Numerical Methods in Cancer ModelsNumerical Methods in Cancer Models Doron Levy Department of...
Transcript of Numerical Methods in Cancer ModelsNumerical Methods in Cancer Models Doron Levy Department of...
Motivation
Properties
numerical methods
Stability in the delays space
Numerical Methods in Cancer Models
Doron Levy
Department of Mathematics
and
Center for Scientific Computation and Mathematical Modeling (CSCAMM)
University of Maryland, College Park
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Plan
1 What is cancer?2 Delayed di↵erential equations3 Agent-based models4 PDEs
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Delayed Di↵erential Equations in Cancer ModelsAnalysis & Numerical Methods
Doron Levy
Department of Mathematics
and
Center for Scientific Computation and Mathematical Modeling (CSCAMM)
University of Maryland, College Park
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Outline
1 MotivationCancer ImmunologyStem Cell Transplantation
2 PropertiesZero CrossingsTime Scales
3 Numerical Methods
4 Stability in the delays space
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Outline
1 MotivationCancer ImmunologyStem Cell Transplantation
2 PropertiesZero CrossingsTime Scales
3 Numerical Methods
4 Stability in the delays space
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
What is leukemia?
Normal cells: stem cells turn intomature cells
Leukemia: A malignanttransformation of a stem cell or aprogenitor cell
Myeloid or Lymphocytic
Acute or Chronic
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
CML
3 phasesChronic: uncontrolled proliferationAcceleratedAcute: Uncontrolled proliferations. Cellsdo not mature
Philadelphia chromosomeTranslocation (9;22)Oncogenic BCR-ABL gene fusionThe ABL gene expresses a tyrosinekinase. Growth mechanismsEasy to diagnoseDrug targeting this genetic defect (atyrosine kinase inhibitor)
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Treating leukemia
Chemotherapy
Bone Marrow or Stem Cell transplantChemo + radiotherapy +transplantation
Imatinib (Gleevec)Molecular targeted therapy -suppresses the corrupted controlsystem$32K-$98K/year
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Problems with existing therapies
Remission vs. Cure: Can CML be cured?Yes! but only with a bone marrow (or stem cell) transplantRequires a (matching) donorA risky procedure (+ unpredictable side e↵ects)
Imatinib? Does not cure the disease: stopping it causes a relapse
New Medical Data: There is an anti-leukemia immune response (Leelab)
The strength and dynamics of the specific anti-leukemia immuneresponse can be measured
Number of cellsActivity (count signaling molecules)
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
The observed immune response
A di↵erent immune response for each patient. However:
At the beginning of the treatment: no immune response
Peak: around 6-12 months (after starting the drug treatment)
Later: waning immune response
Question:
What is the relation between the dynamics of the cancer, the drug, andthe immune response?
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
A mathematical model
Leukemicstem cell (y0)
Progenitorcell (y1)
Differentiatedcell (y2)
Terminalcell (y3)
ay(ay’ ) b y(by’ ) c y(cy’ )
divider y
d0 + qCp(C,T) d1 + qCp(C,T) d2 + qCp(C,T) d3 + qCp(C,T)
sT
dT
n!
T cells (T)p0e-cnCkC
x2n
Ingredients:Leukemia cells: stem cells, , fully functional cellsMutations, Drug (Imatinib), Anti leukemia immune response
Kim, Lee, Levy: PLoS Computational Biology, ’08
Michor et al. (Nature 05). Cronkite and Vincent (69), Rubinow (69), Rubinow & Lebowitz (75), Fokas,
Keller, and Clarkson (91), Mackey et al (99,...), Neiman (00), Moore & Li (04), Michor et al (05), Komarova & Woodarz (05).
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Michor’s model + immune response
Cells without mutations:
y0
= [ry (1� u)� d0
]y0
� qcp(C ,T )y0
,
y1
= ayy0 � d1
y1
� qcp(C ,T )y1
,
y2
= byy1 � d2
y2
� qcp(C ,T )y2
,
y3
= cyy2 � d3
y3
� qcp(C ,T )y3
.
Anti-cancer T cells:
T = st � dtT � p(C ,T )C + 2nqTp(Cn⌧ ,Tn⌧ )Cn⌧ ,
p(C ,T ) = p0
e�cnCkT , C =X
(yi + zi ), Cn⌧ = C (t � n⌧).
