[Codescan] Grow the basin attractor table instead of relying on a fixed 5 percent heuristic

[Stardust0831]

Global scan provenance

This issue was found by the Codex global repository scan of non-unit-test files in Stardust0831/Multiwfn, pinned to commit d2b770a531eb7039153198d04520b5992f15c49d.

Problem

generatebasin() allocates the attractor coordinate table with capacity nint(nx*ny*nz/20D0), assuming attractors can never exceed five percent of grid points.

Multiwfn/basin.f90

Lines 860 to 870 in d2b770a

	use basinintmod
	implicit real*8 (a-h,o-z)
	integer,parameter :: nmaxtrjgrid=3000
	integer igridmethod
	integer ntrjgrid !Recording how many members are contained in trjgrid now
	integer trjgrid(3,nmaxtrjgrid) !The trajectory contains which grids, record their indices sequentially, trjgrid(1/2/3,i) = ix/iy/iz of the ith grid
	! real*8 trjval(nmaxtrjgrid),gradmaxval(nmaxtrjgrid) !******For debugging******
	if (allocated(attgrid)) deallocate(attgrid)
	allocate(attgrid(3,nint(nx*ny*nz/20D0))) !I think the number of attractors in general is impossible to exceeds nx*ny*nz/20
	numatt=0
	write(*,*) " Attractor X,Y,Z coordinate (Angstrom) Value"

New attractors increment numatt and write attgrid(:,numatt) without checking that the heuristic capacity is nonzero or still sufficient:

Multiwfn/basin.f90

Lines 946 to 961 in d2b770a

	end if
	exit cyciatt
	end if
	end do
	end do cyciatt
	!Found a new attractor. Add to list, and assign the grids in the current trajectory
	if (iatt==numatt+1) then
	if (itime==1) then
	numatt=numatt+1
	do itrjgrid=1,ntrjgrid
	gridbas(trjgrid(1,itrjgrid),trjgrid(2,itrjgrid),trjgrid(3,itrjgrid))=numatt
	end do
	attgrid(1,numatt)=inowx
	attgrid(2,numatt)=inowy
	attgrid(3,numatt)=inowz
	call getgridxyz(inowx,inowy,inowz,tmpx,tmpy,tmpz)

The basin grid dimensions are user-configurable, so very small grids can round the allocation down to zero, while noisy or synthetic scalar fields can contain more local attractors than the undocumented five-percent assumption.

Multiwfn/basin.f90

Lines 1688 to 1705 in d2b770a

	if (igridsel==4) dx=spclunaqual
	if (igridsel==5) then
	write(*,*) "Input the grid spacing (Bohr), e.g. 0.08"
	read(*,*) dx
	end if
	dy=dx
	dz=dx
	nx=nint(molxlen/dx)+1
	ny=nint(molylen/dy)+1
	nz=nint(molzlen/dz)+1
	else if (igridsel==6) then
	write(*,*) "Input the number of grid points in X,Y,Z direction, e.g. 139,59,80"
	read(*,*) nx,ny,nz
	dx=molxlen/(nx-1)
	dy=molylen/(ny-1)
	dz=molzlen/(nz-1)
	else if (igridsel==7) then
	write(*,*) "Input X,Y,Z coordinate of original point (Bohr), e.g. 0.1,4,-1"

Impact

The first detected attractor on a zero-capacity table, or any attractor beyond the heuristic capacity, writes out of bounds. Bounds-checked builds terminate; optimized builds may corrupt adjacent memory and later produce incorrect basin assignments.

Suggested direction

• Allocate a documented minimum capacity and grow attgrid geometrically when numatt reaches capacity.
• Check multiplication and conversion for integer overflow before sizing from nx*ny*nz.
• Alternatively collect attractors in a dynamically growing container and materialize the final array afterward.
• Add tests for fewer than 20 grid points and for an adversarial field with more than five percent local attractors.