Fix P::LE::NR2::Field, as it modified haydock coeffs. inadvertently

Changes

@@ -1,3 +1,5 @@
+- Modify P::LE::NR2::Field so as not to modify Haydock coefficients
+  inadvertently
 0.023 2024-10-02
 - Add tests for modules ST
 - Add tensorial modules ST, to deal with anisotropic components

TODO

@@ -1,5 +1,12 @@
 # -*- mode: org; -*-
 
+* TODO [2024-12-04 Wed] Field modifies silently haydock coefficients
+  We found P::LE::NR2::Field uses cgtsv, a routine that modifies its
+  arguments. We used to pass a subdiagonal and supradiagonal which are
+  both given by the b coefficients. We used the same ndarray, so when
+  the subdiagonal is modified, the supradiagonal is erroneously
+  modified. The results were wrong.
+* TODO [2024-11-01 Fri] It seems WE::S::Green doesn't compute antisymmetric part. Fix and test.
 * DONE [2021-09-30 Sun] Add example/test with non Cartesian coordinates
 * TODO [2021-09-25 Sat] Check if it is clear that complexCoeffs is equivalent to non is_hermitian
 * DONE Return as, bs, b2s as pdl instead of array?

lib/Photonic/LE/NR2/Field.pm

@@ -135,8 +135,8 @@ sub _build_field {
     # At first take D=|0>
     my $diag=$u - $as->(0:$nh-1);
     # rotate complex zero from first to last element.
-    my $subdiag=-$bs->(0:$nh-1)->rotate(-1)->r2C;
-    my $supradiag=$subdiag;
+    my $subdiag=-$bs->(0:$nh-1)->copy->rotate(-1)->r2C;
+    my $supradiag=$subdiag->copy;
     my $rhs=PDL->zeroes($nh);
     $rhs->((0)).=1;
     $rhs=$rhs->r2C;