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Accounting for the immune response
0 10 20 30 40 500
0.01
0.02
0.03
0.04
0.05
0.06
T cells
Leukemia
No immuneresponse
Time (months)
Cel
l Con
cent
ratio
n (k
/µL)
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Stopping imatinib (simulation)
0 5 10 15 200
5
10
15
20
25
30
35
40
Leukemia
1000 x T cellconcentration
Time (months)
Cel
l Con
cent
ratio
n (k
/µL)
Imatinib removed at month 12
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Cancer vaccines: a mathematical design
0 2 4 6 8 10 12 14 160
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
Leukemia cells
Anti-leukemia T cells
Time (months)
Cel
l Con
cent
ratio
n (k
/µL)
Inactivated leukemiacells (vaccines)
0 5 10 15-20
-15
-10
-5
0 5 10 15-20
-15
-10
-5
0
0 5 10 15-20
-15
-10
-5
0
0 5 10 15-15
-10
-5
0
5
log 10
[con
cent
ratio
n]
SC PC
Time (months)
log 10
[con
cent
ratio
n]
DC
Time (months)
TC
Inactivated leukemia cells
V = �dVV � qcp(C ,T )V + sv (t)
Anti-cancer T cells
T = st � dtT � p(C ,T )(C + V ) + 2np(Cn⌧Tn⌧ )(qTCn⌧ + Vn⌧ )
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Model populations
Host CellsCancerAnti-donor T cellsGeneral blood cells
Donor cellsAnti-cancer T cells(cancer-specific)Anti-host T cellsGeneral blood cells
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Everything takes time
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Anti-Cancer T Cells
Survive
Perish
Ignore
ReactTC
Death
!!
kTDTC
TD /T
C InteractionT C/C
Inte
ract
ion
n"#Proliferate
Reload
p2TC/C
p1TC/C
p1TD/TC
p2TD/TC
x2n
kCTC
dTC
q1TC/C
q2TC/C
q3TC/C
Die or Become anergic
$
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Anti-Host T Cells
!
"
dTH
kCTH
p2TH/C
TH /T
D Interaction
T H/H
Inte
ract
ion
p2TH/H
p1TH/H
kHTHn# $
x2n
x2n
q1TH/H q2
TH/H
"
$
p1TH/C
q1TH/C
q2TH/C
n#Proliferate
Reload
"
!n#
$Reload
Proliferate
x2n
kTDTH
T H/C
Inte
ract
ion
p2TH/TD
p1TH/TD
q2TH/TD
q1TH/TD
Death!
Killed by TD
TH
q3TH/C
Ignore
React
Ignore
ReactDie orbecome anergic
Survive
Perish
Ignore
Reactp1TD/TH
p2TD/TH
q3TH/TD =0
No flow
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Anti-Donor T CellsT
D /TH Interaction
kTHTD
q2TD/TH
q1TD/TH
Killed by TH
!
"n#
$Reload
Proliferate
x2n
p2TD/TH
p1TD/TH
Survive
Perish
Ignore
Reactp1TH/TD
p2TH/TD
q3TD/TH =0
kTCTD
p2TD/TC
T D/D
Inte
ract
ion
p2TD/D
p1TD/D
kDTD
q1TD/D q2
TD/D
p1TD/TC
q1TD/TC
q2TD/TC
T D/T
C In
tera
ctio
n
No flow
q3TD/TC = 0
"
!
n# $
x2n
x2n
!
$
n#Proliferate
Reload
"
Ignore
React
Ignore
React
dTD
TD
Death
No flow
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
General Donor and Host Blood Cells
Death
dD
TD /D
Interaction
Stem Cells
D
Death
dH
kHTH TH /H
Interaction
p1TH/H
Stem Cells
H
SD SH
kDTDp1TD/D
!
Perish
!
Perish
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Cancer Cells
PerishPerishC
Death
kTCCkTHCp1C/TH p1
C/TC
C/T
C InteractionC
/TH
Inte
ract
ion
rC
Death rate is includedin net logistic growth term
logistic growth
!!
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Equations...
dTH
dt= �dTH
TH � kCTH � kTDTH � kHTH
+ pTH/C2
kC (t � �)TH(t � �) + pTD/TH2
pTH/TD2
kTD(t � �)TH(t � �)
+ pTH/H2
kH(t � �)TH(t � �)
+ 2npTH/C1
qTH/C1
kC (t � ⇢� n⌧)TH(t � ⇢� n⌧)
+ 2npTH/H1
qTH/H1
kH(t � ⇢� n⌧)TH(t � ⇢� n⌧)
+ 2npTD/TH2
pTH/TD1
qTH/TD1
kTD(t � ⇢� n⌧)TH(t � ⇢� n⌧)
+ pTH/C1
qTH/C2
kC (t � ⇢� �)TH(t � ⇢� �)
+ pTH/H1
qTH/H2
kH(t � ⇢� �)TH(t � ⇢� �)
+ pTD/TH2
pTH/TD1
qTH/TD2
kTD(t � ⇢� �)TH(t � ⇢� �).
