diff --git a/pineappl/src/grid.rs b/pineappl/src/grid.rs
index c8341acb..1a4d1d64 100644
--- a/pineappl/src/grid.rs
+++ b/pineappl/src/grid.rs
@@ -868,21 +868,18 @@ impl Grid {
 
     /// Optimize the internal datastructures for space efficiency.
     pub fn optimize(&mut self) {
-        let mut vector: Vec<SubgridEnum> = vec![];
-
-        for subgrid in self.subgrids.iter() {
-            if let SubgridEnum::LagrangeSubgridV1(grid) = subgrid {
-                vector.push(LagrangeSparseSubgridV1::from(grid).into());
-            } else {
-                todo!();
+        for subgrid in self.subgrids.iter_mut() {
+            match subgrid {
+                SubgridEnum::LagrangeSubgridV1(grid) => {
+                    let mut new_subgrid = LagrangeSparseSubgridV1::from(&*grid).into();
+                    mem::swap(subgrid, &mut new_subgrid);
+                }
+                SubgridEnum::LagrangeSparseSubgridV1(_) => {
+                    // nothing to optimize here
+                }
+                SubgridEnum::NtupleSubgridV1(_) => todo!(),
             }
         }
-
-        let shape = self.subgrids.dim();
-        mem::swap(
-            &mut self.subgrids,
-            &mut Array3::from_shape_vec(shape, vector).unwrap(),
-        );
     }
 }
 
diff --git a/pineappl/tests/drell_yan_lo.rs b/pineappl/tests/drell_yan_lo.rs
index 75d5b3ee..95b21536 100644
--- a/pineappl/tests/drell_yan_lo.rs
+++ b/pineappl/tests/drell_yan_lo.rs
@@ -208,6 +208,16 @@ fn dy_aa_lagrange_subgrid_static() -> anyhow::Result<()> {
         assert!(approx_eq!(f64, *result, *reference, ulps = 16));
     }
 
+    // optimize the grid
+    grid.optimize();
+
+    // check that the results are still the same
+    let bins = grid.convolute(&xfx, &xfx, &alphas, &[], &[], &[], &[(1.0, 1.0)]);
+
+    for (result, reference) in bins.iter().zip(reference.iter()) {
+        assert!(approx_eq!(f64, *result, *reference, ulps = 16));
+    }
+
     Ok(())
 }
 
@@ -361,5 +371,95 @@ fn dy_aa_lagrange_subgrid_dynamic() -> anyhow::Result<()> {
         assert!(approx_eq!(f64, *result, *reference, ulps = 16));
     }
 
+    // optimize the grid
+    grid.optimize();
+
+    // check that the results are still the same
+    let bins = grid.convolute(&xfx, &xfx, &alphas, &[], &[], &[], &[(1.0, 1.0)]);
+
+    for (result, reference) in bins.iter().zip(reference.iter()) {
+        assert!(approx_eq!(f64, *result, *reference, ulps = 16));
+    }
+
+    Ok(())
+}
+
+#[test]
+fn dy_aa_lagrange_sparse_subgrid_dynamic() -> anyhow::Result<()> {
+    // suppress LHAPDF banners
+    lhapdf::set_verbosity(0);
+
+    let mut rng = Pcg64::new(0xcafef00dd15ea5e5, 0xa02bdbf7bb3c0a7ac28fa16a64abf96);
+    let mut grid = fill_drell_yan_lo_grid(&mut rng, 500_000, "LagrangeSparseSubgrid", true)?;
+
+    grid.merge(fill_drell_yan_lo_grid(
+        &mut rng,
+        500_000,
+        "LagrangeSparseSubgrid",
+        true,
+    )?)?;
+    grid.scale(0.5);
+
+    let pdf_set = "NNPDF31_nlo_as_0118_luxqed";
+
+    assert!(lhapdf::available_pdf_sets().iter().any(|x| x == &pdf_set));
+
+    let pdf = Pdf::with_setname_and_member(&pdf_set, 0);
+    let xfx = |id, x, q2| pdf.xfx_q2(id, x, q2);
+    let alphas = |_| 0.0;
+
+    // check `read` and `write`
+    let mut file = Cursor::new(Vec::new());
+    grid.write(&mut file)?;
+    file.set_position(0);
+    mem::drop(grid);
+    let mut grid = Grid::read(&mut file)?;
+
+    // some useless scalings
+    grid.scale_by_order(10.0, 0.5, 10.0, 10.0, 1.0);
+    grid.scale_by_order(10.0, 1.0, 10.0, 10.0, 4.0);
+
+    let bins = grid.convolute(&xfx, &xfx, &alphas, &[], &[], &[], &[(1.0, 1.0)]);
+    let reference = vec![
+        5.093090431949207e-1,
+        5.191668797562395e-1,
+        5.467930909144878e-1,
+        4.845000146366871e-1,
+        4.7279670245192884e-1,
+        4.7481525429115445e-1,
+        4.7060007393610065e-1,
+        4.286009571276975e-1,
+        4.2997427448811987e-1,
+        3.911121604214181e-1,
+        3.430494924416351e-1,
+        3.1204501485640446e-1,
+        2.6656633773109056e-1,
+        2.3745772514449243e-1,
+        2.0055415662593068e-1,
+        1.692229208614707e-1,
+        1.4635749293124972e-1,
+        1.1348804559115269e-1,
+        8.934261002200886e-2,
+        6.385631160399488e-2,
+        4.866968827833745e-2,
+        3.379282482821324e-2,
+        1.9494720220040448e-2,
+        6.880478270711603e-3,
+    ];
+
+    for (result, reference) in bins.iter().zip(reference.iter()) {
+        assert!(approx_eq!(f64, *result, *reference, ulps = 16));
+    }
+
+    // optimize the grid
+    grid.optimize();
+
+    // check that the results are still the same
+    let bins = grid.convolute(&xfx, &xfx, &alphas, &[], &[], &[], &[(1.0, 1.0)]);
+
+    for (result, reference) in bins.iter().zip(reference.iter()) {
+        assert!(approx_eq!(f64, *result, *reference, ulps = 16));
+    }
+
     Ok(())
 }