Red blood cell tests

git clone https://github.com/sonwell/ib.cu.git

Collision tests

Each of the following use a PHS (r7) RBF to construct the interpolant. A fictitious force is applied to each cell to drive them into/past one another. The cells are constructed using 1,600 data sites and 15,000 sample sites.

Vertical-vertical collision test

2 red blood cells, initially aligned vertically. The simulation terminated due to timestep restrictions at t ≈ 4.7 ms.

Horizontal-vertical collision test

2 red blood cells, one initially aligned horizontally and the other vertically. The simulation terminated due to timestep restrictions at t ≈ 1.0 ms. The video is very brief, so changing the playback speed to 0.25× might help.

Horizontal-horizontal collision test

2 red blood cells, initially aligned horizontally. The simulation terminated due to timestep restrictions at t ≈ 3.23 ms.

Sliding test

2 red blood cells, initially aligned obliquely. The simulation terminated due to timestep restrictions at t ≈ 2.5 ms.

Whole blood tests

Linear model + Skalak law

12 red blood cells in shear flow with shear rate ɣ̇ = 1000 s-1. The red blood cells use a piecewise linear interpolant with 15,000 points and generate forces according to the Skalak law* with shear modulus E = 2.5×10-3 dyn/cm and bulk modulus G = 2.5×10-1 dyn/cm. I terminated the simulation manually at t = 40 ms.

* I was lazy and areas are approximately 3+𝒪(Δ) larger than they should be. Δ is the difference in reference area across edges.

IMQ RBF model + Skalak law + bending energy

12 red blood cells in shear flow with shear rate ɣ̇ = 1000 s-1. The red blood cells use an IMQ RBF interpolant with 1,600 data sites and 15,000 sample sites and generate forces according to the Skalak law with shear modulus E = 2.5×10-3 dyn/cm and bulk modulus G = 2.5×10-1 dyn/cm and bending energy with bending modulus κ = 2×10-12 erg. While not pictured, there is a bumpy endothelial layer below the cells, which is tethered in place. The simulation terminated due to timestep restrictions at t ≈ 10.3 ms.

PHS RBF model + Skalak law + bending energy

12 red blood cells in shear flow with shear rate ɣ̇ = 1000 s-1. The red blood cells use a PHS (r7) RBF interpolant with 1,600 data sites and 15,000 sample sites and generate forces according to the Skalak law with shear modulus E = 2.5×10-3 dyn/cm and bulk modulus G = 2.5×10-1 dyn/cm and bending energy with bending modulus κ = 2×10-12 erg. While not pictured, there is a bumpy endothelial layer below the cells, which is tethered in place. The simulation terminated due to timestep restrictions at t ≈ 9.5 ms.

Investigations in rough force smoothing

Let F be the force on the surface of the cell. Let 𝒮 be the spreading operator and its transpose, 𝒮, be the interpolation operator. The idea is to get the "rough" part of F by computing
Frough = (I - 𝒮𝒮)F.
The video below plots the rough part of the force on the cell surface.

12 red blood cells in shear flow with shear rate ɣ̇ = 1000 s-1. The red blood cells use a PHS (r7) RBF interpolant with 1,600 data sites and 15,000 sample sites and generate forces according to the Skalak law with shear modulus E = 2.5×10-3 dyn/cm and bulk modulus G = 2.5×10-1 dyn/cm and bending energy with bending modulus κ = 2×10-12 erg. While not pictured, there is a bumpy endothelial layer below the cells, which is tethered in place. The simulation terminated due to timestep restrictions at t ≈ 9.5 ms.

Wendland's compactly supported RBFs

Since r = ‖x-y‖, ∇r = r-1 ∇(x-y). So for RBF φ,

φ = r-1 φ'(r) ∇(x-y) ≔ 𝒟φ(r) ∇(x-y).
Wendland defines
φ(r) = 1r tφ(t) dt,
so that 𝒟ℐφ = -φ. The Wendland RBFs are constructed by repeated application of ℐ:
φk(r) = ℐk(1-r)+,
where (x)+ = x if x > 0 and 0 otherwise. For an object embedded in ℝd, φk is positive definite for ≥ ⌊d/2⌋+k+1. The parameter k controls the degree of smoothness of the RBF at r = 1. The smallest k I consider is therefore 2.

The images below depict the forces on 12 cells in a translated and rotated, but otherwise reference, configuration. The only forces present should be due to bending.

Skalak law forces of 12 RBCs reconstructed with 5th order spherical harmonics
Bending forces of 12 RBCs reconstructed with 5th order spherical harmonics
(a) Skalak and (b) bending forces on 12 translated and rotated RBCs, otherwise in their reference configuration. The cells are reconstructed with 5th order spherical harmonics, to represent the truth. Since the deformations are rigid body motions, the cells generate no tension.
Skalak law forces of 12 RBCs reconstructed with Wendland-4,2
Bending forces of 12 RBCs reconstructed with Wendland-4,2
(a) Skalak and (b) bending forces on 12 translated and rotated RBCs, otherwise in their reference configuration. The cells are reconstructed with the φ4,2 Wendland RBF. Since the deformations are rigid body motions, the cells should generate no tension, but appear to do so.
Skalak law forces of 12 RBCs reconstructed with Wendland-7,5
Bending forces of 12 RBCs reconstructed with Wendland-7,5
(a) Skalak and (b) bending forces on 12 translated and rotated RBCs, otherwise in their reference configuration. The cells are reconstructed with the φ7,5 Wendland RBF. Since the deformations are rigid body motions, the cells generate no tension.
Skalak law forces of 12 RBCs reconstructed with Wendland-7,5
Bending forces of 12 RBCs reconstructed with Wendland-7,5
(a) Skalak and (b) bending forces on 12 translated and rotated RBCs, otherwise in their reference configuration. The cells are reconstructed with the φ7,5 Wendland RBF with its argument scaled by 2. Since the deformations are rigid body motions, the cells should generate no tension. However, they do, and the forces in (a) are scaled down by a factor of 1,000 compared to (b) and to every other figure in this section.
Skalak law forces of 12 RBCs reconstructed with the 7th order polyharmonic spline
Bending forces of 12 RBCs reconstructed with the 7th order polyharmonic spline
(a) Skalak and (b) bending forces on 12 translated and rotated RBCs, otherwise in their reference configuration. The cells are reconstructed with the 7th order polyharmonic spline. Since the deformations are rigid body motions, the cells should generate no tension.

Reduced domain tests

2 RBCs in an 8μm × 8μm × 16μm domain.