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Time Delays
Time for reactive T cell-antigen interaction = 5min
Time for unreactive interactions = 1min
Time for cell division = 0.5-1.5 day
T cell recovery time after killing another cell = 1 day
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Relapse
0 20 40 60 80 1000
0.5
1
1.5
2
Time in Days
Cel
l Con
cent
ratio
n in
103
cells
/µL
0 200 400 600 8000
0.2
0.4
0.6
0.8
1
1.2x 10
!6
Time in Days
Cel
l Con
cent
ratio
n in
103
cells
/µLGeneral host cells H
Anti!host T cells TH
Cancer cells C
never goes to 0
eventually overwhelms TH
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Remission
0 20 40 600
0.2
0.4
0.6
0.8
1
1.2x 10
!6
Time in Days
Cel
l Con
cent
ratio
n in
103
cells
/µL
0 20 40 600
0.5
1
1.5
2
2.5
3
3.5
4
Time in Days
Cel
l Con
cent
ratio
n in
103
cells
/µL
General host cells H
Anti!host cells TH
Cancer cells C
C = 0 at time 25.7276
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Oscillations
0 500 1000 15000
0.2
0.4
0.6
0.8
1
B: Unstable Oscillation
Time in Days
Cell C
once
ntra
tion
in 1
03 c
ells
/µL
0 500 1000 15000
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4A: Stable oscillation
Time in Days
Cell C
once
ntra
tion
in 1
03 c
ells
/µL
Anti!host T cellsCancer cellsGeneral host cells
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Extinction instead of stability
8 8.05 8.1 8.15!25
!20
!15
!10
!5
0
5
10
15
20
25Without state constraint
Time in Days
Cel
l Con
cent
ratio
n in
103
cells
/µL
0 2 4 6 80
5
10
15
20
25With state constraint
Time in Days
Cel
l Con
cent
ratio
n in
103
cells
/µL
The value of C crosses 0at time 8.0192.
Cancer cells C
Anti!host T cells TH
Anti!host T cells TH
Cancer cells C
The value of C crosses 0at time 8.0192, and doesnot recover.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Initial anti-host cells vs. initial host cells
1 1.5 2 2.5 3 3.5 4 4.5 5
2
4
6
8
10
12
14
16
Initi
al a
nti!
host
T c
ell c
once
ntra
tion
T H,0
(cel
ls/µ
L)
Initial host cell concentration H0 (103 cells/µL)
Results up to time 2000SuccessfulUnsuccessfulUnresolved
Higher initial host blood cellconcentrations improve thechances of a successful cure
Greater initial anti-host T cellconcentrations slightly favor thechances of cure
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The immune response to leukemia
Stem cell transplantation
Number of cell divisions vs. cancer growth rate
0.03 0.035 0.04 0.045 0.055
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
5.9
6
Cancer growth rate rC (1/day)
Ave
rage
# o
f T c
ell d
ivis
ions
n
Results up to time 1000SuccessfulUnsuccessfulUnresolved
A higher average number of Tcell divisions favor completeremission
Higher cancer growth rate makecomplete remission slightlymore likely
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Zero Crossing
Time Scales
Outline
1 MotivationCancer ImmunologyStem Cell Transplantation
2 PropertiesZero CrossingsTime Scales
3 Numerical Methods
4 Stability in the delays space
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Zero Crossing
Time Scales
Zero crossing: Example
A simple DDE:
dx
dt= �rx(t � 1), x(t) = 1, t < 0.
Solve: t 2 [0, 1).dx
dt= �rx(t � 1) = �r
Thenx = �rt + c = 1� rt.
If r > 0 then x(t) = 0 for T 2 [0, 1)
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Zero Crossing
Time Scales
Zero crossing: example
A simple DDE:
dx
dt= �rx(t � 1), x(t) = 1, t < 0.
