line 73)
         y = A.applyMultiply(B,1);

Error in PDEfunc (line 368)
            dU = H\(-T(U)*dm);

Error in PDEfunc_dist (line 221)
        dm = PDEfunc(func,v,Q,input,[],model,params,freqsxsy);

Error in oppHGN/multiply (line 67)
           y = PDEfunc_dist(PDEopts.HESS_GN,op.mt,op.Q,x,[],op.model,op.params);

Error in opSpot/applyMultiply (line 96)
                    y(:,i) = op.multiply(x(:,i), mode);

Error in opFoG/multiply (line 120)
              y = applyMultiply(op.children{2},x,mode);

Error in opSpot/applyMultiply (line 96)
                    y(:,i) = op.multiply(x(:,i), mode);

Error in  *  (line 73)
         y = A.applyMultiply(B,1);

Scaling issue when changing the gridsize

Hi Curt! I hope you are doing well. I noticed an issue with scaling when changing the gridsize. It seems that it should be corrected in line 298 of PDEfunc, however when model.d and params.dt are the same, not scaling happens. To make the wavefield amplitude invariant under changes of the gridspacing it should be scaled by 1/prod(model.d). Is that what you intend in this line? See example code attached.

Thanks and best regards,
Tristan

test_grid_scalling.m.zip

;

I got an error while I run this script in Matlab console.

matrix_mode='explicit';
params=struct;
params.coef=[1 2;3 4];
solve_opts=LinSolveOpts();
A=opAbstractMatrix(matrix_mode,params,solve_opts);

slimgroup / waveform Goto Github PK

waveform's Introduction

WAVEFORM - a softWAre enVironmEnt For nOnlinear inveRse probleMs

Requirements

Installation

Examples

Citing

waveform's People

Contributors

Stargazers

Watchers

Forkers

waveform's Issues

Issue with returning Hessian

Scaling issue when changing the gridsize

I got an error while I run this script in Matlab console.

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent