The numeric solution of ODE BVPs related to nanofluids is one of the most-common sources of Questions on MaplePrimes; perhaps the most-common source, with about 3 or 4 Questions per week over the last several years. The papers in this field are often shoddy, and they often use ad hoc methods for numeric solution of the BVPs. These methods usually have no error control. I have not seen any that come close to Maple's ability with BVPs. I have not analyzed any papers in this field that did not have either significant computational errors or significant omissions that make reproducing the entirety of the numeric computations impossible.

These systems often have many parameters, both as input and output. Maple's dsolve(..., numeric) allows symbolic input parameters for IVPs but not BVPs. On the other hand, it allows symbolic output parameters for BVPs but not IVPs. By using the technique that I show in the attached worksheet, these distinctions in the way that parameters are handled can be made mostly transparent to the end user so that all parameters can be treated as symbolic. Once this has been done, it is easy to make plots (both 2-D and 3-D) whose independent variables are the input parameters themselves, and whose dependent variables are the output parameters.

Other techniques covered in this worksheet are

  • how to make a 3D surface plot combined with a dropped-shadow view of the associated contour plot;
  • how to make 3D plots with boxed axes but with the three annoying foreground lines removed;
  • how to do a statisical fit to express an output parameter as a function of the input parameters using only data that's been collected during the production of the plots.


 

Numeric solution of ODE BVPs with many parameters

 

Author of this Maple worksheet: Carl Love <mailto:carl.j.love@gmail.com>, 14-Oct-2018

 

 

This worksheet is a critique of and an attempt to reproduce all numeric computations in the paper:

 

M. Sheikholeslami, D.D. Ganji, M.M. Rashidi, "Magnetic field effect on unsteady flow and heat transfer using Buongiorno model", Journal of Magnetism and Magnetic Materials, 416 (2016) 164-173, Elsevier, 11-May-2016, http://dx.doi.org/10.1016/j.jmmm.2016.05.026

 

That paper's authors' affiliations and email addresses:

M. Sheikholeslami: Department of Mechanical Engineering, Babol Univeristy of Technology, Babol, Islamic Republic of Iran <mohsen.sheikholeslami@yahoo.com>, <m.sheikholeslmai@stu.nit.ac.ir>,

D.D. Ganji: Department of Mechanical Engineering, Babol Univeristy of Technology, Babol, Islamic Republic of Iran <mirgang@nit.ac.ir>,

M.M. Rashidi: Shanghai Key Lab of Vehicle Aerodynamics and Vehicle Thermal Management Systems, Tongji Univeristy, 4800 Cao An Road, Shanghai 201804, China (no email address given).

 

All references to Equation numbers, Figure numbers, and Table numbers in this worksheet are from that paper.

 

restart:

Digits:= trunc(evalhf(Digits)); #generally a very efficient setting

15

Setup of BVP system:

#ordinary differential equations:
ODEs:= [
   #Eq 8:
   diff(f(eta),eta$4)
   - S*(diff(f(eta),eta$3)*(eta-f(eta)) + diff(f(eta),eta$2)*(3+diff(f(eta),eta)))
   - Ha^2*diff(f(eta),eta$2),

   #Eq 9:
   (1+4/3*Rd)*diff(theta(eta),eta$2)
   + Pr*(S*diff(theta(eta),eta)*(f(eta)-eta) + Ec*diff(f(eta),eta$2)^2)
   + Nb*diff(phi(eta),eta)*diff(theta(eta),eta) + Nt*diff(theta(eta),eta)^2,

   #Eq 10:
   diff(phi(eta),eta$2) + S*Sc*diff(phi(eta),eta)*(f(eta)-eta)
   + Nt/Nb*diff(theta(eta),eta$2)

   #All these ODEs are implicitly equated to 0.
]:

<ODEs[]>; #Display the ODEs.
   

Vector(3, {(1) = diff(diff(diff(diff(f(eta), eta), eta), eta), eta)-S*((diff(diff(diff(f(eta), eta), eta), eta))*(eta-f(eta))+(diff(diff(f(eta), eta), eta))*(3+diff(f(eta), eta)))-Ha^2*(diff(diff(f(eta), eta), eta)), (2) = (1+(4/3)*Rd)*(diff(diff(theta(eta), eta), eta))+Pr*(S*(diff(theta(eta), eta))*(f(eta)-eta)+Ec*(diff(diff(f(eta), eta), eta))^2)+Nb*(diff(phi(eta), eta))*(diff(theta(eta), eta))+Nt*(diff(theta(eta), eta))^2, (3) = diff(diff(phi(eta), eta), eta)+S*Sc*(diff(phi(eta), eta))*(f(eta)-eta)+Nt*(diff(diff(theta(eta), eta), eta))/Nb})

#lower and upper boundary values of the independent variable:
(LB,UB):= (0,1): #Eq 11

#boundary conditions:
BCs:= [
   #Eq 11:
   ([f, (D@@2)(f), D(theta), D(phi)](LB) =~ [0, 0, 0, 0])[],
   ([f, D(f), theta, phi](UB) =~ [1, 0, 1, 1])[]
];

[f(0) = 0, ((D@@2)(f))(0) = 0, (D(theta))(0) = 0, (D(phi))(0) = 0, f(1) = 1, (D(f))(1) = 0, theta(1) = 1, phi(1) = 1]

#parameters (input values to BVP system) and their default values:
Params:= Record(
   #Eq 12:
   S=  0.5,  #Squeeze number
   Pr= 10.,  #Prandtl number
   Ec= 0.1,  #Eckert number
   Sc= 0.5,  #Schmidt number
   Ha= 2.,   #Hartman number
   Nb= 0.15, #Brownian motion parameter
   Nt= 0.15, #thermophoretic parameter
   Rd= 4.5   #radiation parameter
):
   

#Named special computed boundary values (output values of the system): Note that each adds
#exactly one parameter and one boundary condition to the system.

NBVs:= [   
   -(D@@2)(f)(1) = `C*__f`,  #Eq 13: Skin friction coefficient
   -D(theta)(1) = `Nu*`      #Eq 14: Nusselt number
]:

#just rewriting the above names in a more-convenient form:
Nu:= `Nu*`:
Cf:= `C*__f`:

#Appliable module that calls dsolve for each
#particular configuration of parameter values. (Maple's numeric BVPs require a separate
#dsolve invocation for each specific set of parameter values (unlike IVPs, which can be
#parameterized into a single dsolve invocation).)
#
#For every dsolve call through this module, the NBVs are evaluated and saved.

Solve:= module()
local
   nbvs_rhs:= rhs~(:-NBVs), #just the names
   Sol, #numeric dsolve BVP solution of any 'output' form
   ModuleApply:= subs(
      _Sys= {:-ODEs[], :-BCs[], :-NBVs[]},
      proc({
         #These are just default parameter values: They can be changed when the
         #procedure is called.
         S::realcons:=  Params:-S,
         Pr::realcons:= Params:-Pr,
         Ec::realcons:= Params:-Ec,
         Sc::realcons:= Params:-Sc,
         Ha::realcons:= Params:-Ha,
         Nb::realcons:= Params:-Nb,
         Nt::realcons:= Params:-Nt,
         Rd::realcons:= Params:-Rd
         #By extracting the default values from record Params, it is possible to change
         #them by changing the record.
      })
         Sol:= dsolve(_Sys, _rest, numeric);
         AccumData(Sol, {_options});
         Sol
      end proc
   ),
   AccumData:= proc(
      Sol::{Matrix, procedure, list({name, function}= procedure)},
      params::set(name= realcons)
   )
   local n, nbvs;
      if Sol::Matrix then
         nbvs:= seq(n = Sol[2,1][1,Pos(n)], n= nbvs_rhs)
      else
         nbvs:= (nbvs_rhs =~ eval(nbvs_rhs, Sol(:-LB)))[]
      fi;
      SavedData[params]:= Record[packed](params[], nbvs)
   end proc,
   ModuleLoad:= eval(Init)
;
export
   SavedData, #table of Records
   Pos, #Matrix column indices of nbvs
   Init:= proc()
      Pos:= proc(n::name) option remember; local p; member(n, Sol[1,1], 'p'); p end proc;
      SavedData:= table();
      return
   end proc   
;
   ModuleLoad()
end module:
 

Now we attempt to recreate all the numeric computations of the paper.

 

Figure 1 is conceptual, not computational, so there's no point in trying to reproduce it here.

 

I can't attempt to reproduce Table 1 or Figure 2 because the paper doesn't specify the values of all the parameters used for those.

 

Recreate Figure 3:

colseq:= [red, green, blue, brown]:

#parameter values that remain fixed for the entire set of plots:
Pc:= Sc= 0.5, Nb= 0.1, Ec= 0.1, Pr= 10:
 

#parameter values that remain fixed with each of the four plots::
Ps:= [
   [S= 0.5, Ha= 2, Nt= 0.1],
   [Ha= 2, Rd= 0.5, Nt= 0.1],
   [S= 0.5, Rd= 0.5, Nt= 0.1],
   [S= 0.5, Ha= 2, Rd= 0.5]
]:

#parameter value for each curve
Pv:= [
   Rd= [0.5, 2, 4, 8],
   S=  [2, 4, 8, 12],
   Ha= [2, 4, 8, 12],
   Nt= [0.01, 0.05, 0.1, 0.2]
]:
      

for i to nops(Ps) do
   plots:-display(
      [seq(
         plots:-odeplot(
            Solve(lhs(Pv[i])= rhs(Pv[i])[j], Ps[i][], Pc),
            [eta, theta(eta)], 'color'= colseq[j], 'legend'= [lhs(Pv[i])= rhs(Pv[i])[j]]
         ),
         j= 1..nops(rhs(Pv[i]))
      )],
      'axes'= 'boxed', 'gridlines'= false,
      'labelfont'= ['TIMES', 'BOLDOBLIQUE', 16],
      'caption'= nprintf(
         cat("\n%a = %4.2f, "$nops(Ps[i])-1, "%a = %4.2f\n\n"), (lhs,rhs)~(Ps[i])[]
      ),
      'captionfont'= ['TIMES', 16]
   )
od;

The 4 plots agree with those in Fig. 3 in  the paper.

 

Recreate Fig. 4.

 

#procedure that generates a 2-D plot of an expression versus a varied parameter:
ParamPlot2d:= proc(
   Y::{procedure, `module`}, #procedure that extracts y-value from Solve's dsolve solution
   X::name= range(realcons), #x-axis-parameter range
   FP::list(name= realcons), #fixed values of other parameters
   {
      eta::realcons:= :-LB, #independent variable value
      dsolveopts::list({name, name= anything}):= []  
   }
)
   plot(
      x-> Y(
         Solve(
            lhs(X)= x, FP[],
            #Default dsolve options can be changed by setting 'dsolveopts':
            'abserr'= 0.5e-7, 'interpolant'= false, 'output'= Array([eta]), dsolveopts[]
         )
      ),
      rhs(X),
      #Default plot options can be changed by putting the option in the ParamPlot2d call:
      'numpoints'= 25, 'axes'= 'boxed', 'gridlines'= false,
      'labelfont'= ['TIMES', 'BOLDOBLIQUE', 16],
      'caption'= nprintf(cat("%a = %4.2f, "$nops(FP)-1, "%a = %4.2f"), (lhs,rhs)~(FP)[]),
      'captionfont'= ['TIMES', 16],
      _rest
   )
end proc:

#procedure that extracts Nusselt number from dsolve solution:
GetNu:= (Sol::Matrix)-> Sol[2,1][1, Solve:-Pos(:-Nu)]:

RD:= [0.5, 2, 4]:
plots:-display(
   [seq(
      ParamPlot2d(
         GetNu, Ha= 2..8, [S= 0.5],
         dsolveopts= [Rd= RD[k], Sc= 0.5, Nb= 0.1, Ec= 0.1, Nt= 0.1, Pr= 10],
         'legend'= [Rd= RD[k]], 'color'= colseq[k], 'labels'= [Ha, Nu]
      ),
      k= 1..nops(RD)
   )]
);

SV:= [0.5, 2, 4]:
plots:-display(
   [seq(
      ParamPlot2d(
         GetNu, Ha= 2..8, [Rd= 0.5],
         dsolveopts= [S= SV[k], Sc= 0.5, Nb= 0.1, Ec= 0.1, Nt= 0.1, Pr= 10],
         'legend'= [S= SV[k]], 'color'= colseq[k], 'labels'= [Ha, Nu]
      ),
      k= 1..nops(SV)
   )]
)

The above two plots match those of Fig. 4 in the paper.

 

#procedure that generates 3-D plots (dropped-shadow contour + surface) of an expression
#versus two varied parameters:

ParamPlot3d:= proc(
   Z::{procedure, `module`}, #procedure that extracts z-value from Solve's dsolve solution
   X::name= range(realcons), #x-axis-parameter range
   Y::name= range(realcons), #y-axis-parameter range
   FP::list(name= realcons), #fixed values of other parameters
   {
      #fraction of empty space above and below plot (larger "below"
      #value improves view of dropped-shadow contourplot):
      zmargin::[realcons,realcons]:= [.05,0.15],
      eta::realcons:= :-LB, #independent variable value
      dsolveopts::list({name, name= anything}):= [],
      contouropts::list({name, name= anything}):= [],
      surfaceopts::list({name, name= anything}):=[]    
   }
)
local
   LX:= lhs(X), RX:= rhs(X), LY:= lhs(Y), RY:= rhs(Y),
   Zremember:= proc(x,y)
   option remember; #Used because 'grid' should be the same for both plots.
      Z(
         Solve(
            LX= x, LY= y, FP[],
            #Default dsolve options can be changed by setting 'dsolveopts':
            'abserr'= 0.5e-7, 'interpolant'= false, 'output'= Array([eta]),  
            dsolveopts[]
         )
      )
   end proc,
   plotspec:= (Zremember, RX, RY),
   C:= plots:-contourplot(
      plotspec,
      #These default plot options can be changed by setting 'contouropts':
      'grid'= [25,25], 'contours'= 5, 'filled',
      'coloring'= ['yellow', 'orange'], 'color'= 'green',
      contouropts[]
   ),
   P:= plot3d(
      plotspec,
      #These default plot options can be changed by setting 'surfaceopts':
      'grid'= [25,25], 'style'= 'surfacecontour', 'contours'= 6,
      surfaceopts[]
   ),
   U, L #z-axis endpoints after margin adjustment
;
   #Stretch z-axis to include margins:
   (U,L):= ((Um,Lm,M,m)-> (M*(Lm-1)+m*Um, M*Lm+m*(Um-1)) /~ (Um+Lm-1))(
      zmargin[],
      (max,min)(op(3, indets(P, 'specfunc'('GRID'))[])) #actual z-axis range
   );
   plots:-display(  #final plot to display/return
      [
         plots:-spacecurve(
            {
               [[lhs(RX),rhs(RY),U],[rhs(RX),rhs(RY),U],[rhs(RX),rhs(RY),L]], #yz backwall
               [[rhs(RX),rhs(RY),U],[rhs(RX),lhs(RY),U],[rhs(RX),lhs(RY),L]]  #xz backwall
            },
            'color'= 'grey', 'thickness'= 0
         ),
         plottools:-transform((x,y)-> [x,y,L])(C), #dropped-shadow contours
         P
      ],
      #These default plot options can be changed simply by putting the option in the
      #ParamPlot3d call:
      'view'= ['DEFAULT', 'DEFAULT', L..U], 'orientation'= [-135, 75], 'axes'= 'frame',
      'labels'= [lhs(X), lhs(Y), Z], 'labelfont'= ['TIMES', 'BOLDOBLIQUE', 16],
      'caption'= nprintf(cat("%a = %4.2f, "$nops(FP)-1, "%a = %4.2f"), (lhs,rhs)~(FP)[]),
      'captionfont'= ['TIMES', 14],
      'projection'= 2/3,   
      _rest
   )
end proc:

Reproduce the twelve 3-D plots of Figure 5. Note that every plot uses a default grid of 25x25, so requires 625 complete solutions of the BVP. So, this might be slower than you expect.

ParamPlot3d(
   GetNu, S= 0..12, Ha= 0..12, [Rd= 4.5, Ec= 0.06, Nb= 0.15, Nt= 0.15, Sc= 0.5],
   labels= [S, Ha, Nu]
);

The above plot matches the paper.

 

ParamPlot3d(
   GetNu, S= 0..12, Rd= 1..8, [Ha= 6, Ec= 0.06, Nb= 0.15, Nt= 0.15, Sc= 0.5],
   labels= [S, Rd, Nu]
);

The above plot matches the paper.

 

ParamPlot3d(
   GetNu, S= 0..12, Ec= 0.01..0.10, [Ha= 6, Rd= 4.5, Nb= 0.15, Nt= 0.15, Sc= 0.5],
   labels= [S, Ec, Nu]
);

The above plot matches the paper.

 

ParamPlot3d(
   GetNu, S= 0..12, Nt= 0.1..0.2, [Ha= 6, Rd= 4.5, Ec= 0.06, Nb= 0.15, Sc= 0.5],
   labels= [S, Nt, Nu]
);

The above plot doesn't match the corresponding one in the paper!

 

ParamPlot3d(
   GetNu, Ha= 0..12, Rd= 1.0..8.0, [S= 6, Ec= 0.06, Nb= 0.15, Nt= 0.15, Sc= 0.5],
   labels= [Ha, Rd, Nu]
);

The above plot matches the paper.

 

ParamPlot3d(
   GetNu, Ha= 0..12, Ec= 0.01..0.1, [S= 6, Rd= 4.5, Nb= 0.15, Nt= 0.15, Sc= 0.5],
   labels= [Ha, Ec, Nu]
);
   

The above plot matches the paper.

 

ParamPlot3d(
   GetNu, Ha= 0..12, Nt= 0.1..0.2, [S= 6, Rd= 4.5, Ec= 0.06, Nb= 0.15, Sc= 0.5],
   labels= [Ha, Nt, Nu]
);

The above plot matches the paper.

 

ParamPlot3d(
   GetNu, Ha= 0..12, Nb= 0.1..0.2, [S= 6, Rd= 4.5, Ec= 0.06, Nt= 0.15, Sc= 0.5],
   labels= [Ha, Nb, Nu]
);

The above plot matches the paper.

 

ParamPlot3d(
   GetNu, Rd= 1.0..8.0, Nt= 0.1..0.2, [S= 6, Ha= 6, Ec= 0.06, Nb= 0.15, Sc= 0.5],
   labels= [Rd, Nt, Nu]
);

The above plot matches the paper.

 

ParamPlot3d(
   GetNu, Rd= 1.0..8., Nb= 0.1..0.2, [S= 6, Ha= 6, Ec= 0.06, Nt= 0.15, Sc= 0.5],
   labels= [Rd, Nb, Nu]
);

The above plot matches the paper.

 

ParamPlot3d(
   GetNu, Ec= 0.01..0.1, Nt= 0.1..0.2, [S= 6, Ha= 6, Rd= 4.5, Nb= 0.15, Sc= 0.5],
   labels= [Ec, Nt, Nu]
);

ParamPlot3d(
   GetNu, Ec= 0.01..0.1, Nb= 0.1..0.2, [S= 6, Ha= 6, Rd= 4.5, Nt= 0.15, Sc= 0.5],
   labels= [Ec, Nb, Nu]
);

The above plot does not match the corresponding one in the paper!

 

Recreate Fig. 6:

#parameter values that remain fixed for the entire set of plots:
Pc:= Sc= 0.5, Ec= 0.1, Pr= 10:
 

#parameter values that remain fixed within each plot:
Ps:= [
   [S= 0.5, Ha= 2, Nt= 0.1, Nb= 0.1],
   [Ha= 2, Rd= 0.5, Nt= 0.1, Nb= 0.1],
   [S= 0.5, Rd= 0.5, Nt= 0.1, Nb= 0.1],
   [S= 0.5, Ha= 2, Rd= 0.5, Nb= 0.1],
   [S= 0.5, Ha= 2, Rd= 0.5, Nt= 0.1]
]:

#parameter value for each curve in each plot
Pv:= [
   Rd = [0.5, 2, 4, 8],
   S= [2, 4, 8, 12],
   Ha= [2, 4, 8, 12],
   Nt= [0.01, 0.05, 0.1, 0.2],
   Nb= [0.1, 0.2, 0.4, 0.6]
]:
      

for i to nops(Ps) do
   plots:-display(
      [seq(
         plots:-odeplot(
            Solve(lhs(Pv[i])= rhs(Pv[i])[j], Ps[i][], Pc),
            [eta, phi(eta)], 'color'= colseq[j], 'legend'= [lhs(Pv[i])= rhs(Pv[i])[j]]
         ),
         j= 1..nops(rhs(Pv[i]))
      )],
      'axes'= 'boxed', 'gridlines'= false,
      'labelfont'= ['TIMES', 'BOLDOBLIQUE', 16],
      'caption'= nprintf(
         cat("\n%a = %4.2f, "$nops(Ps[i])-1, "%a = %4.2f\n\n"), (lhs,rhs)~(Ps[i])[]
       ),
       'captionfont'= ['TIMES', 16]
   )
od;

The five plots above match Fig. 6 from the paper.

 

Generate Eq. 31: At this point, we have a huge database of Nusselt numbers (15,186 records to be exact) computed at various parameter values.

NusseltList:= [entries(Solve:-SavedData, nolist)]:

nops(NusseltList);

15136

Sheikholeslami, et al, call these the "active parameters" (because they're the ones that have varied in the above computations):

V:= [S, Ha, Rd, Ec, Nt, Nb]:

NusseltData:= Matrix(map(R-> [map(v-> R[v], V)[], R[Nu]], NusseltList));