AdditionalSqrt

[HansOlsson]

trafficstars

Follow-up to #4358 Making it a draft PR in case there is more missing.

[henrikt-ma]

Here are some regexp search results, including matches both in code and documentation:

• \^[ ]*[-]?[ ]*0.5 : 58 matches
• \^[ ]*[-]?[ ]*\([ ]*1[ ]*/[ ]*2[ ]*\) : 18 matches

Many of these are not in documentation, and I'm wondering if there could be any cases where raising to 0.5 is more correct than using sqrt?

[HansOlsson]

Will recheck. For the 2nd regexp in terms of unit-checking most don't seem to matter:

• Modelica.Media.Water.IF97_Utilities.BaseIF97.Transport.cond_dTp: lambdaREL0 := TREL^(1/2)*sum(a[i]*TREL^(i - 1) for i in 1:4); Where TREL has unit 1
• Modelica.Thermal.HeatTransfer.Components.Convection documentation Nu = 0.453*Re^(1/2)*Pr^(1/3);
• Modelica.Media.IdealGases.Common.MixtureGasNasa.gasMixtureViscosity (8*(1 + M[i]/M[j]))^(1/2)
• Modelica.Fluid.Dissipation.Utilities.SharedDocumentation.HeatTransfer.Channel.kc_evenGapLaminar Nu_3 = [2/(1+22*Pr)]^(1/6)*(Re*Pr*d_hyd/L)^(1/2) documentation-class
• Modelica.Fluid.Dissipation.HeatTransfer.StraightPipe.kc_laminar_KC(Re*IN_con.d_hyd/IN_con.L)^(1/2)- which has unit 1.

The following seem more concerning - but involve other exponents as well so should probably be a separate PR:

• Modelica.Media.IdealGases.Common.MixtureGasNasa.lowPressureThermalConductivity but there are lots of other unit-issues as well including other exponents.
• Modelica.Media.IdealGases.Common.MixtureGasNasa.mixtureViscosityChung all kinds of exponents

[HansOlsson]

Have now gone through all of them, the remaining ones are:

• Documentation
• Part of class with other non-integer exponents as well. To me the minor benefit of having sqrt(x) instead of x^0.5 is outweighed by the confusion caused by mixing sqrt(x) and x^(1/3)

[casella]

@HansOlsson would you suggest some other reviewers to get this through?

[HansOlsson]

@HansOlsson would you suggest some other reviewers to get this through?

I believe you can review it, as some are part of media and possibly @AHaumer as it relates to complex numbers.

[casella]

I have no strong opinion about this, I need to think about it.

[MartinOtter]

I have no strong opinion about this, I need to think about it.

In general sqrt(x) should be preferred over x^0.5 (assuming both are directly mapped to the corresponding C-expressions):

• If x^0.5 is mapped to pow(x,0.5), then sqrt(x) is a specialized algorithm that is more efficient, might have better precision, and most CPUs have even dedicated square root instructions.
• Readability is better - it is much clearer that the computation is a square root computation

[Esther-Devakirubai]

@casella Can this be backported to maint/4.1.0? No milestone either.

[casella]

@casella Can this be backported to maint/4.1.0? No milestone either.

Given @HansOlsson's comment, you can do that. It's not critical but it doesn't harm.

[Esther-Devakirubai]

Backporting this to maintenance branch by #4452