Proceed: t 2 [1, 2)
dx
dt= �rx(t � 1) = �r + r2t � r2
Then
x = 1� rt +r2
2(t � 1)2.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Zero Crossing
Time Scales
Zero crossing: example
A simple DDE:
dx
dt= �rx(t � 1), x(t) = 1, t < 0.
The general solution: t 2 [n, n + 1)
x(t) =n+1X
k=0
(�r)k(t � k + 1)k
k!
Question:
For what r does that exist a T 2 [n, n + 1) such that
n+1X
k=0
(�r)k(T � k + 1)k
k!= 0?
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Zero Crossing
Time Scales
Zero crossing: example
A simple DDE:
dx
dt= �rx(t � 1), x(t) = 1, t < 0.
The general solution: t 2 [n, n + 1)
x(t) =n+1X
k=0
(�r)k(t � k + 1)k
k!
Question:
For what r does that exist a T 2 [n, n + 1) such that
n+1X
k=0
(�r)k(T � k + 1)k
k!= 0?
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Zero Crossing
Time Scales
Time Scales
A toy problem
xn+1
= (1� yn)xn
yn+1
= �x2n + k + yn
The map iterated twice
xn+2
= (1 + x2n � k � yn)(1� yn)xn
yn+2
= �((1� yn)2 + 1)x2n + 2k + yn
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Zero Crossing
Time Scales
Time Scales
A toy problem
xn+1
= (1� yn)xn
yn+1
= �x2n + k + yn
The map iterated twice
xn+2
= (1 + x2n � k � yn)(1� yn)xn
yn+2
= �((1� yn)2 + 1)x2n + 2k + yn
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Zero Crossing
Time Scales
This corresponds to...
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Outline
1 MotivationCancer ImmunologyStem Cell Transplantation
2 PropertiesZero CrossingsTime Scales
3 Numerical Methods
4 Stability in the delays space
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Approach #1
A mesh in time that is based on the delay.
Numerical methods for ODEs that use the mesh points only(multistep methods)
Example:⇢
y 0(t) = f (t, y(t), y(t � ⌧(t))) , t0
t tf ,y(t) = �(t), t t
0
.
A set of meshpoints:
� = {t0
, t1
. . . , tN = tf },
such that tn � ⌧(tn) 2 �.Forward Euler: yn+1
= yn + hn+1
f (tn, yn, yq), q < n.Same idea with Adams-like methods, Heun, etc.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Approach #2: Feldstein
Free the mesh selection from the delayUse extranodal points for the approximation of the delayed termy(t � ⌧(t)).Example: (t
0
↵(t) t)⇢
y 0(t) = f (t, y(t), y(↵(t))) , t0
t tf ,y(t
0
) = y0
,
Assume h = (tf � t0
)/m. For any tn = t0
+ nh, define
q(n) = floor
✓↵(tn)� t
0
h
◆.
A numerical method:⇢yn+1
= yn + hf (tn, yn, zn),y0
= y(t0
),
where zn = yq(n), i.e., a piecewise-constant approximation of y(↵(t)).Alternatively, a piecewise-linear approximation:
zn = (1� r(n))yq(n) + r(n)yq(n)+1
.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Approach #3: Bellman’s method of steps
Assume a constant delay. In the first interval [t0
, t0
+ ⌧ ] the DDE has theform: ⇢
y 0(t) = f (y(t),�(t � ⌧)),y(t
0
) = �(t0
).
In the second interval [t0
+ ⌧, t0
+ 2⌧ ], define y1
= y(t � ⌧) andy2
(t) = y(t). Then:
8>><
>>:
y 01
(t) = f (t � ⌧, y1
(t),�(t � 2⌧)),y 02
(t) = f (t, y2
(t), y1
(t)),y1
(t0
+ ⌧) = �(t0
),y2
(t0
+ ⌧) = y(t0
+ ⌧)
And so on...For every timestep, a larger system. However, this system can be solvedusing standard methods for ODEs.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Approach #4: Methods based on continuous extensions
Continuous Extension: very low cost method to get an accurateapproximation of the solution at every point in the interval
⌘(tn + ✓hn+1
) = �n,1(✓)yn + . . .+ �n,in+1
(✓)yn�in
+hn+1
(yn, . . . , yn�in ; ✓, g⌘,�0n), 0 ✓ 1.
whereg⌘(f , y) = f (t, y , ⌘(t � ⌧(t, y))).
This is how every Matlab routine provides solutions at the sampledpoints
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Approach #4: Methods based on continuous extensions
Consider the DDE⇢
y 0(t) = f (t, y(t), y(t � ⌧(t, y(t)))) , t0
t tf ,y(t) = �(t), t t
0
.
Using continuous extensions, solving the DDE amounts to solvingthe ODE:⇢
w 0n+1
(t) = f (t,wn+1
(t), x(t � ⌧(t,wn+1
(t)))) , tn t tn+1
wn+1
(tn) = yn,
where
x(s) =
8<
:
�(s), s t0
,⌘(s), t
0
s tn,wn+1
(s), tn s tn+1
,
and ⌘ is the continuous extension interpolant.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Additional Reading
Bellen and Zennaro, Numerical Methods for Delay Di↵erentialEquations, Oxford
Shampine and Thompson, Numerical Solution of Delay Di↵erentialEquations
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Outline
1 MotivationCancer ImmunologyStem Cell Transplantation
2 PropertiesZero CrossingsTime Scales
3 Numerical Methods
4 Stability in the delays space
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Stability in the delays space
d2x
dt2+
dx
dt+
dx(t � ⌧1
)
dt+
dx(t � ⌧2
)
dt+ 8x = 0
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Stability crossing curves
Ref:
Gu, Niculescu, Chen, On stability crossing curves for general systems withtwo delays, J. Mathematical Analysis & Applications, 311 (2005), pp.231–253.
Stability crossing curves: The set of delays for which thecharacteristic equation has at least one imaginary zero (or pair ofimaginary zeros).
Associated with change of stability
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The two delays case
A DDE with two constant delays
x(t)+c1
x(t�⌧1
)+c2x(t�⌧2
)+c3
x 0(t)+c4
x 0(t�⌧1
)+c5x 0(t�⌧2
) = 0.
The characteristic equation:
h(s) = h0
(s) + h1
(s)e�⌧1
s + h2
(s)e�⌧2
s .
Let ak(s) = hk(s)/h0(s). Then
a(s, ⌧1
, ⌧2
) = 1 + a1
(s)e�⌧1
s + a2
(s)e�⌧2
s = 0.
Stability: a question of the number of the roots of the characteristicequation with a real part on the right hand side of the plane.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The two delays case
For an imaginary s = i! to satisfy a(s, ⌧1
, ⌧2
) = 0, the vectorcorresponding to the three terms must form a triangle:
a(s, ⌧1
, ⌧2
) = 1 + a1
(s)e�⌧1
s + a2
(s)e�⌧2
s = 0.For an imaginary s = iZ to satisfy
the vector corresponding to the three terms must form a triangle.
Hence, their magnitudes must satisfy the triangle inequalities:
Hence, their magnitudes must satisfy the triangle inequalities:
|a1
(i!)|+ |a2
(i!)| � 1,
�1 |a1
(i!)|� |a2
(i!)| 1.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The two delays case
The triangle inequalities determine which i! may be zeros of a(s)
The set of all such ! are the crossing set ⌦.
Any given ! defines a collection of pairs (⌧1
, ⌧2
).
⌧1
=\a
1
(i!) + (2u � 1)⇡ ± ✓1
!� 0, u = u±
0
, u±0
+ 1, . . .
⌧2
=\a
2
(i!) + (2v � 1)⇡ ⌥ ✓2
!� 0, u = v±
0
, v±0
+ 1, . . .
where from the law of cosine:
✓1,2 = cos�1
✓1 + |a
1,2(i!)|2 � |a2,1(i!)|2
2|a1,2(i!)|
◆,
and u±0
, v±0
are the smallest possible integers such that thecorresponding ⌧
1,2 are nonnegative.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
The two delays case
The crossing set ⌦ always consists of a finite number of intervals offinite length
• The triangle inequalities determine which iZ may be zeros of a(s).
• We call the set of all such Z the crossing set :.
• The crossing set : always consists of a finite number of intervals
of finite length.
(symmetrical, because imaginary roots come in conjugate pairs)Any interval of !’s defines a collection of curves in R2
The general case is a union of the following sets:
• Any given Z defines a collection of pairs (W1, W2)
• Any interval of Z’s defines a collection of curves in R2.
• The curves glue together in different arrangements, depending on which triangle inequalities correspond to the endpoints of the intervals.
Doron Levy Montreal, May 2013
Motivation
Properties
numerical methods
Stability in the delays space
Example
DDE:
dx
dt= �2x(t � ⌧
1
) + x(t � ⌧2
).Regions are marked with the numberof zeros in right half plane.
Doron Levy Montreal, May 2013