"Out of Memory" whith calcTensor on large EBSD files

[MicAtWork]

Im trying to calculate the stiffnes tensor for some large EBSD files I have. For the files larger than ~690 MB I get "out of memory" in Matlab. Ive tried increasing the memory by setMTEXpref('memory',32000*1024); but it doesnt help.

This is the code I use

% crystal symmetry
    CS = {...
        'notIndexed',...
        crystalSymmetry('m-3m',  [2.8660 2.8660 2.8660], 'mineral', 'Iron BCC', 'color', [255 53 35]/255),...
        crystalSymmetry('m-3m',  [3.6599 3.6599 3.6599], 'mineral', 'Iron FCC', 'color', 'blue')};

% load the data
    ebsd = EBSD.load(FileName,CS,'interface','crc',...
          'convertEuler2SpatialReferenceFrame'); % giving 6836x7116 EBSDsquare 


    o = ebsd(phase).orientations; % resulting in 34426708x1 orientations

    % Define matrix
        M = [
                [C11 C12 C12 0 0 0];
                [C12 C11 C12 0 0 0];
                [C12 C12 C11 0 0 0];
                [0 0 0 C44 0 0];
                [0 0 0 0 C44 0];
                [0 0 0 0 0 C44];
            ];
    C = stiffnessTensor(M, o.CS,'density',MatDensity);
    [CVoigt, CReuss, CHill] = calcTensor(o, C);

The Error Message I get:

Out of memory.
Error in tensor/EinsteinSum (line 81)
    M1 = M1 .* M2;
Error in tensor/rotate (line 31)
  T = EinsteinSum(T,ind,R,[d -d],'keepClass');
Error in  .*  (line 32)
  r = rotate(b,a);
Error in orientation/calcTensor (line 31)
[varargout{1:nargout}] = mean(ori .* T,varargin{:});

The other files are less than 470 MB and work fine for the calculation of the stiffnessmatrix

Using MTEX version 5.10.2 with Matlab 2023b update 4, on Win 11, on a computer with 13th intel i5 and 64 GB RAM, with java heap memory set to maximum.

would it be possible to implement some sort of tall-array calculation for heavy files? or split the calculation in subsets of the orientations and then average the results?

[zmichels]

Those certainly are large datasets.

Not a solution to the actual problem... but wondering if you might entertain using reduce() as a workaround. If you felt like a spatially downsampled version of your map may yield a satisfactory representative stiffness tensor, then you could load your data and subset every other point (or some other factor) to start from a map with a lower point-density and larger effective step-size than your original.

reduce() will output an EBSD dataset with a factor reduction in the number of points. So... reduce(ebsd, 2) yields a map with every other point from the original. Use a greater number factor for greater reduction.

[MicAtWork]

Thanks! that work for now, Another solution was to use a random subset of orientations:
odf_subset = calcOrientations(o, 1000); % 1000 random points, change to any desired number